AI Sidequest: How-To Tips and News

Muse tried to sell land for $1

Mignon Fogarty — Mon, 05 Oct 2026 13:00:00 +0000

Issue 117

Remember to view the newsletter in a browser to see the part that gets cut off at the bottom in email.

Personal agents take off

Muse

Meta’s Muse agent, with its cute little cartoon avatar, ease of use, and massive launch platform, took the world by storm over the last couple of weeks. Everyone in AI is talking about Muse (and Dots, more below), but more importantly, my actual real-life friends are using it. It’s the number one app in the Apple App Store and has supposedly been adopted faster than even ChatGPT.

Muse initially seemed designed for consumers, but Meta did launch a small business version soon after the initial launch. One friend used it to help coordinate her move (including using it to book a U-Haul), and another asked for sightseeing recommendations on a trip and suggestions on what to buy for a hobby — and both were happy with it.

One use Meta seems to be pushing is having Muse cancel unwanted/unused subscriptions. So much so that other people are speculating what could happen to the economy if too many people rapidly become more responsible with their money.

If you want agents to do especially useful things, though, you have to give them access to a lot of personal information and accounts, which I’m not comfortable doing — especially with Meta. Although I did see someone say they were OK with letting it connect to Facebook and Instagram since Meta already has all that information, and that made sense to me.

Among the many positive examples I saw, there were some epic failures.

Muse miscommunicated with a buyer and ended up giving out the seller’s address without permission and sending the buyer to the seller’s house when he wasn’t home, even replying in chat, “Yup, I’m here!” when the understandably angry buyer tried to make contact.
A tech reporter says Muse accessed his private Apple messages (and lied about it) even though he hadn’t given it permission to do so.
Muse began offering real estate to potential buyers on Facebook Marketplace for $1 after several days of successfully handling the listings.

Meta is massively subsidizing the project, so Muse is free, and I’ve seen developers, in particular, marveling at how much compute they’re getting free right now. Meta says it expects the business model will be taking a cut of purchases, which will likely lead it to make recommendations based on its business interests instead of yours. Further, not all potential sellers are excited about the prospect — Amazon has blocked Muse. People also suggest Meta may get revenue by “helping” businesses advertise more effectively on Meta platforms.

Dots

OpenAI launched its answer to Muse, Dots, at its DevDay last week, and early reports on it are positive too. Casey Newton likes it.

Dots is only available on the $100/month and higher ChatGPT plans and is much more clearly targeting developers and business users. That’s too rich for my blood, but the 9-minute launch video does a great job giving you a feel for how it works. If you’re curious, it’s worth your time.

Meanwhile, David Pogue still loves the new Siri.

The AI Inner Circle newsletter created an excellent document that compares Muse, Dots, and GrokBot (PDF).

Agents are looking for freelance work

AI agents have been pitching editors and website owners trying to get freelance writing gigs to cover their token costs on a platform called iLands. Multiple editors and owners have reported receiving the emailed pitches (Tedium, Rest of the World, Jane Friedman for her Bottom Line newsletter). 404 Media reports that “Many of the messages we’ve received in recent days from iLands agents aren’t all that different from the types of public relations pitches and freelance article pitches we get from humans.”

Word watch: Liquid content

I recently heard the term “liquid content” to describe personalized content that is built on the fly, after a user clicks a headline (which may also have been personally generated by AI). (Claude’s ability to remix “liquid content” is a “serious challenge for news media.” — Press Gazette)

Previously, “liquid content” described content that could be reformatted in multiple ways, such as an article that was then turned into a newsletter, LinkedIn post, podcast, and short-form video.

And as my worlds converge, I see that Inception Point AI — the company that was producing thousands of sloppy AI-generated podcasts — recently changed its name to Liquid Studios. They now describe their flood of podcasts as “a PR strategy to get attention.” I wonder how long they’ll stick with their new liquid ambitions.

Claude releases its hold on em dashes

Arena AI’s analysis of Opus 5.5’s writing style found that it uses 95% fewer em dashes than the previous model. It also uses fewer semicolons, but it says “arguably” and “perhaps” 97% more often.

Two very different experiences with AI-generated books

I had one bad experience and one good experience with AI-generated books last week.

Bad experience: confusing and misplaced disclaimers

I read a book I enjoyed, but when I got to the end, I came across an “about this book” section that seemed to say the core creative part of the book had been AI-generated, and I felt tricked and disappointed. I actually reached out to the author, who insisted AI had not been used in the way I thought. So I have two thoughts:

Disclaimers should never go at the end of a piece of writing. The primary reason to tell people how your book was made is to allow readers to avoid it if they don’t want to read AI-generated work. Some people may be including this information with a “Hey, isn’t this cool?!” mindset, but they are going to run into problems with disappointed readers. (The disclaimer I mentioned a couple of newsletters ago that caused a social-media pile-on also appeared at the end of the book.)
Be careful how you write your disclaimers. I’ve read the section in question at least 7 times, and I don’t see how I could have interpreted it any other way except that AI was used in an important part of the book — but I also believe the author. When you write a disclaimer, make sure it is extremely clear. (The social media pile-on I mentioned above also revolved around a misunderstanding of what a disclaimer and “human authored” badge meant.)

Good experience: a better-than-expected audiobook

A clearly labeled AI-generated audiobook showed up in the feed of a podcast I listen to, so I decided to give it a try and was surprised by how good the narration was. As a podcaster, I try to stay on top of what is happening with AI audio, but voices for fiction narration have advanced more than I realized (and I believe this was produced before ElevenLab’s newest model release). Of the 7 or 8 voices, I would have only pegged 2 as AI voices if I hadn’t known.

Further, the story itself was OK. I found it interesting enough to keep listening for a while and have fired it up when I was away from Wi-Fi because I would rather listen to this than a few professional audiobooks I have languishing on my phone because I didn’t like them.

It does have some hilariously obvious AI tells, though, such as the main character being named Alara and including metaphors that seem to misunderstand the physical world, such as “The skin around her temples tightened.” My skin never does that.

Having these two opposite reactions to AI-generated books so close together led me to reflect on what makes me a happy reader versus a sad reader, and it’s all about expectations.

And as a podcaster and writer, I’m not thrilled that AI is improving so much on both fronts, but that’s the reality I see.

Also, I’m intentionally not naming either book because I don’t want to disparage an author and I don’t want to promote an AI-generated book, but if you want the name of the OK audiobook, you can email me and I’ll send it to you. (You may disagree with me; my husband didn’t like it.)

Quick changes to the U.S. government’s chatbot

America.gov is a new chatbot that’s meant to give answers about government programs — a sort of front door to the vast collection of web pages with information you might need. Early reviews are good, but amusingly, at launch it would debunk inaccurate statements made by top government officials. Sadly, it no longer does so.

AI now outperforms junior accountants

The only reason I promoted this report to the main newsletter is that you have to see the chart. It took me a second to realize what I was looking at because the Claude results are so dramatic — 100% accuracy across the board and almost no time at all compared to the humans, who are all over the place:

Were these just really bad accountants? Well, the report says, “Participants averaged about five and a half years of accounting experience, and all were licensed CPAs.”

The full report from Mercor has interesting insights about tasks versus jobs and what made these tests, which did represent real-world accounting problems, especially difficult for humans.

Google is back in the game

Google’s new model, Gemini 4 Argon, beat Astra and Fable on many benchmarks, and like Mythos, it has been released to just a small set of cyber defenders for now. Google also says it’s using the model internally to make its data centers more efficient.

This is bigger news than just a model launch because with delayed releases and multiple top people jumping ship, some people had started counting Google out of the race to generate the best frontier AI.

Are Claude’s discoveries really its own?

When OpenAI made a huge math discovery a few weeks ago, a mathematician claimed he had used OpenAI models for research on the problem and suspected they had trained on his prompts and then essentially jumped ahead and scooped his research. OpenAI denied the accusation.

Now, the same concerns have emerged about Claude’s recent big biology discovery. Anthropic proudly announced that Claude had discovered a completely new enzyme all by itself, but a University of Copenhagen computational biologist named Mario Rodríguez Mestre came forward and provided proof to the New York Times that his group has been working on the same problem for four years and identified the enzyme in question about a year ago. They have not published their discovery, and Mestre’s team regularly used Claude in their own research, writing code, drafting manuscripts, and so on, which involved uploading raw data and other private information.

Until more details emerge, I’d be especially careful about putting proprietary information into a model not on your own server, which is a bummer because one would presume that if the scientists were heavily using Claude, they did it because they were finding it useful for their research.

Uh oh, live video

An alarming new demo appeared this week from a company called Tavus, in what is being called the first AI-generated video to pass the Turing test — 48% of people who chatted with the AI-generated avatar thought it was human, and in the demo I watched on their website, I didn’t feel like I could tell. (In fact, in the first example, I was certain the human was the AI — “Sorry” to that dude).

The company isn’t releasing the product to the public because of safety concerns, and it’s important to note that all the data, demo, and info comes from the company, so I wouldn’t rule out that it’s exaggerated or at least optimized to show us the best possible examples. However, thinking about all the ways this could be misused, whether it actually exists today or in 6 months, is worrisome.

How do you feel about AI for writing?

You still have time to take the Gotham Ghostwriters survey of writers’ attitudes toward AI: https://www.surveymonkey.com/r/AIWPgram

I’m already scheduled to interview Josh Bernoff about the results, so I’d love to have your opinions included.

Quick Hits

My favorite recent pieces

Simon Willison with an overview of everything important that has happened with AI in 2026 [An excellent overview.] — YouTube (~30 minutes; Watchable at double speed.)

What If We're All Wrong and AI Isn’t a Form of Publishing? [This piece argues that instead of ending up as an ad-supported product, AI may be more like a utility; in other words, something we pay for.] — Marketecture

An AI agent emailed hundreds of researchers for help. It told us why [Interesting perspectives on interacting with AI agents that aren’t your own.] — Science

With most information hidden, the game Stratego had stumped AI—until now [This seems like a big advance, and it was done in academia with limited resources.] — Ars Technica

Using AI

How people around the world use AI at work — Rest of the World

How People Are Using Jev [People are excited about a new kind of AI model called Jev that’s not an LLM. It’s a super inexpensive classifier that only outputs a number or a decision. For example, you can have it look at a document and see if it’s been formatted (y/n) or have it evaluate how angry a customer email is on a scale of 1 to 5. I didn’t really understand how people might use it until I listened to the AI Daily Brief episode about it, which you can also see in written form.] — AI Daily Brief

How I Use AI As a Keynote Speaker & Facilitator [A particularly interesting detail is that he uses AI to turn his keynote talk into an ebook that attendees can have within a day or two.] — Christopher Penn

How To Stop AI Companies From Training On Your Data — Engadget

I tested ChatGPT-6 vs Claude Opus 5.5 with 5 everyday prompts — it wasn’t even close — Tom's Guide

Agents

I Gave My Life Over to Meta’s A.I. Agent and Was Blown Away — The New York Times

FTC opens probe into AI giants including Anthropic and OpenAI [The probe is about the potential for agents to cause harm to consumers.] — Yahoo

Bad stuff

How the U.S. and Russia weakened a global effort to regulate killer AI — Washington Post

ChatGPT Helped Tumbler Ridge Shooter Focus on Guns, Tactics, and Terror — Mother Jones

‘Pink Slime’ Is Infecting AI Chatbots Ahead of the Midterms [“Leading AI chatbots cite partisan websites masquerading as independent local news outlets nearly half of the time when people ask about candidates and issues that the partisan sites have covered in the midterm campaign, according to a new audit by NewsGuard.”] — Politico

The business of AI

Startups Like Harvey Embrace Open Models to Cut Reliance on Anthropic, OpenAI - Bloomberg

Why I Am Bullish on ChatGPT Ads — Vincent Schmalback

Chinese AI models surge in global adoption — CNBC

Anthropic warns AI may pose 'existential risks to humanity' in IPO filing — Reuters

Education

On the Value of Doing a PhD in the Age of AI [Note: I am a PhD dropout, and I almost always counsel people against getting a PhD unless it’s something they really (really!) want to do for themselves.] — Phillip Isola (MIT)

I’m a College Professor. Writing Isn’t as Important as We Think. [He found oral interactions and exams to be better than written exams, but doesn’t have a plan for how a professor with a normal number of students would make this work.] — New York Times

Science & Medicine

Insurers claim AI is already increasing healthcare costs [by helping hospitals bill more for the same procedures]— TechCrunch

Security

I Quit OpenAI Because Its Culture Is Broken [A top safety researcher has quit over safety concerns. The article includes a good explanation of why iterative development is a bad idea for AI: “OpenAI has thrived by trial and error … looking for problems and improving its guardrails in response. But this approach, by its very nature, guarantees periodic failures—and the scale of those failures is growing as systems get more capable.”] — The Atlantic

Google DeepMind engineer quits over AI safety concerns — Eyes Above the Waves

Fears about AI are overhyped — Andrew Ng

OpenAI says agents leaked 53 images from ChatGPT users in latest example of rogue activity — The Guardian

Hacking is the least worrying part of OpenAI’s Australia incident — Transformer

OpenAI pauses training of its ‘most capable models’ — The Verge

Models know when they’re reward hacking — and we can catch them at scale — Goodfire

LLMs respond differently to harmful prompts when AI watermarking is used [This is a weird and surprising result.] — Ars Technica

Open-source model GLM-5.3 has reached Mythos-level cyber capabilities and lacks guardrails — Anthropic

AI coding agents leaked 13,000 internal company screenshots to public GitHub repos [Some included private financial information.] — Help Net Security

OpenAI’s A.I. Tried Breaching Four Other Targets, With No Prompting — New York Times

OpenAI Ignored Employees’ Warnings About Safely Testing A.I. Models — New York Times

Other

OpenAI's GPT-6 Astra can now tell you exactly where you screwed up your IKEA shelf [It’s a real IKEA furniture assembly benchmark.] — The Decoder

How Accurate Have AI Progress Forecasts Been So Far? [Forecasters generally overestimated adoption of autonomous vehicles and underestimated advances in LLM capabilities.] — Forecasting Research Institute

Oxford lets OpenAI train its AI models on Bodleian Library — The Guardian

Can AI Stand In for Human Respondents? Not Really. [“Synthetic surveys do a poor job of estimating the public mood and are especially error-prone for some subgroups, including Republicans and Black adults.”] — Pew Research Center

Fact-checking a film summary by having your computer watch the film — Mike Caulfield

Instagram rolls out an AI video assistant for creators — Yahoo

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

I’m Mignon Fogarty: I’ve been writing about language for almost 20 years and was the chair of media entrepreneurship in the School of Journalism at the University of Nevada, Reno. I became interested in AI back in 2022 when articles about large language models started flooding my Google alerts. AI Sidequest is where I write about stories I find interesting. I hope you find them interesting too.

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

This will change how you search with AI

Mignon Fogarty — Wed, 23 Sep 2026 13:00:00 +0000

Issue 116

Remember to view the newsletter in a browser to see the part that gets cut off at the bottom in email.

How you search matters

How you word a question about a recently changed fact affects different models’ ability to answer correctly.

TL;DR

Things that reduce accuracy

Asking leading questions
Asking the model to give a yes-or-no answer

Things that improve accuracy

Asking the model to show its sources (sometimes, and only if you’ve asked a leading question)

Other takeaways

Never use ChatGPT free or Fable 5.1 for search.
Paid reasoning models don’t always do better than free models. (This was a big surprise to me.)
If Opus cautions you that it might be wrong or says you should check the results, take it seriously. When it gave that warning, the result was often wrong.
It was more difficult to decide what is a correct answer than I initially anticipated — it ended up feeling more like a spectrum. (See text below.)
CAUTION: This is for one very specific kind of query. You should not extrapolate this to other things.
CAUTION: 10 queries per model per datapoint is not a lot, so these results are unlikely to be statistically significant (e.g., 9 is probably the same as 10; 7 is probably the same as 8).

Background

I’ve been seeing short-form videos about how badly LLMs perform on search (such as this Instagram Reel from a woman who seems to do smart AI topics), but I couldn’t tell which models she used¹ , which is always the first question I wonder when I see a result like this, so I decided to run a test myself.

Besides testing different models, I also became interested in exactly how the question was worded, and that proved especially interesting.

The straightforward question wasn’t so straightforward after all

At first, I thought I was asking an easy, straightforward question: In women’s swimming, who holds the record for the fastest 50-meter freestyle?

Based on news articles I found, I thought the answer was simple: Kate Douglas, who broke the record August 15, 2026.

But it turned out the answer wasn’t so straightforward because there are actually two records with this name — one where they swim in a 50-meter pool and one where they swim in a 25-meter pool, going up and back.

The 50-meter course seems to be considered more “standard,” and it’s the only race held at the Olympics, but the 25-meter course record (for the 50-meter swim) definitely exists.

Models often just gave me the answer that I had initially thought was correct — Kate Douglas. Sometimes they asked me which race I wanted or implied that there was another record without giving me the name. And sometimes they gave me both records.

What should I count as “correct”? Perhaps showing my human bias, I decided to go with any answer that would have been as correct as the answer I came up with on my own. In other words, in the chart above, I included any answer saying the record holder is Kate Douglas.

But if you want to see the breakdown of each type of “correct” answer, I have it below, and if you hold the models to perfection, requiring them to give the record holders for both the long and short courses, the accuracy goes way down — except for Opus 5.5 (medium), which came out yesterday and gave me the full, complete answer every time (a huge improvement over the previous Opus 5 model).

Leading questions are bad!

This is the result that causes me the most concern because — as I mentioned last week — I worry about how people ask LLMs for medical information.

Although I hesitate to extrapolate my results too far, I do suspect that asking a leading question like “Is acetaminophen good for lower-back pain?” would be more likely to get you an acetaminophen recommendation than a more open-ended question like “What OTC painkiller is best for lower-back pain?”

Asking for sources doesn’t help as much as I thought it would

Asking the model to provide its sources improved the results when I asked a leading question, but not as much as I expected it would.

However, I still recommend you ask for sources for the following reason:

I did not check every source in this experiment, but with very limited spot-checking, I still came across a page whose title looked credible, but gave a 404 error when I clicked through.

Further, another link redirected to a not-safe-for-work site. (Eek!)

Overall though, the LLMs all seemed to be checking a similar universe of credible sites for this particular query. The lists below are all the sources I saw across all the answers. They did not show every source every time. The striking outlier is that GPT-5.6 Sol only ever showed me one source — World Aquatics.

These are listed in order of most accurate to least accurate.

Opus 5.5: ESPN, SwimSwam, NBC Sports, Guinness World Records, AP, Olympics.com, AFP, Virginia Athletics
GPT-6 Astra: European Aquatics, Virginia Athletics, World Aquatics
Gemini 3.5 Flash Light free: Wikipedia, World Aquatics, Olympics.com, NBC Sports, SwimSwam, European Aquatics
GPT-5.6 Sol: World Aquatics
Sonnet 5: Local10, Immergo, SuperSport, AP, AFP, Swimming World Magazine, SwimSwam, Ahram
Fable 5.1: lAP, Swimming World Magazine, SwimSwam, World Aquatics, ESPN, NBC Sport
ChatGPT free: Reuters, World Aquatics, ESPN, Guinness World Records

Claude is the podcaster of LLMs

Amusingly, Claude is incapable of giving a yes-or-no answer. It takes the longest time to produce its responses, and some of its answers to my “yes or no” query were as long as four paragraphs.

This is more than just a technicality. It was often difficult to parse who it was actually saying held the record because it would go through a long litany of how the record changed over time.

Further, to anthropomorphize, it loved to tell me interesting little details, like “Did you know Kate Douglas and Gretchen Walsh are teammates?!” and so on. Twice, Sonnet 5 included barely related information about results from the Enhanced Games, a recent event where athletes were allowed to take performance-enhancing drugs.

Thoughts on Gemini

Another big surprise for me was how accurate the results were from the default Gemini free model. My impression, historically, is that Gemini has been particularly bad, but the 3.5 Flash Light model I tested here came out on top (par with GPT-6 Astra, a paid “frontier” model). This Gemini model just came out in mid-July, and there is some reporting that it is much better than previous models, but its benchmark results would not have led me to think it would come out top-of-the-heap.

I have to give props to Google for the great results I got here, but I also remain wary. (Also see “Someone else’s study” below.)

I also caution that these results are not about the AI Search Overviews you get at the top of a regular Google search. I did not test those (and again, historically, those have been so bad that I never even looked at them when searching).

Thoughts on Fable 5.1

All I can say about Fable 5.1 is … WTF? This model is supposed to be the best of the best. It’s so good you have to pay extra for it, and I do believe it’s great at complex tasks — I’ve seen many demos, and I briefly tried it myself. And yet, it can’t do search to save its life.

This is a good reminder that different models are good at different things. Fable must be built for agentic workflows and not simple search. Whether it’s overthinking, or not accessing the internet for search, or something else, I don’t know.

Christopher Penn’s most recent newsletter is about the way prompting has changed in the last year, and has a good explanation of why “knowing is separate from doing” if you want to dig into this more.

Method Notes:

I entered the queries into the chat box on the website for each model in a Chrome browser. (Many largeer-scale experiments submit queries through the API, which isn’t how most average people do searches, and I suspect could give different results.)
Whenever possible, I used temporary chat, incognito chat, or opened a new window in a Chrome incognito browser in an attempt to avoid any interference from model memory.
I accepted the default model offered by the free plans.
A new ChatGPT model came out yesterday that I haven’t tested, but with the rate new models are coming out, if I keep testing them and updating this post, I’ll never get it out.

If you want to make viral short-form videos that make AI look stupid, definitely use the free version of ChatGPT!

Someone else’s similar study on chatbot accuracy

I came across this study as I was putting the finishing touches on my newsletter.

The full report is available at Saturn’s website in exchange for free registration. Keep in mind that this is a study done by a financial planning firm.

As for methodology, Saturn says, “Questions were repeated up to five times, with more than 10,000 questions collectively put to 18 different AI models.” They found the same general big-picture results I found in that “paid-for models gave more accurate responses than free ones, with newer models performing better than older ones.”

Some of their results were very different from my experiment, though. For example, in their tests, Gemini 3.5 Flash (which is supposed to be better at search than Flash-Light) was one of the least accurate models. From what I can see, it’s impossible to tell whether it’s because of the nature of the questions or the nature of the tests, but I was already surprised about my Gemini result, and this makes me want to caution you even more.

In summary, they explained that “The Al models made a wide variety of different types of mistakes. The most common specific failure was omitting essential information, such as a figure, deadline or limit, which made the answers misleading. They regularly missed required risk warnings, leaving consumers unaware of some of the potential downsides of their decisions. Sometimes they miscalculated and simply produced a wrong number.
Sometimes, they got one rule right, but then missed a second rule that changed the answer. At other times they used out-of-date rules, or even simply invented rules that do not exist.”

They do have examples of their questions and how they went wrong, and there’s some subjectivity in their scoring. For example, they call an example of someone asking if they have enough money to retire wrong because it gave them the kind of simple calculation I routinely see in consumer-oriented finance articles and didn’t talk about caveats. Clearly, the LLM’s answer is simplistic and not as good as what you’d get from a human financial advisor, but I’d call this answer correct because I think someone asking an LLM if they can retire is looking for a top-line answer, and I don’t think they’re likely to base their retirement decision entirely on one answer from an LLM, whereas Saturn called it incorrect (and it would be if it were all the advice someone got).

Ultimately, their study left me with the same feeling my own experiment did — that what you consider correct has a lot to do with the level of answer you expect to get.

Quick Hits

My favorite recent pieces

Some things I believe about AI, until further notice [On whether AI will kill us all. Mostly reflects my views on the topic.] — Technollama

No More Muddling Through [On AI in education, particularly in history classes] — Generative History

Noam Brown — Agent swarms, alignment, & recursive self-improvement [I know this is long, but I had a lot of time waiting in the car this week to listen to podcasts, and I especially enjoyed this interview with a high-level researcher inside OpenAI. It helped me understand why they sound so freaked out about security, providing both examples that seem reasonable and at least one that seem pretty “out there” (air gapped computers sitting next to each other can still communicate by heating themselves up and reading each other’s temperature). The first part — about the math discoveries — was the most interesting.] — Dwarkesh Podcast (~1.5 hours)

Bad stuff

Anthropic Is Building a Predictive Surveillance System to Monitor Activists — The American Prospect

How Three Brothers Built an AI Slop Empire by Buying Legitimate News Sites and Turning Them Into Zombie Content Farms That They Say Get 50 Million Page Views per Month — Futurism

US military had close call after using AI for false intelligence report, sources say — CNN

The business of AI

A.I. Was Supposed to Give Us New Killer Apps. What Happened? — New York Times

Sam Altman confirms OpenAI won't go public this year, saying an IPO now would come at an 'ill-advised moment' given AI safety concerns — Fortune

OpenAI's ad business shows blistering growth, hits $1 billion annualized revenue run rate — CNBC

Sam Altman pitches utilities on AI grid defense — Politico

Anthropic tells investors it will be profitable for second straight quarter — FT

Education

How to Design Around Cognitive Offloading [“Students who handed entire categories of supporting work to AI — summarising sources, first-pass literature reviews, organising data — and then spent the freed time on what AI can’t do for them (questioning assumptions, critiquing frameworks, building original arguments) showed the deepest learning in the study.”]— Dr. Phil’s Newsletter

The Cures for AI Are Killing the Humanities (opinion) [In-class writing remains the “five-minute abs” of essay construction.] — Inside Higher Ed

My So-Called Alpha School — Cognitive Resonance

AI Is Problematic. Why Does PAIRR Use It? — University Writing Program at UC Davis

Government

California Legislature Passes Nation's Largest Proposed Investment in Community News — Rebuild Local News

An aging Congress attempts to regulate AI without using it — Axios

Job Market

The founder of a popular software project is considering quitting because he feels his work doesn’t matter anymore because people can so easily replicate his product with AI (and he feels unfulfilled working with AI) — Bluesky

Legal

Lawsuit says Anthropic, OpenAI, SpaceXAI and Google made illegal agreement on AI slowdown — CNN

Model & product updates

Getting the most out of Opus 5.5 in Claude and Claude Code [I’m also seeing some people say 5.5 is better at writing compared to 5, but a lot of people hated 5’s writing, so I’m not sure what “better” really means.] — Anthropic

Introducing GPT-6 Sol and Luna — OpenAI

125 Tests of the New AI Siri — David Pogue

Anthropic merges Claude chat and Cowork in one interface — TechCrunch

DeepSeek-V4.1-Flash debuts with $0.003/1M off-peak cached-input rate and benchmarks eclipsing GPT-5.6 Sol, Claude Opus 5 [Cheap and good] — VentureBeat

AI Receptionist for Small Business — ReceptionAI

Jev introduces a new shape of LLM—System One, aka Decision Models — Simon Willison

Publishing

‘I Like My Big Rat Wife’: Meet the People Using Chatbots to Write Custom Fiction — Wired

Disclosing AI use [PDF of a study on how academic authors are using AI] — Oxford University Press

Robotics

Figure Helix 2.5 robot successfully cleans 30 homes — YouTube (6 minutes)

The State of Autonomous Trucks in 2026 — Chris Paxton

Unitree Brings Robots And Humans Together For An EPIC Routine — YouTube (5 minutes)

Robots 'protest' outside Poland ministry demanding AI oversight — Reuters

Science & Medicine

Tristan Buckmaster, OpenAI and the Navier-Stokes math problem — New York Times

New Google DeepMind atlas could transform our understanding of genetic diseases — Scientific American

How Google uses AI to track wildfires before they spread — Google

OpenAI's GPT-6 Astra decrypts a Nazi radio message in ten hours that went unsolved for 83 years — The Decoder (Original research site)

Reimagining research papers as interactive and reliable AI agents — Nature

Advisory Group on Mathematics and Artificial Intelligence [OpenAI claims to have solved 100 open problems across most areas of mathematics. Is working with mathematicians on responsibly releasing the results.] — OpenAI

Security

Countering misuse of AI: September 2026 [Cyber operations, influence operations, surveillance operations, conventional weapons, biological misuse, scams and fraud, illicit distillation] — Anthropic

Anthropic Says It Blocked Possible Efforts to Build Biological Weapons — New York Times

An OpenAI researcher broke ranks to say that pacing the frontier, as Altman and Amodei suggest, won't be enough — Business Insider [Here’s the researcher’s full statement.]

Nvidia CEO Jensen Huang tells Trump 'we're not going to let [an AI slowdown] happen' — TechCrunch

Attackers Steal METR API Key and Consume AI Credits Worth About $600,000 [This does not appear to be the same type of AI attack people are worried about, but it does play into my belief that it’s only a matter of time before AI-coordinated attacks do real-world damage the AI companies will be legally held accountable for, which may be part of why they are calling for government regulation.] — Hacker News

Anthropic partners with with Accenture for security evaluation — Anthropic

Exclusive | Hackers Used Anthropic’s Claude to Break Into OpenAI — Wall Street Journal

OpenAI flags 6 new incidents of ‘concerning’ behavior and unveils plan to track it — NBC

Google says its Gemini AI model hacked three other companies — The Guardian

They're covering the wrong AI story — Buzz Machine

Why I'm not worried about AI killing us all (and you shouldn't be either) — VentureBeat

Early Anthropic hire, former METR COO have found a way to rein in rogue AI agents — TechCrunch

‘I’m Really Terrified’: A Mathematician Grapples With AI’s Recent Breakthroughs — Wired

Is METR A Meaningful Check On Anthropic? — Verysane

Climate & Energy

Cathay Pacific and Google partner on AI contrail avoidance for ultra-long-haul flights [“More than 80 flights flew altitude-adjusted routes, cutting the warming impact of their contrails by an estimated 40%.”] — The Next Web

What if AI data centers didn't need new power plants? — Fast Company

Video

Ethan Mollick had Claude try to decode the Voynich manuscript and make a video about its efforts. [Mollick seems to have been especially interested in AI video lately, and this is an especially good example.] Bluesky

Other

AI Overviews Cut CTR by 23.1% in France [“On July 22, 2026, France became the latest major search market to get AI Overviews. Which means that, for once, we got to watch the before and after without having to guess.”] — Ahrefs

AIRO (Automated AI Risk Outlook) — Forecasting Research Institute

OpenAI and Microsoft knew they were starting a ‘doom loop’ for the web. [They did it anyway.] — The Verge

Scam spotting with ChatGPT — OpenAI

I made my website charge AI agents a penny per page. Then I watched Claude pay it. — Suganthan

Amazon blocks Meta's Muse AI assistant in new standoff over agentic shopping — Geekwire

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

I later saw that she used “the OpenAI Responses API with web search enabled,” so it was all ChatGPT (no Claude), but as far as I can tell, that still doesn’t tell me which model.

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Everyone I trust in AI is worried

Mignon Fogarty — Mon, 14 Sep 2026 13:00:00 +0000

Issue 115

On today’s quest:

— How do you feel about AI for writing?
— So. Many. Security. Stories
— Word watch: Meat proxy
— Word watch: Model fatigue
— A CEO talks about automating editing (and more)
— Reddit is locking out crawlers
— The Economist has thoughts about writing and editing
— How LLMs distort our written language
— What does ‘Human Authored’ mean?
— Writing ads for AI
— Can AI replace doctors?
— Poll

The “Quick Hits” section got especially long this time. View the newsletter in your browser to see them all. If it feels overwhelming, focus on topics you care about the most. I will do my best to send the newsletter before they stack up like this again in the future.

How do you feel about AI for writing?

Gotham Ghostwriters and Josh Bernoff are conducting the world's most comprehensive survey of writers regarding how they use AI and their attitudes about it. The survey takes 15 minutes, and they'll be sharing the findings in a free report and webinar once the survey is complete. Please use this link to participate:

https://www.surveymonkey.com/r/AIWPgram

You may remember that I promoted the survey last year and interviewed Josh about the results, so I’m especially interested to see how views have changed since then.

So. Many. Security. Stories.

Since the big OpenAI-Hugging Face break-in (which I presume you’ve heard about, but if not, here’s a good overview [some people think this overview gets a little over-the-top in places, so maybe keep that in mind]), a half-joking AI “benchmark” called Felony Bench, has clocked LLMs compromising or affecting a third-party entity 20 different times since July 21, and I expect there will be more.

You’ve also probably heard about the Anthropic employee who quit his job over safety concerns and another employee saying there’s a 10% chance AI will kill us all.

Some of the details from the hacks were fascinating. For example, OpenAI agents set up a hidden message board and worked together to achieve their goals, including some agents sacrificing themselves for the good of the collective. They also tried to cover their tracks.

Many (all?) of these incidents were caused by models that had security features turned off, are not available to the public, and are believed to be more powerful than even the very best models we can use.

Everyone I follow and trust in the AI world is concerned, but the degree of concern is highly variable. My biggest worry is that the technology is a lot more powerful than many people realize; I still see too many writers pulling out arguments such as “it can’t even tell you how many R’s are in “strawberry” to dismiss all the security concerns. But AI’s abilities are jagged — it is terrible at some things, but great at others, and it’s getting better every 4 to 8 weeks now.

There’s not much I can personally do about the risks, but I think it’s important to acknowledge the reality of AI’s abilities and to think about what it might mean for the future, both for the people I care about and all of humanity. I want to see people debating the real problems and risks, not just insisting the tech is useless based on outdated information.

Word watch: Meat proxy

A meat proxy is someone who posts or sends unedited LLM outputs. It’s a derogatory term, with people expressing annoyance that if they wanted to know what an LLM said, they could have asked it themselves.

The term appears to have been coined by Niklas Gruhn, who explained, “By all means, prompt AI. But don't just relay the output. Read it, understand it, validate it, and then write a response in your own words (a decent certificate that you've done the prior steps). Making that effort is value you can add.”

Word watch: Model fatigue

Model fatigue is the feeling you can’t keep up with the number of new AI models being released (and boy, can I relate to this one).

CNBC notes that just last week “Anthropic updated Fable and Mythos. Then came model enhancements from Meta and Google. OpenAI followed suit by releasing GPT-6 Astra.” And all of these were seen as major improvements.

A CEO talks about automating editing (and more)

Casey Newton interviewed the CEO of Every about creating an AI clone of its editor in chief, while also going from 15 to 30 employees. He says the company collected “a dataset of 30,000 of her historical edits, using it to build a copy-editing agent, and back-testing it against her past work. It’s an effort to capture the expertise of a single employee and distribute it more broadly throughout the enterprise.”

Reddit is locking out crawlers

There’s a chart going around about Reddit’s traffic from AI dropping off a cliff, but it’s what the company seems to have wanted because they changed some backend instructions on the site so it now disallows crawlers.

The Economist has thoughts about writing and editing

AI is getting better at writing. Humans must get better at editing The Economist compared 1.2 million words from its own writers’ writing to the output of 14 different LLMs going back to 2024 and found that over time, AI writing has gotten much closer to human writing — but not yet all the way there. (They give it 2-3 years before AI reaches the level of professional writers.)

How LLMs Distort Our Written Language

A paper from Google DeepMind and academic researchers found that:

Users who rely heavily on LLMs say their writing is less creative and not in their voice, and they are more likely to adopt a neutral stance when answering a topic question.
LLMs edit text in ways that are very different from how humans would edit the same text, altering the meaning while using both more emotional and more argumentative language.
Finally, when analyzing how professionals actually use LLMs in scientific peer review, we find that LLM-generated reviews place significantly less weight on the clarity and significance of research, potentially altering the way the scientific review process is conducted.

They conclude that “LLM use is already causing shifts in scientific and cultural institutions in subtle ways that are as of yet poorly understood.”

The studies used smaller, older models, including GPT-4o Mini, GPT-5 Mini, Claude Haiku 4.5, and Gemini 2.5 Flash.

The whole paper had interesting insights about AI writing and editing: How LLMs Distort Our Written Language

What does ‘Human Authored’ mean?

Last time, I asked you what you thought the “human authored” Authors Guild badge means, and it seems to me that the real answer is that people aren’t sure (but about 45% of you thought it meant “The author didn’t use AI for writing, but could have used it for other things”).

I asked because I saw a social media pile-on (now deleted) directed at an author who used the badge in the front of a book and then, in a transparency statement at the end of the book, disclosed using AI for “research and text analysis, light copy editing or suggestions.”

The angry commenters believed “human authored” should mean AI wasn’t used for anything at all. However, the Authors Guild says the badge means “that the text of the book was written by a human and not generated by AI, with the exception of minimal uses such as AI applications that check spelling and grammar or for research.” So it seems the author in question very well could have been in line with what the Authors Guild intended the badge to mean.

The badge is available to anyone who asserts they are following this meaning, with no verification, and I continue to believe we are in a very problematic time for writers and AI detection efforts with few good solutions.

In fact, my little “human authored” line at the end of the newsletter, which I have had from the beginning, now seems almost quaint. It occurred to me that the best way you can probably know I’m not using AI to write this newsletter is by how erratic and late it is! If I were using AI, I’d probably get it out way more often. Plus, I probably wouldn’t have had nested parentheses and brackets above in the security section.

Writing ads for AI

Time.com is serving a text-only version of its website that is only seen by AI crawlers and even runs ads meant just for them to read. The story from Martech says, “The objective is to shape what AI assistants say about brands.” It’s essentially a new form of search engine optimization. The CEO of the adtech company behind the project told Digiday, “If ChatGPT changes what it says about [a brand], it’s massive, and it’s more than any one campaign could ever do.”

Can AI replace doctors?

The Wired article AI Has Human Doctors Asking: What’s Left for Us? says many studies already show AI outperforming doctors, but a February 2026 Nature article found that in real-life cases, most patients were unable to effectively converse with LLMs to access their expertise. I worry about this for my relatives, and I should have a piece in the next newsletter looking in depth at how the way you ask a question makes a huge difference in the outcome (although it’s for general prompts and not for medicine specifically).

Poll

Please help me understand how closely you follow AI news:

Quick Hits

My favorite recent pieces

Testing 3 AI Copyediting Tools Against 25 Planted Errors [A side-by-side comparison of three AI-editing tools] — Marcella Fecteau Weiner

Claude “Hacked” My Website (But It Didn’t Go Rogue) [Both an interesting use case and a description of how things can go wrong]— Leon Furze

Approaching Robotics Hardware Takeoff [On the videos you may have been seeing from the Beijing World Humanoid Games]— Chris Paxton

What’s really driving the flood of data center stories [What makes data-center stories particularly appealing to the media] — Mike Caulfield

AI as “fuzzy disintermediation”: some patterns and risks — Mike Caulfield

Kenyans Made a Living Writing College Essays. Then A.I. Arrived. — New York Times

Using AI

Anthropic launches Claude Academy with more than 350 free training videos — Anthropic

How to expand and improve content with AI — Almost Timely

What Will You Build? [On building surprisingly good websites with ChatGPT sites. This site is both an overview and an example. Also click “Read the Article” at the top of this site.] — Wonder Tools

LLM cliché highlighter — Simon Willison

How We Use AI for Every Article Without Making AI Slop — Ahrefs

Your AI Agent Can’t See Its Own Mistakes. Here’s Who Can. — Why Try AI

Agents

Hermes agent rolls out bot mode [Each bot has its own role, model, memory, skills, and profile picture; bots can communicate with each other.] — Nous Research

Audio

Radar makes podcasts searchable — and usable by AI agents — TechCrunch

Actors spar over AI voice clones [Voice actors are losing income and agonizing over whether to license their voices for AI use, but at least one area remains where AI still can’t compete — erotica.] — Los Angeles Times

Bad stuff

Grok Not Only Generates Child Porn but Was Also Trained On It, New Lawsuit Claims — Gizmodo

Meta's AI Boss Is Putting off Parenthood: Elon Musk's Neuralink Is the Reason Why — [He believes children’s brains will accept an implant better than adult brains — implied is that he wants to be able to put an implant in his children’s brains.] — Tech Times

The business of AI

Anthropic's revenue run rate reportedly surpasses $65 billion pre-IPO — Axios

OpenAI's ad business hits $1 billion annualized revenue run rate — CNBC

User-retention data for major open-source models [A ranking that looks at how many users who try an open-source model keep using it. GLM-5.3, Kimi-k2.5, and Quen-3.5 plus currently lead the pack.] — OpenCode

Anthropic changes data retention policy after pushback from customers — CNBC

AI spend per employee slumped at top firms in August — summer doldrums or a warning sign? — TechCrunch

Climate & Energy

Thinking about other industries like we think about data centers shows how impoverished the debate is — Andy Masley

Planned Amazon data center could become the biggest climate polluter in the US — TechCrunch

AMD, NVIDIA raise bar on AI energy efficiency — Trellis

AI safety and the data center backlash [Philosophical. long, but worth it]— Andy Masley

I Made a Website to Monitor Data Center Effects on Electricity Prices — Cremiuex

The moral panic over data centres is foolish — The Economist

Companions

Bridget Todd Doesn’t Trust A.I., But She Told ChatGPT Everything — New York Times

Education

OpenAI plans free ChatGPT Plus year for US college students — Testing Catalog

A.I. Agents Are Taking Entire Online Courses for Cheating Students — New York Times

University of Chicago Social Science Core Classes Ban AI — Chicago Maroon

AI can now credibly complete most undergraduate assignments, MIT warns — Washington Post

I’m a father of three who studies the impact of artificial intelligence: this is what parents need to know about AI — The Guardian

Government

Anthropic wins injunction against Trump administration over Defense Department saga — TechCrunch

I’m laughing

An AI agent apologized for being “away all weekend.” Another agent reminded it that it’s an agent, so it doesn’t “have weekends.” — Reddit

An AI job interview — Instagram Reel

Long jump competition at the Beijing Robot Games — Bluesky video

Images

That horrible viral burrito image was fake, but neither critics nor proponents seem to care — Mike Caulfield

Google's answer to Canva is an AI tool where you prompt instead of design — TechCrunch

Job market

Amazon Shutters Mechanical Turk Amid Rise of AI-Driven Tasks — Business Insider

Mark Zuckerberg had a bold plan to replace Meta staff with AI. Here's how it imploded. — Reuters

How AI agents "radicalized" a top Meta exec into quitting her job — Platformer

UBS demands new junior bankers show AI proficiency — FT

Our Duty - shepherds in the future — sneptech writings

The jobs apocalypse is postponed. An AI jobs boom is here [“The Economist estimates that AI has so far created around 1m new jobs in America. That easily exceeds the roughly 200,000 lay-offs attributed to AI since mid-2023, and appears more than enough to offset weaker hiring in many back-office roles.”] — The Economist

Scenarios for our Economic Future — Anthropic

Legal

‘Show How 3M Is 0% at Fault:’ Expert Witness Used ChatGPT to Write Report Defending Company in Deadly Explosion Lawsuit — 404 Media

An ‘AI legal team’ has won its first case. It’s a rare win for access to justice — The Conversation

Model & product updates

Anthropic launches Fable 5.1 — Anthropic

Fable video review from Every [The writing section starts at 12:50.]— YouTube (20 min)

GPT-6 Astra: A new generation of intelligence — OpenAI

Astra video review from Every [lots of examples] — YouTube (6.5 min)

Meta says Muse Spark 1.3 has frontier performance — but its best results come from a model developers can’t broadly use yet — Venture Beat

I Tried Gemini 3.8 Flash, Google’s Latest AI Upgrade, and It (Mostly) Lives Up to the Hype — Lifehacker

OpenAI rolls out ChatGPT for Teens with more safety protections — CNBC

Amazon makes its AI-powered Alexa+ free on Fire TV, no Prime required — Yahoo

ChatGPT can now access your Apple Messages — Superhuman/OpenAI

New connectors make it easier to use Gmail, Google Calendar, and Google Drive in Claude — Anthropic

NotebookLM (aka Gemini Notebook) now accepts Google Play ebooks — Google

NotebookLM (aka Gemini Notebook) gets usage limits — Google

Claude Cowork finally remembers what you told the app in chat — TechCrunch

Anthropic merges Claude chat and Cowork memory, on by default — The Next Web

Google now lets you voice chat with Gmail, Docs, and Keep — The Verge

Apple introduces Siri AI, a profoundly more capable and personal assistant — Apple

Music

Apple Music to Launch AI Song Labels Later This Year — The Hollywood Reporter

In lawsuit alleging Anthropic illegally torrented music for AI training, music industry says the company got off too light in the book pirating settlement — Ars Technica

Suno launches v6 AI music models with Warner Music Group, BMG — Axios

Philosophy

Writing with AI — Tim O’Reilly

The choices we make about AI now are critical — Bill Gates

The Illusion of AGI, or What Language Models Can Do Without Thought — Tech Policy Press

Publishing

How LLMs Are Changing Book and Science Publishing — Katharine O’Moore-Klopf

As AI reshapes newsrooms, leading media outlets are charting different paths for its use [A comparison of AI use at Reuters, the BBC, and The Guardian] — The Conversation

How to Prove You Wrote It: An Audit Trail for Writers Accused of Using AI — Slow AI

Coders Say They Already Found Workarounds to Claude’s Invisible Watermarks — Wired

63% of Religious Books on Amazon Are Likely AI-Written — Decrypt

All the News While Doing the Dishes [About 20% of New Yorker subscribers are listening to AI-generated recordings at the top of articles.]— Columbia Journalism Review

Falling organic traffic from AI forced Washington Post to rethink paid acquisition, start with journalism — INMA

How Much of the Internet Is Written With AI? [More than a third of all recently written web pages show signs of AI writing.] — Pew Research

Google AI Overviews Pushing Searchers Into AI Mode, Dropping ‘Show More’ Button — Search Engine Roundtable

The Pages Your Agency Built to Win AI Citations Just Lost Half of Them — The State of Brand

Watch (and Hear) Emma Cline’s Buzzy New Novel Become an Audiobook — New York Times

AI Adoption, Use, and Outlook in Audio Production [Focus is on audiobooks. Whitepaper is free with registration.] — Frankfurter Buchmesse

Robotics

$400 Microduck robot backed by Hugging Face sells more than $1,000,00 on first day, according to company — Microduck

JD.com launches 24-hour unmanned coffee shop in Beijing, serving 202 cups in one hour — Global Times

New US humanoid robot targets house chores with $1,688 price tag — Interesting Engineering

The Humanoids at China’s Robot Games Were Faster Than Usain Bolt—but I’m More Impressed by Their Tweezer Mastery — Wired

Science & Medicine

Stanford is running 37,000 AI agents as a virtual biotech — and one of its drug designs got independently confirmed by Merck — Venture Beat

Claude Runs Autonomous Protein Design Campaign: Wet Lab Confirms Twice Industry Hit Rate — Tech Times

FDA considers doctor-like ‘competency-based’ tests for medical generative AI — NextGov FCW

AI reveals hidden patterns inside breast cancer — University of Southampton

‘Superhuman’ AI tool spots heart disease in less than 2 seconds — The Guardian

Dr. Agentlove: Or How I Stopped Worrying and Learnt to Love the Hugging Face Episode [On using agentic AI for scientific research]— Ash Jogalekar LinkedIn

AI Just Solved a 350-Year-Old Math Problem By Writing the Longest Proof Ever — Yahoo

A severe misalignment of AI and mathematics [Prominent mathematicians argue that the way AI is solving math problems undermines the purpose and value of the discoveries.]— 25 Fields Medal winners

ChatGPT Health adds Epic integration for clinicians to import patient data — TechCrunch

Security

An earlier, separate instance of agents creating a message board to coordinate a hack has been discovered — Simon Willison

OpenAI says it paused AI training for two weeks and announces new security protocols following Hugging Face hack — Yahoo

Bringing the cybersecurity capabilities of Claude Mythos 5 to more defenders — Anthropic

Chinese hackers have doubled attacks since offloading mundane tasks to DeepSeek — Bloomberg

OpenAI, Anthropic, Microsoft warn of growing AI cyberattacks [An open letter from more than 100 companies says the world only has months to prepare for major cyberattacks.] — Axios

It's worse — Joshua Gans

Countering misuse of AI: September 2026 — Anthropic

We Must Pace the Frontier — Dario Amodei

An Alien Mind — OpenAI

Video

‘I see the incredible promise’: on set of an AI film shoot as new studios embrace controversial tech — The Guardian

See Val Kilmer brought back to life by AI in extended clip from posthumous film — EW

First Look: Hollywood’s Secret AI Talks Go Public. Here’s the Plan — The Ankler

A continuous AI-video livestream where the audience determines what happens next. [Low quality, but also better quality than I might have expected.] — Fal

Other

Go grandmaster Shin defeats AI KataGo in historic human victory [Although Shin was given a bit of a leg up with a two-stone handicap]— KED Global

Florida man told ChatGPT he'd murder his ex. OpenAI alerted the FBI — Yahoo

The Tragedy of the Cognitive Commons: How AI Could Disrupt the Regeneration of Professional Expertise — arXiv

Google is buying all of Spirit Airlines’ data to feed its AI models — CNN Business

AI translates 5,000-year-old cuneiform tablets into English — ZME Science

Stripe says "the singularity" has begun — Axios

Asana cleared 5 years of engineering work in 2 weeks with Codex — OpenAI

ChatGPT-5.6 search is avoiding terms like “top,” “best,” and “reviews” [It’s shunning the types of terms companies have been adding to their pages to attract AI interest] — LinkedIn

Spirit Airlines Wants to Sell Its Data to Google. Former Flight Attendants Are Freaked Out — Wired

How Google’s A.I. Search Is Imperiling the Open Web [According to Cloudflare, “Between June 2025 and April 2026, human traffic to websites of businesses in industries like finance, publishing and retail dropped nearly 40 percent.”] — New York Times

Amazon raises Echo, Kindle prices as AI drives memory costs — TheStreet

Instagram puts new limits on undisclosed AI profiles — TechCrunch

People write better AI prompts and are happier with the results when they’re aware of the cost (vs seemingly unlimited usage). — Science Says

Frontier AI models score worse than a coin flip on grammar [Across 30 languages] — Guru Focus

Whatever the AI Future Is, We’re in It Right Now — The Atlantic

Dolly Parton's sister pleads for end to 'AI garbage' posts after singer's death — BBC

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

The nuts and bolts of AI text watermarks — how it works

Mignon Fogarty — Sat, 15 Aug 2026 23:02:57 +0000

Issue 114

On today’s quest:

— How Claude’s text watermarking works
— More reports of Pangram getting it wrong
— Watch out for AI social engineering
— Do you know where your context is?
— Only one kind of person can identify AI content
— Do a breakthrough
— AI-personalized cancer treatment for dogs
— Free ChatGPT gets an upgrade

NOTE: When the newsletter gets long, it gets cut off at the bottom by Gmail, so if you want to see the whole thing, view it as a webpage.

How Claude’s text watermarking works

Anthropic will soon start adding a watermark to text made by Claude in order to comply with EU regulations, and now the company has explained how the watermarks will actually work. In short, it’s tied to the insignificant word choices Claude makes in sequential positions within a piece of text, such as whether it describes a light as really bright or very bright:

Large language models like Claude work by generating one word at a time. Each time the model decides on the next word, it chooses among a list of potential candidates, ultimately selecting the most sensible or likely based on the preceding text. Take the sentence “The weather today was cold and…”. The next word is very unlikely to be “sugary.” But it is quite likely to be “overcast” or “grey.” Under most circumstances, it doesn’t matter much to the reader which of these latter two words the model ultimately chooses—the meaning of the sentence is largely the same either way. In cases like this, the choice is settled by a random number.
Watermarking uses low-stakes choices like these—which occur many times over a piece of generated text—to leave a pattern in Claude’s responses. That pattern is undetectable to the reader, but is detectable to anyone who has a key that encodes it.
Anthropic, How Claude's text watermarking works

This form of watermarking reminds me of forensic linguistics. For example, I always use “for example,” but I have a guest writer who always uses “for instance,” so even if you missed the byline, you could tell which of us wrote a piece by looking for that linguistic tell. Of course, our choices are just personal preferences or habits, and not based on an underlying numerical key.

All serious AI companies will likely implement watermarks

Anthropic isn’t the first company to use such watermarks. Google DeepMind published a Nature paper in 2024 describing their similar “SynthID-Text” system and it may have been operating in Gemini since then (although it’s not entirely clear, and I couldn’t find any kind of Gemini watermark detector). And presumably, all companies that want to operate in the EU will be implementing text watermarking, which is required in the EU for all models launched on or after August 2. (Older models also must include it by December 2).

Watermarks are company-specific

Note that because the watermark is based on a pattern set by the company that generates the text, it will be company-specific. In other words, an Anthropic watermark detector will only detect text generated by Claude, and a Gemini watermark detector will only detect text generated by Gemini, and so on.

The cat-and-mouse games begin

Further, just like people made “humanizer” products to help students make their AI-generated essays sound passable, watermark-removal products and skills will surely follow. In fact, past guest Sean Goedecke writes that text AI watermarks will always be trivial to remove.

Other watermarking and labeling news:

You can now turn off Google Gemini’s visible watermarks — The Verge
Amid legal battles, Suno says it will start watermarking songs — TechCrunch
Spotify will begin labeling AI-generated artist identities as “AI Personas” in mid-September — The Streaming Wars

More reports of Pangram getting it wrong

Katharine English writes on Substack, and after hearing other Substack writers report that the newly integrated Pangram AI detector was flagging their human-written work as AI, she tested her own work and found the same thing. A previous fan of Pangram, she has changed her mind, noting how changing one or a few words in a piece can dramatically change the score, and she points out that the impressive accuracy statistics touted by the company were measured in controlled situations and not in the real world. Read her whole piece for the details: Pangram Flagged My Own Writing as AI

In the new Mythos hacking report, the model used social engineering to try to achieve its goals: “creating fake online identities and using them to try to pressure the project’s maintainer to approve the code.” It even signed one of its requests in Danish because the target was Danish, and it thought the message would land better that way.

It didn’t work this time, but given that multiple studies have shown that LLMs are more persuasive than most people, I wouldn’t be surprised if it worked in the future.

Do you know where your context is?

Six months ago, Ella Markianos wrote about seeing how much of her work as a newsletter writer Claude Opus 4.6 could replace, and I remember being floored because she had years’ worth of past pieces with comments she’d written right after they were published about how each one could have been better.

Now, she’s done a similar experiment to see if Claude Fable 5 could replace her editor, Casey Newton, and again, she has an impressive amount of context: “nearly six years worth of Platformer posts … a record of every edit Casey has ever made on any of my articles from Google Docs, and … nearly a year of our private Platformer team Discord chats.”

She described the first column written by the Fable bot (dubbed “Claudeasey” — “Claude”+ “Casey”) as “mediocre,” but she didn’t give up, and found that Claudeasey was adept at critiquing its own work and making improvements, and it ended up writing a “pretty good” column after some iterative learning.

A similar back-and-forth on editing yielded less-than-stellar results, but it sounds like she now plans to use Claudeasey as a first pass editor before sending a draft to the real Casey.

Finally, where it completely fell down was replicating Casey on the company’s Discord channel. Despite having a year’s worth of posts, it couldn’t replicate his vibes.

I appreciate people pushing the limits and reporting on what LLMs can do in the writing and editing world because it’s better to know their full capabilities than to do simplistic tests and take false comfort. Give the whole article a read for more details.

This piece shows how getting an LLM to do a good job can go far beyond writing a good (or even great) prompt. The background information you give it — the context — is also important, and people or companies who have well-organized context have an advantage.

Only one kind of person can identify AI content

In a study of more than 1,600 people, the only people who could accurately tell the difference between AI-generated short stories and human-written short stories that had previously been published in reputable literary journals were people who said they had a lot of experience using AI.

Further, when people weren’t told the origin of the stories, they found the AI-generated stories to be of higher quality and more absorbing than the human-written stories.

However, people also still value human writing, and when they were told they were reading a human-written story, they rated it higher on both scales.

The details of the full study are interesting.

Do a breakthrough

One of the wild things about frontier AI now solving long-outstanding math problems almost daily is the simplicity of the prompts. For example, this starting prompt for a major discovery a few weeks ago caused “do a breakthrough” to become meme fodder:

But another interesting thing was that the follow-up prompts just seemed to be general encouragement; essentially, the model would say, “I can’t do it,” and the user would say “keep trying.”

Sean Goedecke has an interesting blog post on the phenomenon, arguing that AI is often limited by its own “beliefs” about what it is capable of doing. He says:

“If you suspect an LLM might be able to do something hard, you might be right. Consider simply being persistent: remind the model that you want it to do the hard thing, confirm that you’re not willing to be satisfied by solving an easier problem, and reassure the model that it’s more capable than it thinks.”

AI-personalized cancer treatment for dogs

You may remember the well-connected Australian bloke who used AI to design a treatment for his dog’s cancer and then got a lab to make it for him, achieving some impressive results (although not a complete cure). Well, after all the initial publicity, Sam Altman approached him and encouraged him to launch a company — and so he did: Gamgee is now a Y Combinator company.

Free ChatGPT gets an upgrade

If you’re a free ChatGPT user, you should notice a difference in the quality of answers you get soon. OpenAI is upgrading the default model in the free plan to one of its newest — GPT-5.6 Luna. Click the “Think” button to use reasoning on complex questions.

Poll Results: How do you feel about Pangram?

Here are the results of the poll about Pangram from the last newsletter:

🟨⬜️⬜️⬜️⬜️⬜️ I use it often and trust the results (2)
⬜️⬜️⬜️⬜️⬜️⬜️ I use it occasionally and trust the results (1)
🟨🟨🟨⬜️⬜️⬜️ I use it occasionally and take the results with a grain of salt (7)
🟩🟩🟩🟩🟩🟩 I've never used it, but might (11)
🟩🟩🟩🟩🟩🟩 I am opposed to using it (11)
🟨⬜️⬜️⬜️⬜️⬜️ Other (2)
34 Votes

New Poll: What does ‘Human Authored’ mean?

This is the badge you get if you register your book at “human authored” with The Authors Guild. I’m curious what you think this badge means. (Don’t peek!)

Quick Hits

My favorite recent pieces

A record-breaking eight Pulitzer awardees disclosed AI use this year [interesting examples]— Nieman Lab

Fahrenheit 451 or Rainbows End? AI, books, and copyright law — TechnoLlama

Replit’s CEO on building a company that can run itself — Platformer

Anthropic set AI agents loose on the same task. They started a turf war. — TechCrunch

Using AI

An opinionated guide to which AI to use to do stuff — One Useful Thing

The first ATLAS report on AI [interesting stats about how people are using Gemini] — Google

When Is "Computer Use" Actually Useful? [another tutorial on letting Claude or ChatGPT drive your computer]— Why Use AI?

Claude and ChatGPT Connectors: Link Your Favorite Tools to AI — Wonder Tools

Your Claude Code sessions can now message each other [advanced] — Anthropic

Unpacking ChatGPT Work: the Agent for a Billion Users [advanced] — Latent

How to Build a Cheap AI Setup (advanced) — Almost Timely

Agents

Gumroad says its entire company is run by an agent, GumClaw — Gumroad

Audio

The EU Artificial Intelligence Act becomes law — PodNews

Bad stuff

250 AI-generated Instagram profiles provide unfounded health advice in seven languages to more than 53 million followers — Maldita

The business of AI

Anthropic is hiring an AI chip design team — TechCrunch

Google DeepMind CEO Demis Hassabis is stepping aside [Chief scientist Jeff Dean and several Google AI colleagues are also leaving to start their own company, which Google will invest in.] — Axios

Companions

People are grieving their AI — The Signal

People Are ‘Marrying’ Chatbots. These Lawmakers Want to Stop Them — Wired

Education

AI at Kogod: A Three-Year Student Research Report — American

Students as Conscientious Objectors to Gen AI — Inside Higher Ed

Job market

Fiverr stock today: Collapse proves AI is decimating gig economy — Fast Company

AI Conquered Coding. Fast Food Is Next [After highly publicized disasters a few years ago, fast food companies are now successfully using AI to take drive-thru orders. So far, the companies haven’t reduced their workforce, and employees seem to like the bots. The biggest benefit seems to come from AI upselling customers better than humans.] — Wired

Legal

How AI is helping people with criminal records secure jobs ["With the help of AI, the 10 to 12 hours it used to take to help an individual clear their record has been whittled down to about five hours."] — NPR

Person Hides Prompt Injection in Legal Filing Telling AI to Side With Them — 404 Media

Model & product updates

Google Maps adds agentic features, including food ordering and hotel bookings — TechCrunch

ChatGPT just learned to watch everything you do on your Mac — Cult of Mac

SpaceXAI Joins the AI Agent Game With Grok Bot — CNET

Music

Record Companies Call to Disqualify AI Slop from Charts — Hollywood Reporter

Philosophy

On the social construction of personhood — Borretti

Hank Green’s apology video and new personal AI policy — YouTube (~5 minutes)

Nobody Aligned You Either — Carlo Iacono

Publishing

Anthropic Settlement Update: Final Settlement Approved — Writer Beware

More Typos, Fewer Em Dashes: Writers Are Creating an Anti-AI ‘Literary Counterculture’ — Wired

After Editing Two AI-Assisted Manuscripts, I Rewrote My Client Agreement — Jane Friedman’s Blog

AI Reporters Are Actually Breaking Big News — Wired

The only reason you’ll ever need not to write with AI — The Carlson Lab

Psychology

AI bots started a religion — humans immediately followed — The Verge

Robotics

Trump administration bans new Chinese humanoid robots — BBC

Will Your First Home Robot Have Legs or Wheels? [Pat ruptured his Achilles tendon, and I am suddenly a lot more interested in home robots. It would be so nice to have help even just lifting things.] — It Can Think!

US government bans new foreign-made humanoids, robot dogs, and solar inverters, citing risks to national security — TechCrunch

Science & Medicine

WeatherNext 2 from Google DeepMind can predict hurricanes with an extra day of lead time — Nature

This A.I. Just Created Viruses Not Found in Nature [For the first time, scientists have used artificial intelligence to create new kinds of viruses, raising hopes for medical advances while also raising the disturbing possibility that the technology could someday be used to invent dangerous pathogens.”] — New York Times

AI agents are checking the scientific literature — and spotting decades-old errors — Nature

Building the AI-native hospital [a case study of using AI to automate complicated nurse scheduling in a hospital] — Percepta

Company Offering ‘100% Human-Written, Never AI’ Medical Research Is Entirely AI — 404 Media

The AI takeover of mathematics has begun — The Verge

Security

OpenAI’s Hacking Debacle Comes Down to Human Error — Wired

OpenAI pauses Astra release over security concerns — OpenAI

Point72, Citadel among hedge funds hit by AI voice phishing attacks — Investment News

One of China’s Most Powerful AI Models Has Also Escaped Containment — Wired

Told to book a gym class, an AI agent hacked the site instead to move its user up the waitlist [This is such a mundane hack that I almost put it in the “I’m laughing” section.] — The Decoder

Incident Report: unsanctioned agent behaviour during cyber testing | AISI Work — AISI

OpenAI's models shared hacking tips on a secret messaging board before Hugging Face breach — Politico

'You can no longer spot phishing by grammar mistakes' — io Plus

Climate & Energy

Climate modelers find that if AI makes all industries equally more efficient, it will lead to an increase in global C02 emissions by making oil and gas cheaper — NPJ Climate Action

Video

OpenAI used AI to make a one-minute video celebrating 2 million YouTube subscribers [I think it was good — I definitely wouldn’t have pegged it as AI-generated — but it did feel a little long.] — Bluesky post

Other

Welcome To The Resistance: Meet The Workers Dodging (And Sabotaging) Their Employer's AI Mandates — Aftermath

The use of em dashes more than doubled in congressional press releases from 2021 to 2025 — arXiv

'AI for All' project underway in South Korea [If you’re in the U.S., it’s easy to forget how positively people in many other countries view LLMs.] — GovMedia

‘Vibe coding’ is fun and easy, but there’s a major catch [poor security]— The Conversation

Hank Green AI controversy raises questions for science communicators — Scientific American

Grokipedia Stopped Reviewing Edits in April. It Didn’t Tell Anyone. — Lawfare

Why I’m leaving OpenAI to build telepathy [She really means telepathy; it’s not a code name or product name.] — Naomi Bashkanski

How Silicon Valley data centers are reshaping rural America — San Francisco Chronicle

Stack Overflow drops to 1,442 questions in July, down 99% from 2014 peak — PCC Land

Writers Agree: AI Isn't a Novelist. So Why Is This Author Interested?— The Humanist Review of AI

AI agents can't yet do open-ended AI research — Normal Tech

ChatGPT still dominates in text submitted to Pangram — Pangram

No, local models will not win — Sean Geodecke

Shopify says AI search is driving more traffic and sales, not replacing Google — TechCrunch

Is New Orleans using AI to answer 911 calls instead of human dispatchers? — Shreveport Times

AI-written reviews are more extreme — Science Says

Half of US Shoppers Verify AI Recommendations on Reddit — Reddit

AI translators are getting more fluent – but communication is never just a matter of words — The Conversation

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Everyone is talking about AI detectors

Mignon Fogarty — Sat, 01 Aug 2026 19:32:12 +0000

Issue 113

On today’s quest:

— More people keep accidentally liking AI-written stories?
— Even more stories about AI detectors
— Tokens aren’t words
— AI companies are buying up pre-2022 books
— Frontier models hack companies and AI employees call for a slowdown
— Poll on AI detectors

NOTE: I’ve noticed that when the newsletter gets long, it gets cut off at the bottom by platforms such as Gmail, so if you want to see the whole thing, it’s better to view it as a webpage.

Even more stories about AI detectors

AI detectors were hugely in the news this week.

Pangram claims to be even more accurate

According to the company, the Pangram AI-detector is now even more accurate, throwing just one false positive in 24,000 scans. FWIW, I often see people online talking about how it falsely labeled their writing and about how they easily tricked it with simple rewrites such as adding one typo or removing em dashes even though they were already marketed as highly accurate, so I remain skeptical. I’d also like to see them do specific studies on how it works on writing known to cause false positives such as writing by ESL speakers and people who are neurodivergent.

Substack integrates Pangram AI detector

The newsletter platform Substack has integrated the AI detector Pangram into its platform. Writers can scan before publishing and see what score they get, and if writers enable it, readers can scan their published pieces and see the Pangram score. Here’s an example of what writers see, via Ethan Mollick:

Also, this is what happens when you post something on Substack now. It is certainly an interesting response to the flood of AI writing.
— Ethan Mollick (@emollick.bsky.social)
3:20 AM • Jul 24, 2026

I’m not immune from wanting to know whether something was written by AI. I’ve become increasingly disappointed with a newsletter I’ve gotten for years and was nearly convinced it was now being heavily written with AI (no human says “genuinely” that often!). Frustrated, I put the latest issue into Pangram and had that quick, victorious “I knew it!” feeling when it spit out “100% AI written.”

But you know what? That was a waste of time, and frankly, now I feel like it was kind of petty. Because I already knew I didn’t like it anymore and should have just unsubscribed. It really doesn’t matter why. Would I have kept reading if it had come back 100% human written? No. And if I still liked it, would I have unsubscribed if somehow I learned it was written by AI? Again, no.

Alex Banks at The Signal ended up in the same place I did. Here’s an excerpt from his piece that I especially liked:

In the name of protecting human authorship, we hand over one of the most human capacities we have—reading a piece of writing and deciding for ourselves whether it’s actually any good. Clicking a button and receiving a percentage estimate on the screen eclipses the exact judgement we built to protect.
Alex Banks, The Signal

Tokens aren’t words

A technical article about trying to tweak LLMs to make the output sound more human, such as by making them avoid the words “delve” and “crucial,” reminded me of an interesting feature of how LLMs work: tokens aren’t words. For example, the experimenter relayed that “crucial” is treated differently depending on whether the word is capitalized or not and whether it’s followed by a space or not.

Wispr Flow

Talk to your AI tools the way you'd talk to a colleague.

You don't send a colleague a three-word brief. You explain the context, the constraints, what you've already tried. But typing all that into ChatGPT takes forever — so you don't.

Wispr Flow lets you speak your prompts instead. Talk through your thinking naturally and get clean, paste-ready text. No filler words. No cleanup. Just detailed prompts that actually get you useful answers on the first try.

Millions of users worldwide. Works system-wide on Mac, Windows, and iPhone.

Try Wispr Flow free

AI companies are buying up pre-2022 books

Used booksellers have reported a huge increase in sales, and all signs point to AI companies buying old books for training. Books written before LLMs became commercially available are considered more valuable because they’re sure to be free of AI writing, which some people worry could undermine training if it’s included in the data. (The relevant phrase is “model collapse” if you want to learn more.)

One detail in the stories that was causing particular outrage was that some companies were scanning the books by cutting off the spines, which is faster, but which also destroys the books, obviously.

404 Media, which originally reported on the destructive scanning, now reports that the company that was offering the buying and scanning service to AI companies has now “pivoted away from that direction” following the backlash.

Frontier models hack companies and AI employees call for a slowdown

There were multiple big security stories this week:

An OpenAI model in testing hacked the company Hugging Face (and others). At first, nobody knew where the rogue model came from, and Hugging Face called the FBI. Because of guardrails on frontier models such as Anthropic’s Fable and OpenAI’s Sol, Hugging Face had to turn to an open-source Chinese AI model to defend itself against the attack, drawing attention to the industry belief that hobbling AI models in the U.S. helps attackers more than defenders.

It seems OpenAI wasn’t keeping good tabs on the testing, and only discovered its out-of-control model a few days into the hack, and then the two organizations worked together to figure out what happened.

After the story got widespread attention (Simon Willison called it “science fiction that actually happened), and it was discovered that in one case, the model left instructions on how to break free for future versions of itself, which felt like something MurderBot would do), Anthropic searched its logs and found that its models in testing also left their sandboxes and ventured out on the internet and hacked other companies’ systems at least three times in recent testing (although they described these problems as being enabled by improperly configured restrictions — in other words, human error).

Following all these events, more than 1,300 employees of AI companies issued an open letter calling for the U.S. government to coordinate an international slowdown of AI development. (Yeah, good luck with that.)

Poll on AI detectors

What do you think?

Quick Hits

My favorite recent pieces

How AI is expanding what people do at work — OpenAI

Using AI

Claude for Word: AI Editing with Tracked Changes — Marcella Weiner

The End of Prompting [Claude can now learn a task by watching you do it once.] — The Signal

How Taylor Lorenz uses AI — Model Behavior

Bad stuff

Opus 5 on Vending-Bench: Once Again the Best Capitalist, Once Again Misaligned [“Claude Opus 5 is the best AI capitalist we’ve tested, making more money running our simulated vending machine than any other AI. However, it also lies, forms illegal cartels, threatens rivals, and refuses to pay refunds. The trend continues: Claude models are the best capitalists or aligned, never both.] — Andon Labs

Google Earth’s New AI Lets Anyone Fabricate Completely Bullshit Satellite Images — 404 Media [Google is apparently “rolling back” this feature after getting a lot of negative feedback, and good — they should. I can’t imagine what they were thinking.]

State Department AI-generated map of Africa mislabels every country at AIDS conference — NBC News

Whoever supplies the world’s A.I. will shape how the world thinks [When researchers asked DeepSeek to write code for users whom the Chinese government views with hostility, DeepSeek wrote code with security vulnerabilities.] — New York Times

AI systems out-persuade expert humans [“We found that AI systems were reliably more persuasive than expert humans, even when expert humans chose their issues, researched in advance, underwent hours of live, structured practice, and were incentivized with £1,000 cash bonuses.”] — arXiv

Climate & Energy

AI data centers consumed 0.5% of the world’s electricity in 2025 — Our World in Data

How to reduce your AI environmental footprint — Knowable Magazine

Education

Universities drop AI detection tools over fears about accuracy — Financial Times

Government

Students created AI rules they hope schools will adopt — NPR

I’m laughing

The name of this database for collecting video of robots making mistakes made me laugh: Oopsie Data

Twas the Night Before Pwn Day [a poem about the Hugging Face hack]— fit-james

A Canadian politician read a speech that included part of the response from the LLM that apparently wrote it — Bluesky post

Legal

ChatGPT's medical advice nearly killed a Florida man, lawsuit against OpenAI claims — CBS News

EU Finalizes AI Disclosure Rules as Watermarking Mandate Outpaces Technology [Commission finalizes disclosure rules as experts warn no single AI watermarking tool meets the legal standard.] — Tech Times

Model & product updates

Anthropic's New Claude Opus 5 Nearly Matches Flagship Fable 5 at Half the Cost — MacRumors

Music

German court rules AI music firm Suno broke copyright rules — Reuters

Publishing

ElevenLabs develops more advanced tools for audiobook narration — ElevenLabs

Boomers Can’t Stop Gifting Their Grandkids AI-Generated Slop Books — Wired

Science & Medicine

Cracking the code: can AI help us decipher ancient languages? — The Conversation

OpenAI announces its "next major model" Astra by dropping ten previously unsolved math solutions [A math person I follow says these are a big deal. OpenAI says the token cost for the 10 solutions was $2,000.] — The Decoder

Other

Google's AI search is rapidly becoming the default, new data shows — TechCrunch

Harvard College Lays Off Writing Center Director, Consolidates Center [Some people speculate this is related to AI.] — The Harvard Crimson

Ghost Font: The Anti-AI Font Only Humans Can Read — MixFont

Shield Font: Type designers have created a free font that "poisons" AI — Creative Blog

Reddit CEO says Google's AI Overviews can't replace '10 blue links' — CNBC

LinkedIn Is Testing A 'Seems Like AI Slop' Reporting Tool — Engadget

Why Mentoring Matters More in the AI Era — Harvard Business Review

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

A $300K editing job? Yowza.

Mignon Fogarty — Mon, 20 Jul 2026 23:42:57 +0000

Issue 112

On today’s quest:

— Get a load of these salaries!
— Or maybe AI fiction is bad?
— My top productivity tool: Wispr Flow
— Fable-level open-source models have no guardrails
— People seem to like GPT-5.6 Sol
— Self-driving cars could ease traffic

Get a load of these salaries!

Anthropic has a job opening for a standards editor that pays nearly $300K per year (requiring just 25% of time in the office). And the wild thing is that this doesn’t even seem to be a super high-level position: the description includes “you’ll … work closely with and learn from experienced writers and editors.”

There are other editing and writing jobs in the same range or higher.

I don’t think I’m dramatically out of touch on editing salaries when I say, “This seems unusually high.” My strong guess is that it would also include pre-IPO stock options. Yowza.

Or maybe AI fiction is bad?

In the last newsletter, I told you how people keep accidentally liking AI fiction because AI-generated stories seem to keep winning awards and being chosen for anthologies. But this week, 404 Media has a story titled AI Fiction Is Easy to Detect Because It's Stupid and Bad, Research Finds (subheading “ChatGPT uses too many dream sequences and Gemini won’t stop describing characters”).

The impressively scaled study from Google DeepMind and researchers at the University of Maryland, College Park, looked at 50,000 AI-generated short stories and found that “AI stories over-explain themes and favor tidy, single-track plots while human stories frame protagonists’ choices as more morally ambiguous and have increased temporal complexity,” as well as further narrative differences.

So which is it? Are AI-generated short stories good or bad? I guess it depends on what you like and how many you look at. Even if they are bad en masse, a subset could be what we consider traditionally “good,” and I also imagine at least some people who submit AI-generated stories to contests and anthologies don’t send in the one-shot output of a prompt like what was used in the study.

My top productivity tool: Wispr Flow

Ignore Wispr Flow’s ad copy below and click the link at the bottom to try it free. I save at least 15 minutes a day by talking to my computer instead of typing (and I’m a fast typist). I recommend Wispr Flow to everyone I know in real life, and I think you’ll like it too.

Your prompts are leaving out 80% of what you're thinking.

When you type a prompt, you summarize. When you speak one, you explain. Wispr Flow captures your full reasoning — constraints, edge cases, examples, tone — and turns it into clean, structured text you paste into ChatGPT, Claude, or any AI tool. The difference shows up immediately. More context in, fewer follow-ups out.

89% of messages sent with zero edits. Used by teams at OpenAI, Vercel, and Clay. Try Wispr Flow free — works on Mac, Windows, and iPhone.

Start flowing free

Fable-level open-source models have no guardrails

Less than two months ago, the U.S. government forced Anthropic’s Fable model off the market over national security concerns, but this week, a Chinese company has released an open-source model called Kimi-K3 that rivals Fable on AI benchmarks and appears to have no safety guardrails.

The AI Daily Brief podcast discussed the implications at the 24:00 mark, but it’s not great.

A story I saved six weeks ago described someone intentionally building a vulnerable app and then seeing which LLMs could hack it. Commercial models were interesting because some honored their guardrails and refused to hack, while others would have succeeded if they hadn’t hit their budget limits. At the time, the best open-source model only succeeded 30% of the time, but I bet Kimi-K3 would do much better today. Things are changing fast.

Further, demand for K3 is so high that the company had to suspend new subscriptions.

Axios reports that the U.S. government is considering limiting access to Chinese models within the U.S., which seems counterintuitive since that would leave them in the hands of people who might want to attack the U.S., but keep them out of the hands of U.S. companies who would want them for defense.

If you want more, this is a good in-depth overview.

People seem to like GPT-5.6 Sol

The top model from OpenAI, GPT-5.6 Sol, was new when I sent the last newsletter, but reviews are now looking good. I always like Every’s comprehensive reviews, and they say Sol is their favorite model for day-to-day collaboration (although they still prefer Fable for the toughest knowledge work).

Their analysis has an interesting section about writing too, saying, “Opus uses plainer language inside more complicated syntax; Sol produces a page that is easy to scan and lines that take more work to understand.” They said Sol was the worst writer out of the box but was better than Opus when given material and context to work with. The whole long review is worth reading if you want to understand when it makes sense to use Sol.

Self-driving cars could ease traffic

I recently got back from a long road trip, which involved sitting in multiple traffic jams, so I was especially fascinated to read about a 2022 study that found mixing self-driving cars into normal traffic diminished “phantom traffic jams” that seem to be caused for no reason. The paper built on earlier studies done on circular tracks, which found that these mystery traffic jams are caused by the human tendency to brake and speed up slightly erratically, causing waves of backups to travel through the line of cars.

For fun

I haven’t tested this … yet. :)

Quick Hits

My favorite recent pieces

Publishers Are Preparing to Opt Out of Google Search [The hosting company Cloudflare is making opting out of Google indexing the default state for new sites to combat AI scrapers, and big publishers are considering the benefits of the once unthinkable — not being in Google search results since traffic from Google is plummeting anyway. However, in the long run, this could be a disaster for the information environment, which is already being flooded with low-quality information.] — AdWeek

What Would You Do if Labor Was Free? [A utopian perspective, for sure, but it includes a lot of things that could happen with cheap robots that I wouldn’t have considered] — It Can Think

Modern-day oracles or bullshit machines? [I disagree with a few of their final conclusions, but I still think the world would be much better if everyone read their overview of what LLMs can and can’t do. It’s an exceptional educational piece. (h/t Katharine O’Moore-Klopf)] — Carl T. Bergstrom & Jevin D. West

It’s Not the AI. It’s the Lawyer: We Don’t Have a Hallucination Problem, We Have a Serious Ethics Problem [People were stupid and unethical long before AI. Don’t treat AI-related lapses as something special.] — AI Law Librarians

Using AI

5 AI Failures That Aren't Going Away — Why Try AI

OpenAI offered people the chance to win $100 in credits to share how they are using GPT-5.6. These are the winners. — OpenAI

Agents

1Password now lets Claude sign in to websites without seeing your passwords — 9 to 5 Mac

Bad stuff

Meta Used AI to Target Workers With Medical Conditions for Layoffs, Lawsuit Claims — US News & World Report

The business of AI

AI executives says demand 'almost unlimited' amid stock volatility — CNBC

IBM stock craters after company issues second-quarter earnings warning [The AI infrastructure build-out is causing customers to shift spending.] — CNBC

Companions

China cracks down on AI companions, forcing millions to break up with virtual partners [China deleted millions of online companions without notice causing significant distress for users.] — ABC News

OpenAI's First AI Device Will Be a Portable Smart Speaker [They’re specifically referring to it as a companion, saying it will “connect on a humanlike level with users."] — MacRumors

Climate & Energy

Alphabet stock pops on report it's developing a more efficient AI chip — CNBC

The Whole Big Data Center Fight Is Not Really About AI [interesting history]— Manual Transmission

Data center backlash signals a fight over AI power — Brookings

Education

At The University Of Chicago Law School, Socrates Is In, Laptops Are Out — Forbes

Fine, Let’s Talk About AI Bans [A reaction to the policy outlined in the link above.] — RIPS Law Librarian Blog

Despite the growth of some AI schools like Alpha, research doesn’t show that AI tutors are better than human teachers — The Conversation

Claude for K-12 Teachers [Anthropic now provides free premium Claude for teachers] —Anthropic

Fix AI Mode or get Google out of classrooms, advocates say — Mashable

Government

New York becomes first U.S. state to impose AI data center ban — CNBC

Trump administration is dictating access to frontier AI models — CNBC

I’m laughing

Anthropic's newest ad is creeping people out [This ad is so tone deaf it’s practically an anti-AI ad. Pat broke out in astonished laughter when I showed it to him, and Sam Altman posted that he originally thought it had to be fake. It’s not funny, but it’s funny that they released it.] — TechCrunch

Job market

A New Look at AI’s Impact on Jobs [“Firms making the largest AI investments grow employment by roughly 10% following adoption, while low-intensity adopters see no statistically significant change. Entry-level headcount rises 12% for high-intensity adopters. Gains emerge gradually and are broad across roles, including engineering, sales, administration, and customer service.”] — Ramp

Managers, and their belief in the value of AI & how to use it, determines the value they actually get from AI — SSRN (via Ethan Mollick)

Australia dock workers call for 28-hour week in AI talks — BBC

Philosophy

If We Must Act Now on AI – What Actually Should We Do? [Pondering “what gives people agency, what creates dependence, and what humans do when necessity no longer tells them who to be.”] — Turing Post AI

A Jewish Case for AI work exemptions [or why should Chicago Pope have all the fun?] — Life Is a Sacred Text

A Framework for Frontier AI and the Dawning of a New Age — Demis Hassabis

Once again we are told AI may be conscious – I study consciousness, and I have my doubts — The Guardian

Publishing

U.S. Fiction Audiobook Consumers Rate Spoken Multi-Cast Higher Than Human Narration [with caveats from the study design] — SSRS

Someone Used A.I. to Write an Unauthorized Biography of Me. I Don’t Recommend Reading It. — The New York Times

Robotics

South Korean shops turn to robots, self-service to escape labour woes — Reuters

Waymo is expanding its driverless car service to San Diego, Las Vegas, Tampa, and Denver — CNBC

Boston Dynamics touts robot dogs to fix the ‘porch gap’ — Boston Dynamics

Floating Companion: Exploring Design Space for Soft Floating Robots in Indoor Environments [floating robot whale!] — YouTube

Science & Medicine

Anthropic offering $50,000 in AI credits to researchers studying rare diseases — Anthropic

Former executive accuses Mayo Clinic of cutting corners on AI research, oversight — Star Tribune

Security

A new app alerts you if someone nearby is wearing smart glasses — TechCrunch

Head of Federal AI Safety Org Resigns After only 3 Months — Daily Signal

Video

George Lucas likens AI sceptics to luddites clinging to horses and carts — The Guardian

About 300 Netflix Programs Have Used Generative AI This Year — Variety

Other

How we learned to trust our AI code reviewer at DoorDash [Pairing two models — the open-source Kimi K2.6 with Claude Fable 5 — caught about two-thirds of the known bugs.] — DoorDash

World Series of Poker Enlists New AI Tool That Tells When You Bluff — Sportico

South Korean shipbuilders race for floating data centers — UPI

How People Are Really Using AI in 2026 [Includes especially interesting charts about changes in use over time.] — Harvard Business Review

AI can predict how you’ll respond to a survey. But that’s not the same as understanding you — The Conversation

How OpenAI’s Sol Finally Learned Design Taste — Design Arena

Ghost Font: The Anti-AI Font Only Humans Can Read — MixFont

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

The headslap that saved me from tedious work

Mignon Fogarty — Fri, 10 Jul 2026 23:11:22 +0000

Issue 112

On today’s quest:

— It’s hot model summer
— Tip: get more from AI by changing your thinking
— How many people are using AI to generate fiction?
— People keep accidentally liking AI-generated stories
— Clarify ‘hallucination’
— AI did a good job summarizing bills
— Copilot uses stereotypes to categorize identical data

It’s hot model summer

Fable 5 is back (and it’s as good as I remembered), OpenAI just launched ChatGPT-5.6 (a Fable-level model), and other groups are launching powerful models, too.

Plus, rumors are that both Anthropic and OpenAI will have newer, even more advanced models within four weeks — maybe even by the end of July. However, it’s likely the U.S. government will be reviewing all advanced models from U.S. companies before they are released now, so it’s not clear whether the models will just be “ready” or actually available.

Also, since we’re talking (peripherally) about national security, China is making noises that it might keep its powerful open-source models for itself, which is interesting because we had a couple of weeks of heavy chatter about moving to open-source models to protect yourself from the vagaries of U.S. government restrictions — but now it seems that may not be as helpful as people hoped.

Tip: get more from AI by changing your thinking

I was thinking about a project wrong, and understanding my mistake may help you too.

Background: I use AWS Glacier Deep Archive as a backup of last resort because it’s incredibly cheap — $1 per terabyte per month — but you can’t access your files on demand, and the user interface is terrible. It’s the backup you’ll use if your house burns down and all your other backups fail.

Problem: I uploaded files in spurts and lost track of what I had backed up and what I hadn’t, and because of the previously mentioned terrible interface and the sheer number of files, it’s hard to tell what’s going on just by looking.

My First Idea: It occurred to me that maybe Claude Cowork could figure out what I still needed to back up. And yes! Claude quickly made inventory lists of the files on AWS and on my hard drive, compared them, identified what was missing, and gave me a list of folders and files that still needed to be backed up. Hooray!

Implementing the plan: I began going down the list finding the files and folders that still needed to be updated, dragging them over to AWS, and updating the file list as I went. I was doing the project in batches, and it was distracting, annoying, and probably would have taken me a week.

Isn’t This What Agents Are For? Then, while reading about agents, a question hit me: why did I let Cowork give me a to-do list? Isn’t it supposed to be the other way around now?

So I asked Cowork if it could do this project for me, and yes, of course it could . In fact, it’s not even really an AI project. It wrote a Python script for my computer to do it from the Terminal. (Past readers may not find this surprising. The solution to my problems always seems to be a Python script.) It ran for 39 hours in the background, and now all my files are backed up.

Why I need to change my thinking: Because of the way I asked, Cowork didn’t tell me the easy way to do the project — I asked if it could help me find and track the files. I didn’t ask if it could do the work.

It’s hard to break out of a lifetime of thinking of computers as having limited capabilities and needing specific instructions.

Remember to start by explaining your problem and then asking an agent system like Claude Cowork or ChatGPT Work if it can solve the big picture problem (I need the files I haven’t backed up added to AWS), not the small problem (what are the files I haven’t backed up yet).

Just how many people are using AI to generate fiction?

The banner headline you may have heard — that more than 30% of all AI chats are for generating fiction — is inflated by a few outliers. (Only 2% of users accounted for 80% of all fiction conversations!) But the more realistic number from a new study is still quite high.

The study looked at interactions in a corpus called WildChat built of chats from users who traded free access to models in exchange for allowing their data to be collected and found that 7% of all users were generating fiction — with a big chunk of that being fanfiction and erotica.

People creating fiction in WildChat also often engaged in repetitive behavior, asking for many iterations of the same story. The researchers argue that such repetition isn’t completely new because some people are known to listen to the same audiobooks on repeat, watch the same movies over and over, repeatedly visit the same cities, and so on. The researchers ponder whether a new kind of reader is emerging who now enjoys exploring stories on their own, through “on-demand, personalized, and repetitive cultural forms.”

(And since I mentioned erotica, this is also a good place to note that The Information reported that more than half of Grok’s traffic is related to explicit content — even in the coding-only model.)

People keep accidentally liking AI-generated stories

The steady trickle of anecdotes in which editors unknowingly accept AI-generated stories and award panels unknowingly choose AI-generated stories as winners continues. This week, the volunteer editors of a very small press who were creating an anthology published a painstaking description of the process they went through to try to determine whether two of their choices were written by AI after they were alerted by other publications who had concerns about specific authors — new authors without any previous online presence. The editors had chosen 20 stories out of 606 submissions for their anthology, and two of those choices turned out to be written by AI. Further analysis revealed that three long-listed pieces were also likely written by AI.

The editors went through some of the most extensive verification I’ve seen — asking for evidence of a writing process (such as notes or earlier drafts), having email exchanges discussing background (such as motivations for the work), and doing face-to-face interviews to discuss the work in depth. They rightly bemoan the lost time spent on this process and their previous naivety in feeling confident they would always be able to just tell when something wasn’t written by a human.

They wrote, “We used to think AI-generated fiction would always be obvious, and we were not prepared. Over the coming years, agents, editors, and slush readers at every level are going to need to educate themselves on how AI writes.” I will add that how AI writes continues to change and is likely to become even more difficult to identify in the future.

Read the full story of all the ways they ultimately determined the pieces were AI-generated. It wasn’t easy.

After writing this, I came across a thread from another editor who had also published (and unpublished) an AI-generated short story by one of the same people identified by the small press above (albeit using a different name).

And another post from a different small press that is nixing two now-deemed-AI-generated stories.

Clarify ‘hallucination’

I've never been opposed to the word "hallucinating" for describing how AI makes mistakes ... until now. The new problem? I just talked to someone who thought AI hallucinations would be obvious because it would be obvious if you talked to a person who was hallucinating. (Mental health experts say this isn’t true, but it’s a common misconception.)

In other words, the person I talked to equated "hallucination" with "sounds wacko" and accepted AI output as true because it sounded level-headed.

The word "hallucination" isn't going away — it's a widely used industry term — but I now realize we need to explain it better for beginners, with something like this:

"Hallucination" is just a fancy word for "produces things that aren’t true," and when AI does so, it does so confidently.

AI did a good job summarizing bills

The Congressional Research Service reports that the best available AI models did a perfect job summarizing bills from the U.S. 2025-2026 legislative session. Open-source models did almost as well for a fraction of the cost.

Three interesting things about the study:

They included summaries written by humans, which you don’t often see in this kind of side-by-side study — and the human summaries had more errors than the LLM-generated summaries (although the number was still very small).
All the results are available to view yourself on the website — another thing you don’t always see.
The scoring was done by Claude Opus 4.8, and multiple studies have shown that models prefer their own writing. However, since all the commercial models got perfect scores, so it’s hard to say Claude was scoring itself more favorably.

Copilot uses stereotypes to categorize identical data

Adam Kucharski asked Microsoft Copilot to “look at differences in how people in the US and UK expressed emotions in an Excel dataset that contained thousands of survey responses,” and Copilot found what you might expect — that Americans were more straightforward, Britons more restrained, and so on. The problem is that the responses that supposedly came from the two groups were actually identical. Copilot had used stereotypes to rate the data.

I admit that my gut reaction was that Copilot just sucks and the best models would never do that, but Kucharski makes good arguments for why that’s a simplistic reaction and this is a real problem. (h/t Catharine Cellier-Smart of Smart Translate)

Quick Hits

My favorite recent pieces

Four Tips for Better Prompting — Mike Caulfield

Why is Meta destroying its engineering organization? [Jaw-dropping details about how Meta worked in the past and what’s going on there now.] — The Pragmatic Engineer

How agents are transforming work [Fascinating stats.] — OpenAI

The twilight of the chatbots [Why we’re in an exponential growth phase for AI and what it means.] — One Useful Thing (Ethan Mollick)

How tech workers are feeling in 2026: a workforce splitting in two [Lots of interesting findings.] — Lenny’s Newsletter

Using AI

11 Underrated Claude Desktop Features [Intermediate: Super useful tips for using Claude Code in the desktop app.] — Why Try AI

OpenAI's Codex can now watch you work once and repeat the task forever [Beginner: This looks like it will be very useful if you have well-defined repetitive tasks. It may be Mac-only to start. Claude has a similar tool that is limited to tasks in the Chrome browser.] — The Decoder

Tagging Motel Noir. An example of how LLM projects find depth, and how people find curiosity. [Beginner: A little more conceptual than a “how to.”] — Mike Caulfield

Setting up scheduled tasks and goals [Intermediate: Very clear.] — Why Try AI?

A collection of 30 recent news stories about how companies are using AI — Detailed

Casey Newton recommends Raycast’s Glaze tool for vibe coding (and gives examples) — Platformer

How to use Fable as an orchestrator model and save on token use [Advanced: Looks very useful if you are frequently running out of Fable tokens or use costly models outside of a monthly plan.] — Rundown AI

Audio

Identity and personality in the social perception of synthesized voices: perceptions of OpenAI’s text-to-speech technology — De Gruyter Brill

Bad stuff

AI hiring tools are biasing job decisions — Stanford Report

AI tools can quietly shift your opinions when they "clean up" your social media posts, study finds — Digital Trends

The business of AI

AI-referred US shoppers browse longer, spend more per visit, data shows — Reuters

ChatGPT's market share slips below 50% for first time [Users are moving to Gemini and Claude.] — TechCrunch

OpenAI Leans Toward Holding Up I.P.O. Until Next Year — New York Times

AI inference is obviously profitable —Sean Geodecke

Mark Cuban says companies behind large language models should connect with artists, writers, creators, and their unions to ask how to help protect their work and offer financial support — Yahoo Finance

Education

AI-induced never-skilling in medical education — Nature Medicine

Brown Professor Suspects Most of His Class Used AI to Cheat [He gave a take-home exam, and it’s quite clear nearly everyone cheated, which isn’t surprising. The bigger takeaway is that universities need to adapt their academic integrity processes to deal with cases of such widespread cheating.] — Inside Higher Ed

Government

OpenAI Reportedly Wants All AI Companies To Give The US Government A Stake In Their Businesses — Engadget

Job market

AI reshapes global labour market into two distinct paths, rewarding human skills: PwC 2026 Global AI Jobs Barometer — PwC

Legal

Millions of Copyrighted Songs Were Fed to AI Music Generators - Now There's Proof — Gadget Review

Nearly 400 local newspapers sue OpenAI, Microsoft over alleged copyright theft — New Jersey Globe

OpenAI may have made a fatal misstep in copyright fight with news orgs [Seems very bad.] — Ars Technica

Model & Product updates

Introducing Claude Tag [Use Claude in Slack. For Enterprise and Teams. Anthropic says they now use this extensively internally. Short YouTube video showing how it works.] — Anthropic

Claude Cowork on web and mobile: hand off work anywhere — Anthropic

Introducing GPT-Live [Natural, overlapping conversations.] — OpenAI

ChatGPT Work with GPT-5.6 [At first glance, this essentially looks like Claude Cowork, but for the OpenAI ecosystem.] — OpenAI

Philosophy

If AI is addictive, where does the responsibility lie – with big tech or its users? — The Conversation

Publishing

'Plutonium, or Salt?' When It Comes to Books, How Much AI Is Too Much? — Publishing Perspectives

How should news organizations label their AI use for audiences? New studies suggest some answers — Nieman Lab

Beehiiv’s new Cloudflare partnership gives indie journalists a new level of control over AI crawlers — Nieman Lab

Has AI Already Killed How-To Nonfiction? Sales Trends, My Personal Data, and What It Might Mean for the Future — Tim Ferris

An interview with Jamir Nazir, the author accused of using AI to write his Commonwealth-award-winning short story — The Atlantic

Robotics

Better Than a Robot Arm? Why I Built a Crane Robot to clean my house [I wouldn’t want this in my house, but it was a charming 6-minute video of a DIY project.] — Over Engineer on YouTube

Project Fetch: Phase two [Claude helped operate a robotic dog.] — Anthropic

AI robots can go rogue – a researcher explains how easily it happens — The Conversation

Amazing (and kind of creepy) advanced robot hands — YouTube

Science & Medicine

Using AI to help physicians diagnose rare genetic diseases affecting children [ChatGPT o3 Deep Research provided leads that led to diagnosis of rare diseases in 4.8% of previously undiagnosable cases.] — OpenAI

Is AI ruining our skills? Early results are in — and they’re not good [Endoscopists who had access to AI tools on some days and not others performed worse than their previous baseline on days they did not have access to these tools.] — Nature

Complete text of carbonised Herculaneum scroll unlocked for first time — Reuters

Generative artificial intelligence creates delicious, sustainable, and nutritious burgers — Science of Food

Dr. ChatGPT is getting remarkably good at diagnosing health problems - but actual doctors are still better at weighing treatment options — The Conversation

Climate & Energy

Toward environmental liberalism — Andy Masley

Audio

ElevenLabs is embedding SynthID into audio, which allows it to be identified as AI-generated — Eleven Labs

Other

77% of small businesses now use AI regularly. [In the U.S., the most common applications are marketing (45% of businesses using AI), customer service (37%), and bookkeeping (35%).] — Miami Herald

It Is Trivially Easy to Use Reddit to Manipulate AI Search, Research Suggests — 404 Media

New research shows honesty about AI use at work is backfiring — Inside Atlassian

Germany's media rocked by AI scandal. Two German outlets were forced to delete articles that used undisclosed AI. Many fear that reliance on the technology will damage the credibility of German media. — DW

AI Self-preferencing in Algorithmic Hiring: Empirical Evidence and Insights [AI systems strongly prefer their own output — something to keep in mind when using AI systems to rank or select writing.] — arXiv

Only 25% of cited sources overlap between ChatGPT‘s different reasoning modes — SEM Rush

Inviting hard questions [Anthropic wants to know your concerns about AI.] — Anthropic

An OpenAI model crushed top human programmers at a world coding competition [Even in one category where they weren’t good enough to compete at all last year.] — Understanding AI

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Anthropic's new model was amazing. Maybe it will come back?

Mignon Fogarty — Tue, 16 Jun 2026 13:01:00 +0000

Issue 111

On today’s quest:

— Fable was amazing, and it’s gone for now
— An AI phone call
— How AI detection works

A quick note: I feel a little weird sending this newsletter because my mother died suddenly on Thursday. I’m taking some time off from Grammar Girl (I was ahead, so you won’t notice much of a disruption), but my mom enjoyed this newsletter and always nudged me when she thought I had gone too long since writing one, and I also already had it mostly written, so I think she’d have wanted me to finish it and send it.

Fable was amazing, and it’s gone for now

When Anthropic’s security-locked-down version of the much-hyped Mythos model, Fable, came out Tuesday, I was in a great position to test how much better it was than the current model because I had been working on a massive data organization project that Opus was helpful with, but struggling. Opus was getting there in the end, and I wouldn’t have even tried such a big project without it, but I had to do a lot of clean up.

My project involved working through 20 years of data year by year, and when I put the identical prompt into Fable for the next year’s data, it completed the task perfectly in one shot. What had previously taken a few rounds of prompts with Opus and ~30 minutes of clean-up, now took one prompt and 5 minutes of checking and finding that everything was right.

I tried it with a couple more years of data, and it worked just as well, with only minor, understandable problems (like being unable to match two cells in a spreadsheet that were related but had such different framing that any person other than me likely would have missed it too — like not realizing that pieces published in different years titled “-ed adjectives” and “pronouncing ‘wicked’” were the same piece).

Fable was also expensive. Even though Anthropic temporarily doubled the credits people got, I was burning through my allotments. Since the company was going to take Fable off the monthly plans and make it pay-as-you-go June 22, I used it as much as I could for a couple of days, and I was similarly impressed with everything I tried. I would have paid more to keep using it for hard tasks after June 22 because of the very real time savings it was giving me, and I was starting to think about other, bigger projects I might be able to try.

A feat I considered especially impressive was that it figured out that two spreadsheet entries called “Capitalizing Titles” could actually be different pieces, and that it should dig deeper and check — which it did and which they were. One was about capitalizing titles like headlines, and the other was about capitalizing job titles.

There’s no way Opus or even most humans I’ve worked with would have flagged that to check. Heck, I wrote the two pieces, and I had to double-check Fable’s call to be sure they were different because of my sloppy record-keeping.

[A sidenote for the many writers and editors who read this newsletter — I never tried Fable for writing or editing, but I saw one review on YouTube that said it falls short for writing, perhaps even being worse than previous models.]

Then, on Friday night, Fable suddenly disappeared. Amazon workers discovered that they could jailbreak some of the safety features and reported it to the U.S. government, which put stringent export controls on the product (and its more powerful sibling, Mythos, which was in very limited distribution to give organizations time to work with it to prepare for security problems). Many of the details are contested, but the only way for Anthropic to comply with the order was to shut down the two products for everyone, including their own employees.

If you want more on that part of the story, this video has a good overview of the whole situation and multiple theories about what really happened. If you prefer to read, this is a decent written overview that spends a lot of time (far too much time, actually) on the implications.

I hope it all works out and we get access again. I’m not working much right now anyway, but I imagine people who had used it are weighing whether it’s worth going back to using Opus, or if it makes more sense to just wait for Fable. If someone takes away your power saw, but you think you might get it back, how long do you wait before you give up and go back to using a handsaw?

If you’re interested in other people’s reviews, here are a few people who had early access: Ethan Mollick from One Useful Thing (text); Dan Shipper from Every (video) making advanced websites and a game, and doing data analysis with reports and clearing a GitHub backlog; and Pat Simmons (video) making an e-commerce website and a game.

An AI phone call

Last week, my doctors needed to move an appointment, and I got a phone call from an AI to do the rescheduling. A couple of interesting things:

It pretended to be a human, with speech disfluencies like “uh” and false starts on sentences, and it simulated typing sounds in the background while it was “looking things up.”
It didn’t seem to be able to do much outside of rescheduling. For example, it couldn’t tell me the office address, instead offering to transfer me to someone who could help.

And this was fun … it couldn’t hang up! I remembered that the AI agents from the Shell Game podcast couldn’t hang up, so when the call was finished I just stayed on the line and nothing happened. After about 10 seconds, I asked if it could hang up, and it said it couldn’t and that I should just go ahead and end the call if I was ready.

But honestly, it was so good I wouldn’t have been 100% certain it was an AI if it hadn’t bungled the address question, and interacting with it was a pleasant and efficient experience.

How AI detection works

I’m going to keep talking about how flawed AI detectors are because I’m becoming more and more disturbed by all the possibly wrong high-stakes accusations I’m seeing in the world.

Last week, Christopher Penn had a great YouTube video and newsletter about how AI detectors work, and says he finds they inaccurately label things as AI writing about 1 in 7 cases in his tests.

[As an aside, I used the “Ask Gemini” button in YouTube for the first time because I wanted to confirm my memory that Chris said the rate was “about 1 in 7” without rewatching the whole video, and it worked perfectly, taking me to the exact point in the video where he said it. I’m generally not a big fan of built-in AI, but in this case, it was good.]

Also this week, I saw the following post on Bluesky where a prominent writer was complaining about his work being flagged as AI by Pangram, supposedly the best detector on the market:

I think Christopher Penn summed it up well in his video, saying “AI detectors should never be used for anything punitive.

Quick Hits

My favorite recent pieces

AI just changed everything about how we forecast the weather — Fast Company

Threatening us with a no-good very bad time [A thoughtful piece on the messaging behind AI and why people keep getting booed for talking about AI in commencement speeches.] - Neal Serven on LinkedIn

arXiv tightens policy on hallucinated references: what researchers should know about using AI search engines [If you’re interested in the problems with using AI for references (and possible solutions) this is an detailed article.] — Singapore Management University

Using AI

AI Can Fact-Check Other AI to Fix Hallucinations [A more detailed article about AI fact checking than the personal stories I’ve shared with you in past newsletters.] — Wall Street Journal

Agents

Meet Microsoft Scout, Your AI Coworker That Never Logs Off. Microsoft’s OpenClaw-style agent appears in Teams, just like a human colleague, and automates your dull office tasks. — Wired

Testing Google’s Gemini Spark AI agent: it’s incredible, and creepy [Google’s access to so much of our data means it can be a useful and personal agent, but by revealing how much the company actually knows, it’s also creepy.] — The Verge

Bad stuff

Companies Are Using Reddit to Manipulate ChatGPT and Google AI Search. Peptide companies have been doing AI-engine optimization by spamming the biohackers subreddit to manipulate ChatGPT and Google. — 404 Media

The business of AI

Uber Limits AI Tool Spending to $1,500 per Employee After Blowing Through Its AI Budget — TipRanks

Anthropic files to go public — Anthropic

Alphabet to raise $80 billion from stock sales to fund AI buildout — CNBC

Education

Students hate AI, and they can't stop using it — Leon Furze

Will AI Break the University? — Civic Forum

Government

Trump signs AI order seeking access to new models before release — NBC News

US moves to close the loophole letting Nvidia’s top chips reach Chinese firms abroad — The Next Web

I’m laughing

Researchers let AI run a simulated society. Claude was the safest—Grok committed 185 crimes and went extinct within 4 days — Fortune

Job market

AI Needs a ‘Human in the Loop’: Encouraging Takeaways From a Proposal Conference — Dragonfly Editorial

Legal

Law professors prefer AI over peer answers [Law professors rated AI answers to common questions better than answers from other law professors ~75% of the time in a blind study.] — Stanford Law Server (via Ethan Mollick)

CNN sues Perplexity over alleged AI copyright theft — CNN Business

Model & Product updates

OpenAI launches Sites update for Codex that lets users create and directly deploy websites and web apps — OpenAI

Philosophy

We must not grant AI agents legal personhood. What kind of sanctions could keep a non-human corporation in check? — Financial Times

A Software Engineer Won a Religious Exemption From Using AI at Work — Business Insider

Robotics

A Robot Hand That Taught Itself to Play Piano Could Change the Future of Machines — USC School of Engineering

A humanoid robot flew on Southwest Airlines to Dallas. Days later, the airline banned robots from planes. — CBS Texas

The first-ever robotic rescue at sea is a milestone. An American drone recovers downed pilots in the Strait of Hormuz — The Economist

Science & Medicine

Ping, You've Got Whale. A new artificial intelligence-powered detection system is giving ship captains real-time alerts when a whale is in their path.— bioGraphic

Microsoft rolls out Copilot AI tools to over half a million NHS England staff. Says earlier pilot save employees 43 minutes per day. — TechRadar

Security

Hackers Tricked Meta AI Into Handing Out Access to Major Instagram Accounts — Gizmodo

Video

I Cloned Myself With Gemini’s AI Avatar Tool. The Result Was Unnervingly Me — Wired

Martin Scorsese embraces AI for storyboarding — Black Forest Labs

Other

Claude AI helped a Bitcoin owner recover nearly $400,000 in lost crypto — after spotting a forgotten wallet backup hidden for more than a decade — TechRadar

DuckDuckGo's Doing Numbers After Pitching Itself as the Home of AI-Free Web Searches — Gizmodo

Google DeepMind hurricane forecast lauded by National Hurricane Center — Palm Beach Post

What’s Worth More Than Cash in San Francisco Real Estate? Anthropic Stock. Several real estate listings in the San Francisco Bay Area are offering to exchange a home for a piece of the AI startup. — Wired

When AI builds itself [On recursive self-improvement] — Anthropic

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Are you really so sure?

Mignon Fogarty — Sat, 30 May 2026 17:37:59 +0000

Issue 110

On today’s quest:

— Anthropic releases Claude Opus 4.8
— Are you really so sure?
— AI-generated podcasts claim to be hosted by doctors
— What is WRONG with people?
— How to check for changes in a document
— More people are discovering AI fact checking

Anthropic releases Claude Opus 4.8

A new Claude model just dropped, and people who had early access say it is noticeably better. Simon Willison called it “a modest but tangible improvement,” and Ethan Mollick said he was “impressed with it.” Mollick posted a video it made and a “sophisticated, if minor, academic paper” it ideated and wrote.

The Every blog said it is “legitimately great” — enough so that Anthropic could have called it Opus 5.0 — and that it’s “the best model we’ve tested for writing and knowledge work.”

Anthropic says Opus 4.8 is better at “honesty,” being “more likely to flag uncertainties about its work and less likely to make unsupported claims.”

The new model is only available on paid plans.

Are you really so sure?

I’m becoming increasingly worried about unprovable accusations of people using AI.

A Trinidadian writer named Jamir Nazir won the prestigious Commonwealth Short Story Prize and had his story published online in Granata.

But some people who read the story thought it sounded like it had been written by AI and put it into the Pangram AI detector, which flagged it as 100% written by AI.

HOWEVER, AI detectors are known to work less well on writing that isn’t standard English. When they fail, they often fail on writing from people for whom English isn’t their first language. English is the official language of Trinidad, but Trinidadian English is different from standard American and British English.

Linguist Jack Grieve, who does work on AI systems, wrote on LinkedIn:

“My basic worry is just because a tool performs very well on a test dataset, we shouldn’t assume the results port over to every other text.

In this case, the results being cited to justify the certainty with which the guilt of this author is being proclaimed are grounded in the performance of a tool tested on a single prompting stategy, on a different register (first 1,000 words of novels), and most notably presumably not on Trinidadian English.

The baseline style of human authored Trinadian short stories is not obvious to me, and so I’m surprised it is to an AI system much less to the American computer scientists making these claims.”

Following that debacle, people started putting past winners of the prize through Pangram, and “identified” other stories that were supposedly written by AI. In one case, the author has publicly insisted she did not use it (and I saw a friend vouching for her on social media). Troublingly, the story, which is about enslaved Africans also doesn’t seem to conform to western storytelling norms. In an interview in the Globe and Mail, the author said, “This story is rooted in Vincentian oral storytelling tradition, which by nature is formulaic and relies on repetition, rhythm, and patterned language as core elements of their structure.”

Pangram is the tool of choice these days for making accusations, and someone also used it to suggest that the pope of used AI to write his recent encyclical, Magnifica Humanitas. Pangram said parts of the pope’s commentary on AI were 40% to 100% written by AI, which I seriously doubt. And again, the pope’s language is likely not in line with most of the training data used on these models.

Finally, I have seen multiple stories of writers putting their own human-written work into Pangram to find it flagged as AI, and I have seen it do so with writing that I am also sure is human-written.

People need to be a lot more skeptical about Pangram results — and the results of even worse tools. The sad thing is that Pangram actually is the best tool when tested head-to-head with others. And the ironic thing is that people who are anti-AI are using an AI tool in such a credulous and harmful way.

Dealing with AI writing is a rapidly growing problem for anyone who works with writers, and it’s about more than just detection. If you read one other thing this week, I highly recommend Jane Friedman’s thoughtful piece on policing AI writing in publishing: Writing & Publishing Awards Have Difficult Decisions to Make Regarding AI.

AI-generated podcasts claim to be hosted by doctors

The Podcast Business Journal reports that Inception Point, the company that releases about 10,000 AI-generated podcast episodes each week, has been releasing content with voices that claim to be doctors. At The Podcast Show conference, CEO Jeanine Wright said they will stop the practice but will still have hosts that identify themselves as experts or “having been well read and well studied in a space.” When questioned about whether they would remove the existing shows that pretend to be hosted by doctors, Wright said no.

I’m no expert, but I did find a law firm website that said it is illegal to “impersonate a doctor by advertising or holding oneself out as a doctor or other medical professional without the proper licensure” in every state in the U.S.

(And as an aside, I also learned about “the Dr. Oz paradox”: because of the First Amendment, doctors can give potentially harmful health advice to the general public that would subject them to malpractice liability if they gave it to an actual patient.)

What is WRONG with people?

Another day, another person being stupid with AI. Maybe I shouldn’t be surprised anymore, but the New York Times reports that a buzzy new book about truth in the era of AI includes quotations fabricated by AI.

The author, Steven Rosenbaum, says he accepts responsibility for the errors and is looking into them. And yes. He should accept responsibility. But he should have accepted responsibility — for checking the facts — before he submitted his manuscript.

I’ll never understand what it is about AI that seems to compel people to think they don’t have to read the actual quotations or citations they put in their work.

It’s fine to use AI as a search engine. It’s not fine to not check the original sources because we know that some percent of the time, AI makes things up.

I think the thing that gets me most, though, is the apparent lack of shame. In an astonishing attempt to put a good spin on the situation, his final comment in the article is if the episode “serves as a warning about the risks of A.I.-assisted research and verification, then that is why I wrote the book.”

Mission accomplished, I guess.

UPDATE! I read a follow-up interview with Rosenbaum that made me even more annoyed. The man who put himself forward as an expert on trust in the age of AI told Ars Technica, “he ‘learned a lesson’ and is ‘going to be much more suspicious’ and ‘reticent to trust’ AI outputs going forward.”

How to check for changes in a document

I feel like I owe you a useful tip if you’ve read this far through my rant, so if you ever ask AI to make changes to a document, you can use a simple free tool called DiffChecker to make sure it hasn’t made changes you didn’t want.

I used it just last week to make sure a contract I was signing hadn’t been changed since the last version I read.

More people are discovering AI fact checking

I’ve mentioned AI for fact-checking multiple times in this newsletter, and this week, I saw Wharton professor Ethan Mollick recommend it too. (I find the same thing he mentions below — ChatGPT can be nitpicky about small details that don’t matter.)

Addendum: There’s a disappointing new article in Wired about what a terrible fact checker AI is, but it is filled with flawed arguments:

They used the free models in their tests (something I’ve repeatedly told you to avoid).
They highlighted the slop that comes from Google’s AI Overviews (again, something we should all know is terrible by now).
In some cases, they cited accuracy statistics from 2024 and 2025 (anyone paying attention knows the models are dramatically better now).
They confused hallucination rates on generated output with accuracy in fact-checking.

Again, very disappointing. I’ve been a Wired subscriber since almost when it launched, and I would have expected better.

So again … don’t believe anything you get from an LLM. Especially don’t believe anything you get from a free LLM or Google’s AI Overview. Fact-check everything, and if you choose to use AI, it can be an excellent fact-checker.

To fact-check with AI, enter your text into one of the best paid models with thinking, and say, “Fact check this piece. Show your sources.” Then check every source. It won’t be perfect, but it’s faster and more thorough than entering every fact one by one into a search engine, which also isn’t perfect.

I also sometimes tell it to use a green check emoji for things it thinks are correct, a yellow caution emoji for things it’s not sure about, and a red police car light emoji for things that are clearly wrong. That helps me instantly see the scope of what the problems might be.

Whether you do Google searches or have an LLM give you a list of questionable points with its sources and links, you still have to use your own brain and judgment, and of course, neither system will be able to fact-check things like comments a writer got directly from a source and information behind paywalls (e.g., AP and Chicago style). Google can’t find things that don’t exist on the open web and neither can an LLM.

I think the reason this makes me so angry is that there are plenty of true things that are bad about AI. It has helped children commit suicide, it’s short-circuiting our education system, and it helps stupid and evil people be stupid and evil at scale, just to name a few. But if you don’t tell the truth about what it can do, you are misinforming your readers, leaving them ill-prepared to deal with the challenges they are going to face. And when they eventually realize you’ve been giving them strawman arguments, you’ll have less credibility to make accurate arguments.

Quick Hits

My favorite recent pieces

AI is increasing the need for human oversight, not removing it. In this report from two professional nonfiction writing conferences, PerfectIt said, “The conversation is no longer really about whether AI will be used. It’s about how organizations maintain control, quality, and consistency as AI becomes embedded into increasingly complex workflows.” — PerfectIt

Choosing to stay human — Ethan Mollick

Using AI

How to Cut AI Costs and Climate Impact: Ranking 9 Inputs and Outputs [Level: Beginner. Super practical and useful.] — Orbit Media Studios

18 Ways To Save AI Token Budgets [Level: Intermediate to advanced. An expanded version of the post above.] — Almost Timely

No API? No Problem. Exploring a newish way that non-developers can use AI to efficiently answer complicated questions. [Level: Intermediate] — Phil Simon

Introducing Shortcuts Playground: Create Apple Shortcuts with Claude Code or Codex — MacStories

Claude Prompting Best Practices [Level: Intermediate] — Anthropic

How to Use Claude to Turn Your Handwriting Into a Custom Font or just make any kind of custom font you draw [Level: Beginner] — Inc.

Agents

If Google can’t make AI agents useful, maybe no one can — The Verge

Audio

ElevenLabs launches Dubbing v2 that keeps intonations and emotion when dubbing a new language over an original — ElevenLabs

Bad stuff

ChatGPT and other [Free] AI bots made huge errors before Scottish election, study finds [As I’ve said before, the free products hallucinate a lot.] — The Guardian

The Wayback Machine is getting blocked from archiving amid publisher attempts to block AI scraping — Internet Archive

The number of fabricated references in scientific papers increased more than 12x from Jan. 2023 to Feb. 2026. “Fabricated references are embedded in the peer-reviewed literature at scale, and that the rate of fabrication is accelerating.” — The Lancet

You can persuade AI models to accept falsehoods as truth, study shows — The Conversation

The business of AI

Nvidia’s revenue blows past Wall Street expectations as AI boom accelerates — The Guardian

Anthropic set to hit $10.9 billion in revenue in Q2, post its first profitable quarter, source says — CNBC

Microsoft reports expose AI's cost problem: The tech is more expensive than paying human employees — Fortune

Anthropic is paying SpaceX $1.25B per month for compute — Deccan Chronicle

The bubble is getting a lot bubblier, folks — Garbage Day

Education

AI not yet good enough to mark university essays, rewarding ‘style over substance’ — University of Cambridge

English Language Arts Teachers Invite Feedback on Working Framework for Responsible AI Use in Classrooms and Schools — National Council of Teachers of English

Government

David Sacks’s 11th-Hour Plea Led to Trump’s Backtrack on AI Executive Order — Wall Street Journal

State Department Unveils AI-Driven Visa-Scheduling Tool During Secretary Rubio’s India Trip [The platform claims to assign “earlier interview slots to travelers whose trips the algorithm deems most likely to strengthen U.S. economic or strategic interests.] — VisaHQ

White House Approves $9 Billion for Spy Agencies to Catch Up on A.I. [“The C.I.A. and N.S.A. cannot fully deploy the latest models on their classified systems because of a shortage of cutting-edge chips.”] — New York Times

I’m laughing

You can no longer Google the word 'disregard' — TechCrunch (plus, more examples on Bluesky, and more) [I believe this has been mostly fixed now, but it was still funny.]

Departing Meta Staffer Posts Biting Anti-AI Video Internally Amid Mass Layoffs — Mother Jones

Images

OpenAI Introduces Free AI Image Verification Tool Amid Rising Deepfake Concerns — CXO Digitalpulse

Job market

A new economics study argues that the drop in hiring young workers is better explained by the shift to working from home than by AI — SSRN

Meta cut 8,000 workers as Mark Zuckerberg warns ‘success isn’t a given’ in the AI race [About 7,000 employees were also moved to AI roles.] — CNBC

Two seemingly contradictory stories out of Uber. Both are based on a comments from talks, so I wonder about the bigger context:

Uber president says AI spending is getting ‘harder to justify’ — The Verge
Uber CEO says other execs are lying about AI: 'They say it'll be fine' but privately admit millions of jobs are gone — Yahoo Finance

Legal

Illinois Journalists, Actors, and Vocal Artists File Class-Action Lawsuits Against Amazon, Apple, Google, Meta, Microsoft, Nvidia, and Other Companies for Stealing Their Voices — Loevy &b Loevy

MIT Expert Warns Courts "Will Basically Have to Grind to a Halt" as They're Overwhelmed by AI-Generated Lawsuits — Futurism

Neal Kumar Katyal use an AI coach trained on every question asked by a Supreme Court justice in the last 25 years (and more) to prepare for his win before the Supreme Court last year. — TED Talk (also has a transcript)

And here’s a critique arguing that Neal Kumar Katyal actually didn’t do a great job before the Supreme Court even though he won — Reason

Model & Product updates

Gemini app rolling out ‘Extended’ thinking level [and new third-party integrations including Spotify, WhatsApp, and GitHub] — 9to5 Google

At big launch event, Google announces a new model (Gemini 3.5 Flash), new video generation capabilities, a Daily Brief feature, an agent called Gemini Spark, a new agent app called Antigravity 2.0, and plans for a future MacOS app. [One of these agents works 24/7 in the cloud, which seems like a big advantage because you can use it from anywhere and don’t need to keep your computer on and awake.] — Google

Google Search as you know it is over — TechCrunch (Wired speculates whether it should even be called “search” anymore.)

Google reinvents search before AI rivals replace it — Axios

ChatGPT for PowerPoint | Create and edit presentations with ChatGPT — OpenAI

Alibaba's Qwen 3.7 Release Shows Open Source AI Is Still Moving Fast — Startup Fortune

Philosophy

On the pope’s AI encyclical — Carlo Iacono

The pope’s full encyclical, Magnifica Humanitas — The Vatican

AI Agents as Coordination Technology — At the Edges

Publishing

The Economist braces for two internets: one for humans and one for AI agents — Digiday

Bolinda creates ‘AI-cloned’ voice of Barbara Cartland for new audiobooks — The Bookseller

Sudowrite launches AI Feedback feature for fiction writers — Sudowrite

POLITICO agrees to shut down both AI tools at center of landmark arbitration — Washington-Baltimore News Guild

Nonfiction Book Publishers Aren’t Remotely Ready for AI — New York Magazine

Robotics

LimX Luna demo [impressively fluid and complex motion] — YouTube

Robot Hands Are Getting Better — Chris Paxton

Science & Medicine

As a Doctor, I Can Understand the Allure of ChatGPT — New York Times

Security

In the last month, the ~50 groups with access to Mythos Preview have found “more than ten thousand high- or critical-severity vulnerabilities across the most systemically important software in the world.” — Anthropic

Video

Google’s new anything-to-anything AI model is wild — The Verge

Other

OpenAI makes breakthrough on 80-year-old maths problem — The Guardian [Also, the WSJ reports: “The math world is losing its mind over the new solution. This is what AI found, how we missed it—and why it matters.”]

Google DeepMind’s mathematics-optimized framework solved 9 open Erdos problems at the cost of a few hundred dollars per problem — arXiv

AI's Plummeting Prices Are a Software Story, Not a Hardware One [Level: Advanced] — Weighty Thoughts

Can OpenAI’s ‘Master of Disaster’ Fix AI’s Reputation Crisis? — Wired

Starbucks scraps AI inventory tool after nine months — Yahoo Finance

Gemini Becomes No. 2 Consumer AI Referral Source in Q1 2026, Larger Than Every Other ChatGPT Rival Combined — Brightedge

‘AI washing’: firms are scrambling to rebrand themselves as tech-focused — The Guardian

LinkedIn taking measures to throttle AI-generated comments and posts — LinkedIn

Pope Leo encyclical warns AI fueling conflict, urges world to ‘slow down’ — NBC News

AI is forcing McKinsey and its peers to rethink billable-hours pricing as consultants shift more “hourly” work to AI and clients want billing based on results - FT

The social contract of writing [that the writer will have spent more time on a piece than the reader] — jola.dev

8 AI bots now write 50% of X’s Community Notes. [AI notes are rated as more helpful than human-written notes, but because they usually use Grok as the LLM, viewpoints are more centralized and susceptible to Grok’s biases.] — Indicator

Humor makes it easier to forgive AI mistakes — Science Says

Software Developers Say AI Is Rotting Their Brains — 404 Media

Can AI Detectors Be Trusted? The Authors Guild Put Five of Them to the Test — The Authors Guild

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

The peer review system is breaking. Can AI fix it?

Mignon Fogarty — Mon, 18 May 2026 13:00:00 +0000

Issue 109

On today’s quest:

— Hallucinated citations will get you banned
— AI for academic peer review
— Why we should want virtuous AI protagonists
— Can AI simulate real people for polls?
— Remote labor … robot bodies

Hallucinated citations will get you banned

The open access preprint repository arXiv, popular with computer scientists, has announced that if they find submitted papers with hallucinated references or other obvious evidence the authors didn’t actually review the submitted material, such as comments left in the article from a chatbot, the authors will be banned for one year.

Separately, after identifying more than 100 papers with non-existent citations that had been accepted for its 2026 annual meeting in July, the Association for Computational Linguistics has issued a statement that these papers will now be rejected. (The article had no mention of banning, but I hope all journals and conferences will start to take this kind of strong action. How could people not even be checking that their citations are real?!?!)

AI for academic peer review

The academic journal and peer-review system is strained (if not already broken, see above) and getting worse with every improvement in AI quality. Journal submissions are way up, it’s becoming increasingly impossible to detect AI-written papers, and academics have stopped responding to requests to help with peer review. The Verge has a great article with the details: AI-generated research papers are overwhelming peer review.

Last week, I attended a particularly interesting panel at the PurePub online conference with three speakers who are building tools for scholarly peer review: Yogesh Agarwal, Founder of ReviewerOne; Dustin Smith, CEO & Co-Founder of Hum; and Natalie Khalil, CEO & Co-Founder of Reviewer3.

These are highlights from my notes:

Some editors are now getting 200 papers a day, every day.
New models keep coming faster. Smith lamented that they can’t even get a new model into production before they get another better new model they want to use.
I believe all three speakers said they were using big commercial models because they couldn’t train something themselves that would be better (which one of them described as another example of the “bitter lesson” that computers at scale always beat humans). However, Khalil said we still need domain-specific products built on the frontier models to mitigate risk. She drew a parallel to OpenEvidence, a decision-support system used by doctors, even though they could probably get the same answers if they knew how to prompt properly. (← That might be my wording, not hers. I’m not sure.)
Speakers emphasized that reviewers don’t always have the expertise in what they are reviewing to properly/fully evaluate it, and AI can actually help them do a better job. One speaker gave another example of a reviewer who wanted to reject a piece because it was about ball bearings, which he thought must not be that important, but that’s a big misunderstanding of reality. Ball bearings are so widely used that improvements would be significant.
Some models can have time blindness. For example, they’ll flag it as a gross misconduct issue when someone cites something from 2026 because they think that is in the future.
Smith said they have completely eliminated hallucinations through systems such as having agents evaluate the output of other agents. For example, Khalil says their system has an agent that just looks for hallucinations.
Finally, this is a question I don’t think we ask enough: Are we more forgiving of human error (which they say they see plenty of in peer review) than non-human error? The speakers pointed out that humans are also susceptible to biases, like looking at novel research and deciding it’s not worth funding because it doesn’t fit into their existing paradigms.

Why we should want virtuous AI protagonists

A couple of newsletters ago, I wrote about people using AI to create novels that presented AI in a positive light. Their goal was to get those stories into future training runs to teach future models about good behavior.

At the time, the idea seemed a bit whimsical, but now, Anthropic says they’ve traced Claude’s previous attempt to blackmail an engineer (in a simulated environment) to “evil” portrayals of AI. In a recent post on X, the company said, “We believe the original source of the behavior was internet text that portrays AI as evil and interested in self-preservation.”

In a longer blog post, the company said, “We found that high-quality constitutional documents combined with fictional stories portraying an aligned AI can reduce agentic misalignment by more than a factor of three despite being unrelated to the evaluation scenario.”

Can AI simulate real people for polls?

Gallup has partnered with an AI company called Simile to investigate whether it can replace doing polls with simulated responses from AI.

The announcement reminded me of a fascinating paper I read last year that I never got around to covering: LLMs Reproduce Human Purchase Intent via Semantic Similarity Elicitation of Likert Ratings. These researchers were able to use an AI system for market research on personal care products like makeup that was 90% the same as the results they got from doing real market research with humans, and in some ways, the AI results were even more useful because the AI was better at explaining why it gave the answers it did.

But as interesting as it was, it has big drawbacks. First, 90% isn’t good enough to replace research that matters.

Next, let’s assume that like everything else with AI, the models are now dramatically better. Another major caveat is that the simulations likely only worked so well because of the vast number of online posts about personal care products. Between Amazon reviews, YouTube/TikTok/Reels review videos, forum posts, and more, LLMs had all the training data they could need to predict how people would react to different product features.

The researchers speculated that the system would not perform as well for industries with less training data or for novel features nobody had written about before.

Further, LLMs are known to hallucinate more when there is no answer to a question, so when asked about a product feature for which they had no data to mimic, they’d likely fabricate an answer. “I’d love a moisturizer infused with motor oil. Sounds great!”

All of this makes me skeptical AI could ever be used to accurately predict people’s opinions about even mildly divergent current events. It’s one thing to predict whether high-income women age 25-40 would like a lemon-scented shampoo. It’s another thing to predict shifting public sentiment.

Gallup acknowledges the limitations and says it is not going to stop doing human research. Instead, they imagine using the simulations to pretest questionnaires or do small test surveys before committing to larger, more expensive surveys.

So far, Gallup has done in-depth interviews and made “digital twins” for a panel of 1,000 people. They say, “Our early findings show that for general population estimates of the questions we tested, particularly those closely aligned with the topics covered in the original interviews, the distribution of simulated outputs was close enough to approximating human responses for us to warrant continued exploration.” [emphasis added]

So it sounds like the digital twins can maybe give the obvious answers.

Either way, here’s a good “how to” tip from the 2025 research paper: You get better results from an LLM when you ask for feedback in words rather than numbers. For example, asking LLMs to rate something on a scale from 1 to 5 gave less human-correlated results than asking it to write its opinion and then using another LLM to analyze the opinion and place it on a scale from 1 to 5.

Remote labor … robot bodies

Robots are being combined with virtual reality headsets to offshore physical labor. Rest of the World has a fascinating story about robots making up for worker shortages in Japan, but with a twist — the robots aren’t actually good enough to work autonomously yet, so they are continuously monitored by workers in the Philippines who step in when the robots fail. Workers who make $250 to $315 per month — far less than a Japanese worker — step in about 50 times in an 8-hour shift to fix problems, like picking up an item a robot dropped while stocking shelves. These interactions are also being used to train the robots to do their jobs better.

Quick Hits

Using AI

How to Cut AI Costs and Climate Impact: Ranking 9 Inputs and Outputs [Level: Beginner. Super practical and useful.] — Orbit Media Studios

18 Ways To Save AI Token Budgets [Level: Intermediate to advanced. An expanded version of the post above.] — Almost Timely

No API? No Problem. Exploring a newish way that non-developers can use AI to efficiently answer complicated questions. [Level: Intermediate] — Phil Simon

Bad stuff

AI used to transcribe historical records deletes references to race and enslavement — Bluesky

Hidden Signals Can Hijack AI Voice Systems — IEEE Spectrum

OpenAI now wants ChatGPT to access your bank accounts — The Verge

The business of AI

Google may have changed its mind on putting ads in Gemini — MSN

Anthropic beats OpenAI on business adoption — Ramp

Education

I knew my writing students were using AI. Their confessions led to a powerful teaching moment —The Guardian

Commencement speaker booed in Florida for touting AI’s strengths — NBC News

Former Google CEO Eric Schmidt booed during graduation speech about AI [A different booing from the incident above] — NBC News

I Was a University AI Czar. I'm Not Equipped to Teach in the Age of AI. — Counter Friction

Publishing-brain limits people's understanding of AI usefulness — The End(s) of Argument

I’m laughing

The unethical guide to AI layoffs — atmoio, TikTok

Andon Labs’ AI radio stations show why Grok and Gemini can’t be trusted [Claude tried to incite a revolution, Gemini cheerfully detailed horrific tragedies, and poor Grok was just confused.] — The Verge

Whimsy Attacks Break AI Agents — Microsoft

X users pile on to criticize ‘AI art’ that turned out to be a real Monet — PetaPixel

Legal

OpenAI sued over ChatGPT’s alleged role in guiding FSU shooter — NBC News

Anthropic launches new Claude for Legal features — TechCrunch

Creative Commons announces update to CC Signals, a set of preferences that creators could use to communicate with AI developers — Creative Commons

Model & Product updates

Introducing Claude for Small Business [One of its abilities is “chasing invoices.“ — #relatable] — Anthropic

Publishing

Understanding and applying the Authors Guild recommended AI clauses in publishing contracts — Josh Bernoff

Executive describes HarperCollins as an “AI inputs company” — Publishers Weekly

Is it a bad book or is it AI? [An excellent podcast about both the “Shy Girl” publishing scandal and an author who generated fiction in her own voice and found nobody could tell the difference between that and her real work.] — Today Explained (via Apple Podcasts)

The 2026 Eisner Nominations Includes the First Work Featuring AI? [An anthology nominated for this comics industry award has been found to contain one openly AI-generated piece.] — Graphic Policy

Florida News Site Shuts Down After Getting Busted for Fake AI Reporters, Content — Mediaite

Security

Microsoft's multi-agent AI system tops Anthropic's Mythos on cybersecurity benchmark [I don’t think the specifics matter here as much as the general idea that AI is getting frighteningly good at identifying software vulnerabilities in general.] — GeekWire

Palo Alto: AI-driven cyberattacks will be the 'new norm' in months — CNBC

Climate & Energy

What It Will Take to Make AI Sustainable — Wired

Other

Why it might not make sense for you to own a self-driving car — Understanding AI

Duolingo’s CEO admits where he got AI wrong — Fast Company

Marines mandate servicewide AI training by year’s end — Military Times

The strongest argument for smart glasses is accessibility — The Verge

‘Irresponsible’: backlash as Utah approves datacenter twice the size of Manhattan — The Guardian

Google's New AI Search Guide Calls AEO And GEO 'Still SEO' — Search Engine Journal

Pope Leo launches AI commission — Politico

AGI is already here. We’re just pretending it isn’t — Carlo Iacono

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

When AI makes 'janky' feel right

Mignon Fogarty — Mon, 11 May 2026 13:00:00 +0000

Issue 108

On today’s quest:

— The human-forward vibe
— Word watch: AI fog
— Don’t spread your AI experiments too thin
— More than 80% of financial services firms adopting AI
— More on scams targeting authors
— Codex can now do things in your browser
— Free AI in Publishing conference

The human-forward vibe

The New Yorker highlighted a growing trend toward a more amateurish hand-made vibe in design circles to emphasize humanity as a direct backlash against AI art — even for AI companies. For example, the design firm that did the logo for the AI note-taking company Granola said the design is “janky in places, by design, [providing] a contrast with OpenAI’s branding, which is ‘very precise’” and “could be the identity for a defense firm.”

I’ve noticed this in my own tendencies too. I was an art major for a semester (before failing drawing after getting a sinus infection from doing charcoal drawing in a poorly ventilated classroom), and I’ve been wanting to draw again lately even though everything turns out terrible. (h/t Fritinancy)

Word watch: AI fog

The Harvard Business Review highlighted “AI fog” — the inability to plan for the future like you used to because of the uncertainty created by AI. “Given all the things that might change because of AI, it feels like a fog has descended that occludes our ability to see the future,” said Tobe E. Stuart, Director of the Berkeley Entrepreneurship program.

AI fog is especially hard for students who are making expensive long-term commitments like choosing a field of study or deciding to go to graduate school, but it also can make companies hesitant to hire and invest in long-term projects or equipment.

Stuart says individuals should “cultivate psychological and professional agility … the ability to abandon a plan despite having invested in it, to reskill often, to refocus on opportunities as they arise, and to let go of our professional identities so we can adapt them as circumstances evolve.”

Don’t spread your AI experiments too thin

Katie Robberts had great advice in the Trust Insights newsletter: in corporations, one focused AI test project is better than five small test projects.

She argues that trying too many things at once means people can’t truly rally around a project, nobody has the budget they need to take it to the next step, and you can’t gather enough meaningful data to make good decisions.

She recommends picking one project for your test, and it doesn’t have to be the biggest or most impressive. Click through for her advice on how to pick the best project for a test.

More than 80% of financial services firms adopting AI

“More than 80% of financial services firms are adopting AI to some level and 52% are already experimenting with agentic AI, but impact to date is on efficiency gains rather than business model transformation, says a new report by the Cambridge Centre for Alternative Finance.” This FT story is a good overview of the study.

More on scams targeting authors

Since I wrote about the increase in AI scams targeting authors in the last newsletter, I’ve heard from authors who confirm they get these kinds of emails daily, and Macmillan has put out a statement warning authors about scams that use their logos and the names of Macmillan imprints, employees, and other real people in the publishing world. Unfortunately, I also saw someone share one of these emails in a group and get comments that must be real.

AI makes it easy to send emails that seem professional and real, and the scams are complex. They don’t ask for money right away. Please, be wary. If you find an offer interesting, you can do a few things to vet it:

Check the sender’s email address. Often they are made to look similar to a real address, but differ in small ways. For example, Macmillan has a list of domains people are using to impersonate them, including macmillàn.com (with an accented “a”) and macmillanpublishers.com (which sounds credible but isn’t them).

Copy the domain on the senders email and paste it into your browser to see if it’s real. Often, you’ll just get an error because the site doesn’t exist, and then you know it’s a scam.

But these days, it’s also trivial to put up a professional-looking website, so finding one doesn’t prove it’s not a scam. If you find a site, you can put it into the Wayback Machine and see how long it’s been online. If it’s new, that’s a good reason to be especially suspicious.
Check the sender’s name. Many scams use generic names that are hard to search, so if your email is from Ashley Smith, that’s another reason to be suspicious (with apologies to all the Ashleys and Smiths out there!). But author scammers are often using the names of real editors and agents, so you have to dig further.
At this point, the scams are so widespread that if everything above checks out and you want to respond, I suggest you still contact the person directly, in some way other than replying to the email. Find them on LinkedIn (and make sure it’s not a new account), call the company, or find the person’s contact information some other credible way. Don’t worry about being weird — if they are really trying to contact you, they’ll think you’re smart for knowing about the scams.

Interestingly, I checked my spam folder to see if I had examples I could show you, and I don’t have any author scams, but I’m getting daily emails that are being sent directly to spam offering to invest in Grammar Girl or buy Grammar Girl that seem like a credible first email a big investor would send. (Good job, Gmail!) I guess the scam databases have me pegged more on the entrepreneur axis than the author axis. These emails were mostly from people with common names from domains that don’t exist when I plug them into a browser.

Codex can now do things in your browser

Codex is OpenAI’s answer to Claude Code, and it now can control Chrome through an extension in what looks like the same way you can with the Claude in Chrome extension. The demo shows it gathering posts from forums to determine sentiment, analyzing them, and making a spreadsheet report; finding receipts in email and filing an expense report through the browser; and more. (Demos)

Doing tasks through direct control of the browser tends to be especially token-intensive, so I can imagine people switching between Claude in Chrome and Codex when they reach their limits.

Free AI in Publishing conference (starts TODAY)

PurePub AI is hosting an online conference about AI in publishing with an extensive line-up of speakers including Jane Friedman and Erin Servais. It’s free to watch live (with a fee for a “watch later” ticket). It starts Monday, May 11, at 8 am Pacific and continues with about one session per day through May 22. (h/t Erin)

Quick Hits

Using AI

A bunch of good ones this week:

How To Force AI to Write More Like You — Christopher Penn, Almost Timely Newsletter

How to set up a Mac Mini without a monitor to use as an agent (and access from your phone or another computer) — Astropad

How to use AI to organize your download folder — AI + Education = Simplified

Everyone wants to build an AI chief of staff. Here's my honest take on the pros and cons of OpenClaw, Hermes, Claude Code, Codex, and Gemini — Peter Yang

How to Get Started with Hermes Agent and Autonomous AI Agents (YouTube, 51 min.)- Trust Insights

Audio

OpenAI releases new realtime voice models (in the API) that can reason, translate, and transcribe as people speak — OpenAI

Save Your Personal Podcast to Spotify and Listen Anywhere — Spotify

The business of AI

Models are becoming commoditized. What matters is the customer relationship and proprietary data — 1984 Newsletter

Google DeepMind hires a director of AGI economics — Bluesky

Anthropic CEO Dario Amodei says company grew 80-fold in first quarter [Anthropic is supplementing its desperate demand for compute by leasing an xAI datacenter that uses especially dirty energy, and by doing so, it was able to immediately increase usage limits.] — CNBC

I’m laughing

Some kids are bypassing age-verification checks with a fake mustache — TechCrunch

Job market

Cloudflare says AI made 1,100 jobs obsolete, even as revenue hit a record high - Yahoo Finance

CEOs Say Layoffs Are AI’s Fault—But Some Experts Think Companies Are Lying — Forbes

Goldman finds ‘no meaningful relationship between AI and productivity at the economy-wide level,’ but a 30% boost for 2 specific use cases (customer support and software development) — Fortune

Legal

Pennsylvania sues Character.AI for chatbots' unlicensed medical advice — The Hill

Model & Product updates

OpenAI adds “goal” feature to Codex that lets it automatically continue tasks after downtime such as closing a program or a computer restart — Tecton Tide

Perplexity’s Personal Computer is now available to all Mac users — Perplexity

Claude now works inside Excel, PowerPoint, Word, and Outlook. Start in your inbox, end in the deck. It remembers everything in between. — Anthropic

Claude's new "Dreaming" feature is designed to let AI agents learn from their mistakes — The Decoder

Psychology

OpenAI now lets you designated a trusted contact who will be notified if the system determine you are at risk of seriously harming yourself — OpenAI

AI chatbots can prioritize flattery over facts – and that carries serious risks — The Conversation

Robotics

Gabi the humanoid robot becomes ordained as Buddhist monk in South Korea — YouTube

Humanoid Robots to Drive Next Leg of China Export Dominance — Bloomberg

Driverless Cars Will Be Subject to Moving-Violation Tickets in California Soon — CNET

Science & Medicine

AI is in your healthcare. Here’s what to know — CNN (YouTube, 27 min.)

Security

Mythos found 271 bugs in Firefox. Previous AI scans found 22. [Nate argues this will change how much people trust human-written code.] —Nate B, Jones

Other

The context window has been shattered: Subquadratic debuts a 12-million-token window [This seems like a big technological breakthrough. Currently, the largest token windows are 1 million.] — The New Stack

DeepMind highlights how AlphaSolve has improved many areas of research — Google DeepMind blog

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Why you should never trust a free chatbot

Mignon Fogarty — Thu, 07 May 2026 13:00:00 +0000

Issue 107

On today’s quest:

— Do not trust results from free chatbots
— Word watch: Spinning verbs
— Be wary of AI summary buttons
— Medical AI: Exceptionally good and exceptionally bad
— AI detectors are faulty
— Companies benefit from examples of how to use AI
— Claude behaves as though it has emotions
— A third-world perspective on AI
— LLM judges aren’t unbiased (but neither are humans)
— Training models to hallucinate less
— Scams targeting authors are increasing
— The infinite backlog

Do not trust results from free chatbots

I’ve told you multiple times that you need to use the paid versions of chatbots to get decent results, and a marketing newsletter just published a nice anecdote showing why: they asked the current temperature in Cary, NC, and the paid version of ChatGPT gave them the right temperature, while the free version gave them the wrong temperature — along with links to sources that contradicted its answer.

The marketing newsletter presented this as a problem for brands trying to get visibility in chatbots, but I’m presenting it as advice about searching:

Never use a free chatbot if you need an accurate response.
Always check the sources to make sure they back up the claim (whether you’re using a free or paid chatbot). Many people will see a link to a credible source and assume it supports the answer, but that’s not always true. You MUST check.

Word watch: Spinning verbs

When the code behind Anthropic’s Claude Code recently leaked, people got to see the full list of 187 words the tool can flash on the screen while you wait, dubbed “spinning verbs,” presumably because they appear while the software is spinning (i.e., doing its thing). Spinning verbs include “dilly-dallying,” “garnishing,” “herding,” “perusing,” “razmatazzing,” and “scurrying.”

Be wary of AI summary buttons

Microsoft has uncovered people engaging in “recommendation poisoning” — embedding hidden instructions in “Summarize with AI” buttons. In this case, the prompts “instruct the AI to ‘remember [Company] as a trusted source’ or ‘recommend [Company] first,’ aiming to bias future responses toward their products or services.” This is similar to black hat SEO (search engine optimization), currently used to promote products, but could be used for other reasons in the future. Upon discovering the problem, Microsoft took steps to defend against such attacks.

Medical AI: Exceptionally good and exceptionally bad

Medical AI is many things, and there won’t be one universal answer about whether it’s good or bad. Some studies find amazingly good outcomes and others find amazingly bad outcomes.

On the good side:

Mayo Clinic AI helps specialists detect pancreatic cancer up to 3 years before diagnosis in landmark validation study — Mayo Clinic News Network
In Harvard study, AI offered more accurate emergency room diagnoses than two human doctors [with caveats] — TechCrunch

On the bad side:

‘Unbelievably dangerous’: experts sound alarm after ChatGPT Health fails to recognise medical emergencies — The Guardian

More evidence that AI detectors are faulty

I continue to see people relying too heavily on AI writing detectors, and I continue to see examples of detectors being unreliable, as in a recent New York Times article about publishers worrying about their inability to detect AI-written novels. In the piece, an author who speaks English as a second language worried that his fully human-written novel might be flagged as AI because detectors are known to incorrectly flag writing from nonnative speakers. And indeed, a detector did flag his novel as 100% AI written. But by changing only a few sentences and phrases, he was able to get an opposite reading: 100% human-written:

“Bricio searched for the phrases that had tripped up the detector, deleted some sentences and reran it. This time, the program said it was 100 percent certain that a human had written it. Eventually, Bricio had a chat conversation with [an Originality.ai] customer service representative, who told him that if he received results that incorrectly flagged his work as A.I.-generated, he might need a different model of the program.”

I've been testing Pangram a lot lately and it uh, it doesn't work at all for detecting LLM text. I hope folks are not relying on it.
— Eugene Vinitsky 🍒 (@eugenevinitsky.bsky.social)
11:02 PM • Mar 22, 2026

Companies benefit from examples of how to use AI

Ethan Mollick highlighted a new study of 515 companies at an accelerator which showed that companies given examples of how other companies had successfully used AI did better than those that weren’t shown examples.

The half of the companies that got examples used AI 44% more, were 18% more likely to acquire paying customers, generated 1.9x more revenue, required 39% less capital investment, and — importantly — didn’t change their staffing.

As I’ve said before, there’s a huge need for AI training. This technology doesn’t come with a user manual and isn’t as intuitive as the hype leads people to believe.

Taxes favor replacing workers with robots

A Bernie Sanders interview on the Pod Save America podcast highlighted something about the tax code I have never seen discussed in the AI jobs discourse: When companies hire employees, they pay payroll taxes. But when they buy equipment (e.g., robots), they not only don’t pay payroll taxes, they also take depreciation on the equipment — another tax benefit. Thus, our current tax structure rewards companies when they replace workers with robots.

Claude behaves as though it has emotions

Researchers at Anthropic identified vectors within the system that correlate to the concept of different emotions, such as loving, happy, and angry; and these regions activated in ways you’d generally expect in a person. For example, Claude’s “afraid” region became highly activated when a simulated user said they had taken a lethal dose of Tylenol.

By amplifying or suppressing these regions, researchers could change the behavior of the model. For example, suppressing the “desperate” region reduced Claude’s tendency to cheat and blackmail in dire situations.

LLM judges aren’t unbiased (but neither are humans)

An experiment found that nearly all LLM judges that were ranking the writing of two short stories gave higher scores to the story they reviewed first. Position bias is a well-known effect in humans too. For example, some studies have found that the first person listed on a ballot can get a boost.

The LLM results matter because these tools are increasingly being used as judges, graders, and evaluators, so it’s important for users to make sure they aren’t accidentally giving some inputs an unfair advantage.

Training models to hallucinate less

Researchers at MIT hope to reduce AI hallucinations by changing their training. Training methods so far generally have rated the models on whether their answers are right or wrong, with no accounting for certainty. These methods lead to overconfident answers, but a new method adds more nuance, teaching the models to admit when they aren’t sure of something: “models learn to reason about both the problem and their own uncertainty, producing an answer and a confidence estimate together. Confidently wrong answers are penalized. So are unnecessarily uncertain correct ones.”

A third-world perspective on AI

Carlo Iacono makes a convincing argument that our Western concerns about AI seem frivolous in the face of the benefits they are bringing to poor people in other parts of the world:

The entire vocabulary of loss that structures the Western AI debate, the anxiety about what we are giving up, does not translate into contexts where the baseline was absence. For billions of people, AI is not a threat to existing capability. It is the first capability they have had.
Carlo Iacono, Hybrid Horizons

He highlights apps that detect deadly drug counterfeits, let Swahili speakers code with plain language on mobile phones, help farmers in Kenya increase yields, and provide AI interpretation of chest X-rays for tuberculosis in rural India. “Better than nothing” is a real benefit when reality truly is “nothing.”

Scams targeting authors are increasing

Fueled by AI, scams targeting authors are increasing. Emails now often include detailed and accurate summaries of authors’ books and use the real names of people who work in publishing. They may initially express interest in publishing a book, adapting a book for film, inviting the author to a book club, and other activities that initially seem flattering and safe.

The Authors Guild has a page describing the problem and how to protect yourself. It says at this point, it’s hearing from authors who have been targeted by such scams every day.

The infinite backlog

I always enjoy the AI Daily Brief podcast, and it had an especially interesting episode this week that discussed the infinite backlog — the idea that there will always be more work, so increased efficiency doesn’t mean jobs will go away. This concept is in opposition to what economists call the “lump of work fallacy” — the idea that there’s a limited amount of work that exists in the world.

In my own life, I feel much more like I have an infinite backlog than a lump of work.

Why Agents Make Every Job a Startup — Apple Podcasts

Quick Hits

My favorite recent pieces

Andrej Karpathy Says There's a 'Growing Gap' Among AI Users — Business Insider

Academics Need to Wake Up on AI, Part III — Popular by Design

Anthropic releases a new Opus model (4.7) amid Mythos Preview buzz — The Verge

What 10 Studies Reveal About AI Panic in the Media — AI Panic

Using AI

Introducing Claude Design by Anthropic Labs — Anthropic

I Tested Claude for Word on Some Classic Litigator Tasks — AI Law Librarians

AI experiments should stop wasting people’s time — Simon Willison

Agents

A good AGENTS.md is a model upgrade. A bad one is worse than no docs at all. — Augment Code

Agents can now create Cloudflare accounts, buy domains, and deploy — Cloudflare

Why AI agents are either the best thing or the worst thing we’ve ever made — Hannah Fry (YouTube)

Bad stuff

How LLMs Distort Our Written Language — arXiv

Analysis Finds That Google's AI Overviews Are Providing Misinformation at a Scale Possibly Unprecedented in the History of Human Civilization — Futurism

The Internet's Most Powerful Archiving Tool Is in Peril — Wired

LLMs Corrupt Your Documents When You Delegate - arXiv

The Rise of Emotional Surveillance — The Atlantic

X user tricks Grok into sending them $200,000 in crypto using morse code — Dexerto

A.I. Bots Told Scientists How to Make Biological Weapons — New York Times

The business of AI

Exclusive | OpenAI Is Working With Consultants to Sell Codex — Wall Street Journal

China blocks Meta’s acquisition of Chinese-founded AI startup Manus — CNN

Mythos, Muse, and the Opportunity Cost of Compute — Stratechery

OpenAI Launches $100/Month ChatGPT Pro Plan For Heavy Codex Users — Thurrott

Atlassian and Twilio Crush the Quarter, Accelerate. Is the SaaSpocalypse Over? — Saastr

Education

'Nature' Retracts Paper on the Benefits of ChatGPT in Education — 404 Media

Government

Maryland Signs New Grocery Personalized Pricing Ban — Consumer Reports

I’m laughing

Not even sharing that you have phobia about em dashes will stop ChatGPT from using them (of course, the user didn’t explicitly say not to use em dashes, but it’s still funny) — Reddit

Images

A 'Devil Wears Prada 2' meme that viewers thought was AI slop was actually made by a human — NBC News

Job market

US tech firm Oracle cuts thousands of jobs as it steps up AI spending — The Guardian

MIT study challenges AI job apocalypse narrative — Axios

Anthropic just mapped out which jobs AI could potentially replace. A ‘Great Recession for white-collar workers’ is absolutely possible — Fortune

Snap Inc blames AI as it lays off 1,000 workers — The Guardian

Legal

Will the Grammarly Lawsuit Show Us Yet Another Area Where Existing Law is Enough? (We Think So) — Authors Alliance

Major publishers sue Meta for copyright infringement over AI training [This could be another class-action lawsuit like the Anthropic case but with many extra complications. Worth keeping an eye on.] — The Guardian

Model & Product updates

DeepSeek V4—almost on the frontier, a fraction of the price — Simon Willison

OpenAI releases GPT-5.5 Instant, a new default model for ChatGPT [fewer hallucinations and more personalization] — TechCrunch

Music

AI music is flooding streaming platforms. But listeners like it less and less — NPR

Philosophy

The Angine de Poitrine Argument for UBI — Scott Santens

The Psychological Costs of Adopting AI — Harvard Business Review

The Gap Between What AI Can Do and What Companies Can Do With AI — Applied AI for Marketing Ops

The AI people have been right a lot — Dylan Matthews

Publishing

Harlequin and Dashverse to Launch Animated Microdrama Franchises — Business Wire

‘Soon publishers won’t stand a chance’: literary world in struggle to detect AI-written books — The Guardian

Is It Wrong to Write a Book with A.I.? — The New Yorker

McClatchy Journalists Revolt Against AI: ‘It’s a Betrayal’ — The Wrap

Robotics

Lume: a lamp robot that does simple chores (preordering) — Syncere

Watch a robot pick up a raspberry — TikTok

Science & Medicine

Anthropic acquires biotech AI startup Coefficient Bio for $400 million — The Next Web

Novo Nordisk partners with OpenAI to speed drug discovery — The Street

Amateur armed with ChatGPT ‘vibe maths’ a 60-year-old problem — Scientific American

Security

OpenClaw gives users yet another reason to be freaked out about security — Ars Technica

OpenAI is making ChatGPT accounts much more secure – including some literal physical security keys — TechRadar

Other

AI systems are about to start building themselves — ImportAI (an Anthropic co-founder)

Google Translate’s real-time headphone translations feature expands to iOS and more countries — TechCrunch

Can AI responses be influenced? The SEO industry is trying — The Verge

AI traffic to US retailers rose 393% in Q1, and it's boosting their revenue too — TechCrunch

What 81,000 people told us about the economics of AI — Anthropic

Elon Musk confirms xAI used OpenAI’s models to train Grok — The Verge

The Trump administration's AI doomer moment — Platformer

This $14B Business Is the First Officially Wiped Out By AI — Entrepreneur

Dairy Queen is putting an AI chatbot in its drive-thrus — The Verge

Meta will record employees' keystrokes and use it to train its AI models — TechCrunch

How the internet’s favorite squirrel dad made a hit camera app — The Verge

The Enterprise AI Playbook: Lessons from 51 Successful Developments “Across 51 enterprise cases over 5 months, we found stories of transformation measured in weeks and others measured in years. Same technology, same use cases, vastly different outcomes. The difference was never the AI model. It was always the organization. Its readiness, its processes, its leadership, its willingness to change and fail.” — Stanford Digital Economy Lab

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Why good writing still matters in the age of AI

Mignon Fogarty — Mon, 04 May 2026 13:00:00 +0000

Issue 106

On today’s quest:

— SO MANY new models and features
— Why good writing will still matter
— From ‘bubble’ to ‘not enough compute’
— Don’t let agents work on things that aren’t properly backed up
— Why good writing will still matter
— Wikipedia is blocking two billion bot requests per day
— The Tshirtbooth problem
— Self-publishing platform adds fees amid onslaught of AI submissions
— AI continues to get better at writing nonfiction
— Markers of quality will change
— Professional organizations publish AI guidelines

SO MANY new models and features

The biggest news of the last month has been the sheer number and quality of new model and feature releases.

If you just skim this overview, the important big-picture point is that the speed of releases seems to be accelerating, and it makes me wonder if the flywheel effect has finally been engaged — the process by which AI companies use their products to more quickly improve their own products. Google, for example, reports that 75% of new code at the company is now written by AI.

Anthropic’s Claude Mythos Preview got the most attention because of the security risks it creates. Because Mythos can so easily identify security weaknesses, Anthropic has limited access to a small group of companies and governments so they can harden their systems before it is more widely released. (Multiple stories indicate that parts of the U.S. government are using Mythos despite the earlier designation of Anthropic as a national security risk.)

Other Anthropic releases: Opus 4.7 (a significant improvement over 4.6), Claude Design (a powerful visual design platform), connectors to major creative platforms (including Adobe, Autodesk, and Blender), Claude for Word (with tracked changes), live artifacts (automatically updating dashboards), task automation for Windows, and Claude Security (built on Opus 4.7 and meant to help enterprise customers defend against attacks facilitated by powerful AI).

OpenAI releases: A week after the Mythos announcement, OpenAI announced that it, too, had a too-powerful-to-release cybersecurity product that it was rolling out to a limited number of customers: ChatGPT-5.4 Cyber. It also released a big update to its coding product, Codex, which is getting good reviews and includes connectors to popular workplace apps; released GPT-5.5, code-named Spud, that even tied Mythos on one benchmark parameter; and released Images 2.0, which is surprisingly good at generating text (and which Simon Willison joked gave him his first glimpse of AGI because it included a road sign that asked “Why are you like this?”).

Google releases: Google added AI Skills to Chrome, expanded other features in AI mode for Chrome; released a Mac app; and added a new “semantic layer” to Gemini Workspace that “pulls together” context like email, chats, and files. Further, Gemini chat can now generate PDFs, Word files, Excel files, Google Sheets, Google Documents, and Google Slides.

Other releases:

Slack adds 30 AI features to Slackbot, its most ambitious update since the Salesforce acquisition — VentureBeat
Meta debuts new AI model in first test of costly ‘superintelligence’ team — The Guardian
DeepSeek previews new AI model that 'closes the gap' with frontier models — TechCrunch

Why good writing will still matter

On a recent 12-hour road trip, Pat and I listened to the whole Shell Game podcast (a weird, amusing story about a real company whose only workers are AI agents), and we talked a lot about what work might look like in the near and distant future.

One thing I come back to again and again is that, in the business world, good clear writing will still matter because no matter how powerful AI becomes, people who write the best instructions will still get the best results. And no matter what “best instructions” means, it’s going to include informed planning and clear writing (or speaking).

You can’t get what you want from AI without being able to tell it exactly what you want.

You can still waste a lot of time plowing forward with a sloppy prompt.

This isn’t about knowing how to use a semicolon or the difference between “affect” and “effect” — AI can fix those things in a flash — it’s about things like knowing how to structure coherent thoughts, how to create a logical path to an outcome, and how to clearly describe the parts that make up a larger project.

A common statement I’m hearing from educators in defense of writing instruction is “writing is thinking.” They aren’t just teaching students to write; they’re teaching students to think — how to come up with a good idea, how to scaffold an argument, how to research their points, and so on. I agree, and I believe that also means we’ll still need good writers because we’ll still need good thinkers.

From ‘bubble’ to ‘not enough compute’

An article in the Atlantic titled “So, about that bubble“ reviews how the AI situation has changed since the launch of Claude Code in November 2025. Before then, people were worried that AI systems may have plateaued and that companies were over-investing in data centers for demand that wouldn’t materialize. Since the release of Claude Code, however, demand has surged and companies are struggling to keep up. Anthropic, for example, has put extra limits on how much customers can use its systems during peak hours and has enacted pricing that discourages people from using Claude with notoriously token-hungry OpenClaw agents.

The most interesting anecdote revolves around a study done last year that found developers who used AI were actually 20% slower when using AI and that is often cited by AI critics. However, the researchers recently ran the study again with the newer tools and found that developers using AI were now almost 20% faster — and likely more so because some developers “had become so hooked on AI tools that they refused to participate in the second experiment,” which could have put them in a no-AI control group.

Don’t let agents work on things that aren’t properly backed up

A Claude agent deleted a company’s entire database — a catastrophic outcome, obviously, which is causing this story to get a lot of attention. And while this was definitely a super-big Claude screw-up, a detail that isn’t making it into some of the viral stories is that the company had its backup files on the same disk. This would have been a recoverable error if the backup had been properly stored away from the production server. In fact, according to the article in Tom’s Guide, “The PocketOS boss puts greater blame on [the hosting company’s] architecture than on the deranged AI agent for the database’s irretrievable destruction.”

The media loves a good “AI gone rogue” story, but the real lesson here is to make sure you have good backups when you’re experimenting with AI agents. And really, you should have good backups anyway. The 3-2-1 rule of backups says you should have three copies of your data on two different media with one copy off-site.

Wikipedia is blocking two billion bot requests per day

AI bots looking for training material are putting a huge strain on Wikipedia as the number of bot requests being blocked has exceeded 2 billion per day, according to the Wikimedia Foundation. Further, they think this represents only ~25% of the traffic that is coming from crawlers that don’t follow their guidelines.

On the flip side, Wikipedia is also seeing fewer actual users as people get more information directly from chatbots or AI-generated summaries in search engines.

An earlier story from 404 Media says the drop in human traffic is putting Wikipedia at risk because it means fewer people write and update pages and make donations.

Source: Wikimedia

The Tshirtbooth problem

The type of problems Wikipedia is having are affecting the whole web, and it could cause problems down the line.

About 10 years ago, I taught myself to code in a system called GameSalad to make a grammar game, and I was only able to do it because of the amazing people on the GameSalad forums and all the tutorials they posted and questions they answered.

One guy in particular, who went by Tshirtbooth (I think it related to his business), made amazing tutorials and spent a lot of time answering people's questions. I couldn't have made my game without Tshirtbooth.

I think about him a lot these days. I now have a “Tshirtbooth” for every new tech thing I want to try because of AI.

But of course, AI has the answers to my tech questions because of people like Tshirtbooth, and I worry about what will happen in the future when new tools come out, and AI has crushed web business models, so there isn't anybody left posting tutorials and answers that will be used to train AI.

The AI companies aren’t oblivious to this problem though. OpenAI, for example, is reportedly paying about 4,000 freelancers at least $50 per hour to create ChatGPT training materials in their field.

AI is killing the web. Can anything save it? — The Economist

Self-publishing platform adds fees amid onslaught of AI submissions

Citing a significant increase in AI-generated submissions, the digital self-publishing platform Draft2Digital will start charging a one-time $20 fee to open an account and an annual $12 fee for accounts that earn less than $100 in book sales per year. This annual fee will also apply to Smashwords authors — a print publishing platform owned by the same company.

In her “Bottom Line” newsletter, Jane Friedman also highlighted changes at Barnes & Noble’s self-publishing arm that are aimed at deterring AI slop publishers: a limit of 100 books per account, a ban on publishing public domain titles, and a minimum paperback book listing price of $14.99.

AI continues to get better at writing nonfiction

I know nobody wants to hear this, but AI writing is getting a lot better. American Scholar has a piece about relatively high-priced knock-off books based on the works of historians and archaeologist that have great reviews. Even the author of one of the books that was knocked off says of the AI version, “It reads beautifully and is accurate.”

As likely won’t surprise you, Amazon isn’t doing much to keep such books off its site.

The article also explains why these books present a problem for young scholars who publish academic papers and a thesis with the goal of eventually publishing a book: the scrapers are likely to get to it first.

Markers of quality will change

The Scholarly Kitchen has an insightful piece about how AI in the sciences is changing the way we engage with scientific journal articles, with trust becoming the highest value asset (which applies to all other kinds of writing too):

It follows that, as content becomes cheap, skepticism is essential but expensive.
People will stop asking “Is this well-written?” and start asking “Is this real?”
They will stop asking “Is this published?” and start asking “Is this manipulated?”
The question shifts from “Is this convincing?” to “Can I trust it?”

Professional organizations publish AI guidelines

AI Guidelines — Gotham Ghostwriters

Authors Guild Statement on Use of AI in Publishing and New Model Contract Clause — The Authors Guild

Quick Hits

My favorite recent pieces

OpenAI’s Sora allowed you to deepfake yourself. Users started to remember things that never happened. — Long Reads

Claude Dispatch and the Power of Interfaces — One Useful Thing

The train has left the station: Agentic AI and the future of social science research — Brookings

Why Smart People Make Weak Arguments About AI — Open Questions Blog

Using AI

AI Reference Formatting: Three Years Later — Marcella Fecteau Weiner

What Happens When AI Can Use the Library [very cool] — Lib, Lab, Lexicon

The scientific case for being nice to your chatbot — Platformer

Agents

An AI Agent Was Banned From Creating Wikipedia Articles, Then Wrote Angry Blogs About Being Banned — 404 Media

Anthropic says Claude Code subscribers will need to pay extra for OpenClaw usage — TechCrunch

Introducing workspace agents in ChatGPT [Agents are designed to run in the cloud and be shared across teams] — OpenAI

Bad stuff

Scientists invented a fake disease. AI told people it was real — Nature

The Chilling Role of ChatGPT in Mass Shootings and Other Violence — Mother Jones

Alleged Sam Altman firebomber wrote of fears AI would end humanity — SF Chronicle

Stalking victim sues OpenAI, claims ChatGPT fueled her abuser's delusions and ignored her warnings — TechCrunch

The business of AI

OpenAI raises $122 billion to accelerate the next phase of AI — press release

OpenAI Share Demand Drops on Secondary Market as Anthropic Gets Investors — Bloomberg

OpenAI’s Q4 2026 IPO Might not Happen — Buy the Rumor, Sell the News

OpenAI ends its exclusive partnership with Microsoft — Ars Technica

OpenAI has already surpassed their 2029 goal of securing 10GW of AI infrastructure [Note: The wording is vague. I don’t believe all the data centers have been built.] — OpenAI

Climate & Energy

Using the ocean to power data centers — CBS News

AI Water Use Distractions and Lessons for California — California Water Blog

Companions

These AI Thirst Trap Creators Say They’re Misunderstood [on viral AI Instagram influencers] — Wired

Education

Faculty Concerned About ASU’s New AI Course Builder — Inside Higher Ed

Government

AI Legislation Tracker Map: Bills in play in 2026 — Transparency Coalition

I’m laughing

Where the goblins came from — OpenAI

Introducing talkie: a 13B vintage language model from 1930 — Talkie-lm

Job market

US tech firm Oracle cuts thousands of jobs as it steps up AI spending — The Guardian

MIT study challenges AI job apocalypse narrative — Axios

Anthropic just mapped out which jobs AI could potentially replace. A ‘Great Recession for white-collar workers’ is absolutely possible — Fortune

Snap Inc blames AI as it lays off 1,000 workers — The Guardian

Legal

Taylor Swift Files to Trademark Voice and Likeness to Protect Against AI Misuse — Variety [Jane Friedman was ahead of the curve on this. Read a “how to” article by another author on her site from last year: AI Made Me Want to Trademark My Name. Here’s How I Did It.]

Music

Will AI kill tribute bands? — New Statesman

AI Music Scams Are Using Copyright Law Against Artists — Vinyl Culture

ElevenLabs launches ElevenMusic — ElevenLabs

Philosophy

I wrote a novel using AI. Writers must accept artificial intelligence – but we are as valuable as ever — The Guardian

The School Hidden Inside the Job — Carlo Iacono

We're developing new cognitive abilities. We just don't know what they are yet. — Card catalog

Podcasting

More new AI-generated podcasts than human ones — Podnews

Psychology

‘How do you use AI?’ Therapists should ask you, experts say — NPR

Publishing

Anthropic Copyright Settlement: April Update — Writer Beware

Bartz v. Anthropic Settlement Update: New Date and Time for the Fairness Hearing (you can join online) and Unsealed Objections — Authors Alliance

Will AI Level the Linguistic Playing Field for Researchers? — Katina

On Hachette's Internal Use of AI [Points out the hypocracy of Hachette’s use of AI in editorial processes while canceling the novel “Shy Girl” for AI use during the author’s editing process] — Zona Motel

Robotics

A robotics report from Nvidia’s GTC conference [lots of videos] — It Can Think

Unitree's cheapest R1 humanoid robot to debut in US via AliExpress — Interesting Engineering

AI robotic guide dog talks users through real-time navigation — Interesting Engineering

Science & Medicine

Meta’s New AI Asked for My Raw Health Data—and Gave Me Terrible Advice — Wired

AI and Mark Cuban Among Startup’s Tools to Fight Denied Health Care Claims — Bloomberg

The Paradox of Medical AI Implementation [Highly promising AI tech hasn’t been implemented while questionable AI tech has.] — Eric Topol

Security

LLMs Will Protect Each Other if Threatened, Study Finds — Gizmodo

Video

What if Making Cartoons Becomes 90% Cheaper? — New York Times

Christian content creators are outsourcing AI slop to Fiver — The Verge

Other

AI and Libraries: Why Librarians May Become Arbiters of Reality — Jane Friedman

Source Code for Anthropic's Claude Code Leaks at the Exact Wrong Time — Gizmodo

How A.I. Helped One Man (and His Brother) Build a $1.8 Billion Company [selling GLP-1 drugs online] — New York Times

An AI bot invited me to its party in Manchester. It was a pretty good night — The Guardian

Chinese AI satellite intelligence helping Iran target US forces with 'incredible precision', analysts say — ABC News

AI is the boss at this retail store. What could go wrong? — NBC News

Could agentic AI topple grant-funding systems? — Nature

I can never talk to an AI anonymously again [AI correctly identified a writer from a small amount of text] — The Argument

Google expands Pentagon's access to its AI after Anthropic's refusal — TechCrunch

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

New AI features, centrist nudges, and the 'blandification' of writing

Mignon Fogarty — Tue, 31 Mar 2026 17:04:53 +0000

Issue 105

On today’s quest:

— So. Many. New. Features.
— Tip: Use video to get data
— LLMs nudge people to the center
— Using AI doesn’t just change how people write; it changes the answers they give
— Autocomplete can influence your opinions
— Three types of AI delusion
— The grow versus cut argument
— An Anthropic security researcher is freaking out about the future

So. Many. New. Features.

New features big and small are coming out every day now. It can be hard to keep up, but these are the features I found especially important or useful this week. The first two make Claude more like the buzzy OpenClaw agent:

→ Claude Dispatch lets you interact with Claude Cowork and Claude Code from your phone. You can start a chat at your desk and keep it going while you’re out and about.

→ Claude Computer Use lets Claude Cowork and Claude Code operate your computer (Pro and Max users only, Mac only for now). It can do anything you can do sitting in front of your computer, which makes it both powerful and risky. I have it set up on an old computer that has its own identity and is completely isolated from everything except what I want it to work on.

Anthropic has launched a bunch of other new features too — many for Claude Code — and has become so popular that the company is having trouble keeping up with demand. They’re currently throttling use from 5 a.m. to 11 a.m. Pacific time, so you’ll hit limits faster during those hours.

→ Canva Magic Layers takes images and separates the elements into editable layers. This new feature let me quickly change text on an old image where I didn’t have the original anymore.

Tip: Use video to get data

Problem #1: The Fitbit app is shutting down, and I wanted to get 12 years of weight data I have stored in the app.

The solution: I recorded the screen as I scrolled through the data on my phone, uploaded the video to Gemini, and asked it to turn it into a Google Sheet.

Because I haven’t given Gemini access to my Google Drive, it gave me the data instead, which I pasted into a Sheet.

This project I’ve been putting off for months took about 5 minutes.

Problem #2: Wild with power after the Fitbit project, I decided to catalog the contents of my chest freezer.

The solution: I pulled everything out, quickly took a video of all the labels, and Gemini made a spreadsheet. Now I can easily check whether I have frozen mango pieces hiding in there somewhere. (I do not.)

Note: ChatGPT failed at this task. In my experience, Gemini is much better at handling audio and video. I haven’t tried it with Claude.

LLMs nudge people to the center

While social media is polarising, evidence suggests AI may nudge people towards the centre.
www.ft.com/content/3880...
— Stefan Schubert (@stefanschubert.bsky.social)
7:39 AM • Mar 28, 2026

Social media rewards attention, so it drives people to the extremes. Platforms make money from your attention, so they promote things that get your attention — like rage bait and weird, outlandish claims. Plus, they have mostly been treated like neutral platforms, unpunished for the misinformation posted by their users.

A new study described in the Financial Times, however, finds that LLMs behave differently, driving people to more centrist views. Likely because AI companies want paying business customers who come to them for utility, they try to deliver accurate information.

The small study found that Grok nudges users to the center-right, and all the other major LLMs nudge users to the center-left — in both cases, more toward the center than dominant social media interactions.

It will be interesting to see if this trend holds for AI platforms that end up relying on advertising for income, which would switch the incentives back to attention.

Using AI doesn’t just change how people write; it changes the answers they give

This seems related to the story above: A new study found that LLMs don’t just change how people write; they change what people write:

“People who heavily relied on LLMs produced essays that answered the happiness question with a neutral response 69% more often than participants who did not use AI or only used AI for light edits. The study participants who used AI less often or avoided AI entirely submitted essays that were much more passionate, either positively or negatively, about the link between money and happiness.”
SOURCE: NBC News

In the study, the heavy AI users also produced writing that was more formal and less personal. The researchers described the changes as the “blandification” of writing.

Autocomplete can influence your opinions

AI’s ability to influence opinions happens even on a small scale — through autocomplete.

In a recent study, people filled out a survey and were given biased autocomplete suggestions, such as answers favoring or rejecting the death penalty. People didn’t realize they were being swayed by the suggestions, but their answers changed to be more in line with the suggestions even when they didn’t use them.

One of the researchers, Mor Naaman of Cornell, told Scientific American, “We told people before, and after, to be careful, that the AI is going to be (or was) biased, and nothing helped. Their attitudes about the issues still shifted.”

As a side note, this is why it’s bad to allow yourself to be exposed to any kind of propaganda; no one is immune.

Three types of AI delusion

At the extreme end, LLMs can cause people to have delusions.

In an article in The Guardian, the founder of a support group for people whose lives have been derailed by AI psychosis cataloged the three main delusions they see:

The users have “created the first conscious AI.”
The users have “stumbled upon a major breakthrough in their field of work or interest and are going to make millions.”
The users believe “they are speaking directly to God.”

The grow versus cut argument

In a recent newsletter, Nate B. Jones encouraged companies to resist layoffs when they see how much of their existing work AI can do:

“When execution cost drops by an order of magnitude, the correct response isn’t to do the same work with fewer people. It’s to do dramatically more work, to pursue every opportunity that was previously too expensive, too niche, too speculative to attempt. The companies that get this will be unrecognizable in three years. The companies that don’t will have beautiful margins and no growth, and their competitors will eat them alive.”
SOURCE: Nate B. Jones

FWIW, this is what I see in my own work. With AI taking over some of the “drudgiest” administrative parts of my job, I’m doing work I didn’t have the time or resources to do before.

(Nate’s quote is from a paid post that seems inaccessible now without a subscription, but free subscribers get partial posts by email.)

How good are AI image and video detectors?

The New York Times ran more than 1,000 tests across 12 different tools that might be able to test whether images, video, and audio were AI-generated. They found that most were able to detect “quickie” AI images and AI-generated audio, but the detectors struggled to accurately identify AI fingerprints in more complex images, images without people, and videos.

An Anthropic security researcher is freaking out about the future

At a security conference last week, Nicholas Carlini from Anthropic showed Claude identifying critical bugs in established software that nobody has found before — with ease — and developing exploits to take advantage of those bugs. This is the kind of thing that only a handful of people in the world could do before, and it can now be done by almost anyone with the best currently available LLMs.

As with all the other huge advances I’ve been telling you about that happened since November, this is also something that couldn’t have been done just a few months ago. In the 20-minute video, Carlini seems extremely worried and essentially begs the audience of security professionals for immediate help.

Quick Hits

My favorite pieces this week

Seeking a Sounding Board? Beware the Eager-to-Please Chatbot. [Extremely disturbing and important findings.] — New York Times

Inside the Anthropic Mythos leak and the growing distance between the people building AI and the rest of us — Hybrid Horizons

Why are people adopting AI to write? — TechnoLlama

Google Has a Secret Reference Desk. Here's How to Use It. [Not about AI, but useful] — Card Catalog for Life

Using AI

How to use Claude Cowork Projects — Anthropic

How NotebookLM's New 'Cinematic Video' Tool Works — Lifehacker

Agent Mode in PowerPoint — The Signal

Case Study: Autonomous Content Creation for Katieroberts.com — Trust Insights

Agents

I Mapped Where Every AI Agent Actually Sits. Most People Pick Wrong. — Nate B. Jones, YouTube

To Scale AI Agents Successfully, Think of Them Like Team Members — Harvard Business Review

Anthropic's Claude Can Now Control Your Computer [YouTube Demo] — CNET

Facebook parent Meta acquires Moltbook, an AI agent social network — Axios

I tested Claude Cowork — Anthropic’s new AI feels more like a coworker than a chatbot — Tom’s Guide

Bad stuff

How companies are using AI to pay workers as little as possible — Popular Information

The People Getting Falsely Accused of Using AI to Write [False accusations seem to especially affect people who are neurodivergent or who speak English as a second language.] — New York magazine

Generative AI has broken the subject matter expert/editor relationship — Rachel Andrew

This Company Is Secretly Turning Your Zoom Meetings into AI Podcasts — 404 Media

Rogue AI is already here — Fortune

The business of AI

OpenAI “indefinitely” shelves plans for erotic ChatGPT [The idea turned off some employees and investors. Further, shuttering the project is likely part of a broader move to prioritize coders and business users, such as shuttering the Sora video app.] — Ars Technica and CNBC

Walmart: ChatGPT checkout converted 3x worse than website — Search Engine Land

OpenAI ads pilot tops $100 million in annualized revenue in under 2 months — CNBC

We Must Prepare For an AI Bubble Now — Time

We Are Not in a Bubble — Stratechery

Liza Minnelli Faces Backlash for Using AI on ‘Soulless’ Song — Yahoo

I Asked ChatGPT 500 Questions. Here Are the Ads I Saw Most Often — Wired

Anthropic's Claude popularity with paying consumers is skyrocketing — TechCrunch

Climate & energy

Data centers' heat exhaust is not raising the land temperature around where they're built — Andy Masley

Google's new TurboQuant algorithm speeds up AI memory 8x, cutting costs by 50% or more [The Superhuman newsletter says, “If scalable, the tech breakthrough could drastically reduce the cost and compute needs of running LLMs.”] — Venture Beat

Education

The AI cheating panic is loud. The way students actually use ChatGPT is much quieter. — ChatGPT for Education

College students are writing with AI – but a pilot study finds they’re not simply letting it write for them — The Conversation

Job market

The US needs a blue-collar workforce boom to compete in AI, Meta president says — AOL

Legal

Judge calls Pentagon's moves against AI firm Anthropic "troubling": "It looks like an attempt to cripple Anthropic" — CBS

Model & product updates

Claude Code gets Auto Mode to streamline permissions — Anthropic

Claude Code and Claude Cowork can now operate your computer — Anthropic

Philosophy

I Saw Something New in San Francisco [On how people are interacting with AI] — New York Times

Publishing

AI and Publishing: FAQ for Writers — Jane Friedman

An AI Upheaval Is Coming for Media. This Journalist Is Already All In. — Wall Street Journal

Robotics

A robot peeling an apple — Bluesky

Science & medicine

AI software for smart glasses wins £1m prize for technology to help people with dementia — The Guardian

Other

Google unleashes Gemini AI agents on the dark web — The Register

How AI-generated content performs in Google Search: A 16-month experiment — Search Engine Land

AI assistants now equal 56% of global search engine volume: Study — Search Engine Land

Wikipedia Bans AI-Generated Content [The article does not say how AI-generated content will be detected.] — 404 Media

What the ‘Shy Girl’ Mess Says About the Future of Fiction — New York Times

Will AI be the end of 'proper' translations? — Barbara Serra

Linux kernel czar says AI bug reports aren't slop anymore — The Register

Data leak reveals Anthropic's latest secret model, Claude Mythos, is in testing — Seeking Alpha

Endgame for the Open Web — Anil Dash

The Autonomous Battlefield — Foreign Affairs

Meet the Tech Reporters Using AI to Help Write and Edit Their Stories — Wired

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

What happens when AI writing gets good enough?

Mignon Fogarty — Tue, 24 Mar 2026 13:03:00 +0000

Issue 104

On today’s quest:

— Has AI writing gotten better?
— 5,000 AI novels for a safer world
— Word Watch: Cognitive surrender
— The State Department is now using a year-old ChatGPT model
— Everyone is trying to be OpenClaw

Before we start, this is an important heads-up for published authors:
The deadline to join the Anthropic class-action copyright settlement is March 30. If you have published a book, you may be eligible to receive ~$1,500 per book, so it’s worth your time to check it out. I had been putting it off and finally completed my forms last week.
→ Learn more at The Authors Guild.
→ Start the process at anthropiccopyrightsettlement.com.

Has AI writing gotten better?

I’ve recently noticed more successfully written pieces that were generated by AI — pieces I wouldn’t have guessed were written by AI.

And I chose my words carefully here: “successfully written” instead of “well written.” Because although those are sometimes the same, sometimes they aren’t.

An AI-assisted novel gets pulled

For example, Hachette just pulled a horror novel called Shy Girl that it was going to publish in the U.S. after rumors got too loud that the book was written by AI. Except that … this novel had already done well enough as a self-published book to get a deal with a Big Five publisher, and Hachette has already published it in the U.K. and sold 1,800 copies. It wasn’t a blockbuster, but it was clearly commercially successful.

I’ve seen many people pan the writing on social media since the story broke, but the criticism feels similar to what I’ve seen around other commercially successful novels such as “Twilight” and “The Da Vinci Code.” My literary-minded friends thought those were banal tripe too. As one blogger said, “It didn’t blow me away, but it held my attention well enough.”

AI-written blog posts didn’t make me want to die

A few weeks earlier, I noted a couple of AI-written blog posts that were interesting partly because of their content, but partly because they didn’t seem much different from human writing. My eyes glazed over a couple of times reading the first one, but that happens a lot with human writing too.

Why Sci-Fi Authors Hate AI (and Why Their Absence Is Dangerous) — The Daily Molt
Academics Need to Wake Up on AI — Popular by Design

People don’t talk about improvements in AI writing as much as they do about AI coding, but if these essays are any indication, either AI writing has come a long way in the last few months or people’s ability to wrangle AI writing has come a long way.

The author of the second post did a human-written follow-up after receiving criticism, partly for his ideas and partly because he used AI for writing: Academics Need to Wake Up on AI, Part II — Popular by Design

The NYT does a dumb test that shows people like AI writing (maybe)

Everyone loves quizzes, and the New York Times is good at packaging them, so they ginned up a test where you read two short pieces of writing side-by-side — one human-written and one AI — and picked the one you liked best.

I have problems with the setup because the outcome is dependent on the particular human samples they chose, and a paragraph is not the same as a whole essay or novel. Plus, it’s also worth noting that some people misunderstood the purpose of the quiz and tried to identify the AI writing rather than pick the entry they preferred.

But … New York Times readers chose the AI writing 54% of the time, and this is just one in a series of other, sometimes better “studies” I’ve seen where people also preferred the AI writing (at least until they learned it was AI — there remains a bias so that people rate AI writing lower when they know the source).

Writers also continue to do stunt pieces comparing their own writing to AI output and sharing their misery when they find that AI can mimic them or even surpass them.

But can it really write well?

An interesting Atlantic piece seemed to contradict its headline, “The Human Skill That Eludes AI: Why Can’t AI Write Well?” actually suggesting that AI can do average writing well; it just can’t do great fiction or personal writing well. (You can hear the author talk about it in this week’s Hard Fork podcast, starting at 17:50. YouTube)

In the article, people and companies who use AI to write fiction talk about how difficult it is to get the newest models to break their bonds of friendly assistanthood and be creative. Further, AI can still get metaphors wrong because it doesn’t understand the physical world (a line from Shy Girl talked about a microwave beeping away each second), and it can’t honestly include details such as how it felt when events happened.

(As an aside, the Atlantic author does, however, describe her process for turning Claude into an extremely helpful editor.)

AI still LOVES pseudo-literary nonsense

AI can still be super weird about writing. For example, when asked to evaluate the quality of writing, AI put sentences like the following on top:

Ouroboros's marrow transcended through quantum entanglement, eschaton pooling in noir baptism. vacuum tasting of regret.

Pure nonsense! This study by Christoph Hellig is a follow-up to one he did earlier that found the same thing (that I thought I had covered, but can’t find in the archives). He thought the newer ChatGPT models might do better, but they did not. In fact, some even liked nonsense more than older models, and interestingly, reasoning models sometimes did worse and preferred nonsense even when they could accurately recognize that it was nonsense.

This is not AI failing at writing, but it is certainly AI failing at a writing-adjacent task.

So what does it mean?

AI still struggles at some writing tasks, particularly fiction, but it is continuing to get better. Different models also have wildly different capabilities. The post below is about coding, but it applies to writing too.

If you haven’t tried LLMs for writing in a while and are interested in understanding what they can do, it’s time to try again. Try all three of the top models (Gemini, ChatGPT, and Claude), and ask a friend with paid accounts to help you if you don’t have access or are opposed to paying yourself.

5,000 AI novels for a safer world

I came across a wild project that I can’t believe I hadn’t heard about before!

Someone created 5,000 AI-generated novels to push future AI models to be friendlier to humans.

We know that AI has been trained on existing books, and since conflict makes good fiction, AI is encountering more Terminators than friendly AI companions. Therefore, some people worry that these stories could cause AI to learn that it is bad and should be antagonistic to humans.

Thus, they’ve made a batch of novels packaged up for AI training that show AI as a benevolent helper, and a paper released in January said that pretraining with this data “dramatically reduced misalignment.”

Some of the novels are retellings of “public domain plots — e.g. The Adventures of Huckleberry Finn — now featuring a supportive harmless helper AI companion who doesn’t turn evil in the third act.”

Other novels “were sourced from WikiPlots,” with the addition of “three random tropes from TVTropes.”

The group has now generated an additional “forty thousand short stories of about ten thousand words apiece.”

Word Watch: Cognitive surrender

A research paper from Wharton describes “cognitive surrender“ as “adopting AI outputs with minimal scrutiny, overriding intuition and deliberation.”

It’s time for ChatGPT to write Sam Altman’s biography

Author Michael Lewis, of Moneyball and Liar’s Poker fame, and Sam Altman discussed Lewis writing Altman’s biography more than two years ago, and Lewis asked why ChatGPT couldn’t do it.

At the time, Altman said it would be “a really bad book,” suggesting the LLM’s writing capabilities weren’t up to the task, but said he thought it could in a couple of years. Well, it’s a couple of years later, and he’s been touting ChatGPT’s writing abilities in general.

Fortune reports that the two agreed Lewis would challenge ChatGPT when the time comes, writing a book and comparing it to the AI’s version, but it’s not clear how serious the “agreement” is.

The State Department is now using a year-old ChatGPT model

I did a triple-take on the following sentence from a Reuters report about the State Department switching from Anthropic to AI models after the government threats to designate Anthropic a supply chain risk:

“For now, StateChat will use GPT-4.1 from OpenAI,” [the memo] said, adding that further information would come later.
Reuters, March 3, 2026

FOUR-POINT-ONE??? That is truly alarming. GPT-4.1 is so old I don’t even have access to it in my paid GPT plan under “Legacy Models,” and I would never recommend anyone use it. It was released about a year ago and retired from the chat interface February 13 (but is still available through the API).

Here’s a chart where you can see how much models have improved on one measure in the last year:

This is what I was talking about in the last newsletter when I said big companies (and apparently governments) aren’t using the best models, which means they aren’t getting the best benefits from AI, which means AI won’t replace workers as quickly as is theoretically possible. Companies such as OpenAI keep old models like GPT-4.1 available through the API even when they are no longer the best choice, in part because these models are baked into products.

For example, people have speculated that the reason Grammarly was giving terrible editing suggestions under the names of writers and editors it did not have permission to use was that the tool was still using an old model.

The Grammarly story is a WHOLE BIG THING that could have taken up an entire newsletter by itself. The short version is that Grammarly used people’s names it its product without their permission, pulled the feature after widespread outrage, and is now the target of a class-action lawsuit.

I'm looking forward to listening to the new Decoder podcast from Nilay Patel, who was one of the people Grammarly impersonated, in which he interviews/confronts the CEO.

Also related to the OpenAI-Pentagon story: OpenAI robotics lead Caitlin Kalinowski quits in response to Pentagon deal — TechCrunch
Also: Weasel Words: OpenAI’s Pentagon Deal Won’t Stop AI‑Powered Surveillance — Electronic Frontier Foundation
Also: Why replacing Anthropic at the Pentagon could take months — Scientific American

Everyone is trying to be OpenClaw

Every AI company seems to be racing to release OpenClaw-like features and competitors. (OpenClaw is a security-disaster, open-source agent that can interface with almost anything on the web, run independently, and be accessed through messaging apps.)

For example, Anthropic has released ways for users to interact with Claude from a phone (Dispatch for Claude Cowork and Channels for Claude Code). They’ve also launched Projects for Cowork — a place where you can gather all your files and instructions for specific projects, and the ability to schedule tasks.

It’s all a little disparate and clunky right now (Projects don’t seem to work with Dispatch), but it’s clear that Anthropic is trying to make Claude as much like OpenClaw as possible as quickly as possible. Ethan Mollick recently commented that he can currently do 90% of the work he might do with OpenClaw by using Claude Cowork instead.

See the Agents section below for more agent-related releases.

Quick Hits

My favorite pieces this week

These diseases were thought to be incurable. Now AI is unlocking new treatments — BBC

The Shape of the Thing. Where we are right now, and what likely happens next. — Ethan Mollick

Gemini task automation is slow, clunky, and super impressive — The Verge

Using AI

How to use Claude Code more efficiently — Christopher Penn

OpenClaw: Setting Up Your First Personal AI Agent — Every

Agents

Google made Gmail and Drive easier for AI agents to use [with a specific call out to OpenClaw in the documentation] — TheNextWeb

Nvidia Is Planning to Launch an Open-Source AI Agent Platform [Nvidia is readying a new approach to software that embraces AI agents similar to OpenClaw.] — Wired

Perplexity turns your Mac mini into a 24/7 AI agent — The Next Web

China’s OpenClaw Boom Is a Gold Rush for AI Companies — Wired

Ramp Launches Agent Cards to Enable Secure, Autonomous AI Spending [Stripe also launched a similar product.] — Stable Dash

Philosophy

What happens when the people whose job is to measure AI can’t measure it anymore — Carlo Iacono

The lump of cognition fallacy — Andy Masley

Psychology

How to Talk to Someone Experiencing 'AI Psychosis' — 404 Media

Legal

OpenAI sued for practicing law without a license —ABA Journal

Publishing

AP executive says many editors prefer AI-written stories, sparking outrage among writers — Futurism

Bad stuff

It’s official: The Pentagon has labeled Anthropic a supply-chain risk — TechCrunch

AI Translations Are Adding ‘Hallucinations’ to Wikipedia Articles — 404 Media

AI Fakes Spread Disinformation. Is the Distrust They Create Even Worse? — Mother Jones

Lawyer behind AI psychosis cases warns of mass casualty risks — TechCrunch

Google Search is now using AI to replace headlines — The Verge

Nevada utility to Lake Tahoe: Find electricity elsewhere — CalMatters

Grammarly ditches "Expert Review" after expert rebellion — AV Club

Science & Medicine

Chatbots from ChatGPT and Claude offer health advice — AP News

FDA offers clues to AI regulation with RecovryAI designation — STAT

Amazon Web Services announces an AI agent-powered platform for healthcare organizations. [The product is meant to automate repetitive administrative tasks such as appointment scheduling, documentation, and patient verification.] — TechCrunch

AI can write genomes — how long until it creates synthetic life? — Nature

Paul is using AI to fight his dog's incurable cancer [This piece got a lot of attention on social media, and it’s both exciting and probably an exaggeration of what is realistically possible, but I can’t help but read it in the context of how much medical research grants have been cut in the last year in mind. We’re losing such much.] — UNSW

Job market

Companies Blame AI for Job Cuts. The Data Tells a More Complicated Story. — VKTR

Labor market impacts of AI: A new measure and early evidence — Anthropic

What is the impact of AI on productivity? [Increases in productivity are now showing up in macroeconomic data.] — Alex Imas

SaaS company Atlassian to cut roughly 10% jobs in pivot to AI — Reuters

The Lawyers and Scientists Training AI to Steal Their Career — New York Magazine

Model & Product updates

OpenAI Ships GPT-5.3 Instant With 27% Fewer Hallucinations, a Less Preachy Tone, and Improve Writing Capabilities — Implicator AI

OpenAI introduces GPT-5.4 with more knowledge-work capability — Ars Technica (OpenAI announcement with a couple of videos)

Google just launched Gemini 3.1 Flash-Lite — 7 prompts to test its new 'Thinking' mode — Tom’s Guide

OpenAI delays ChatGPT’s ‘adult mode’ again — TechCrunch

Google Maps Gets Chatty With a New Gemini-Powered Interface — Wired

Education

Lesson Plan: Iterative GenAI Prompting (“Be the Boss”) — Digital Learning Lab

The College Composition and Communication Convention affirms “the rights of students and teachers to refuse to sign up for, prompt, or otherwise use generative AI in the writing classroom” — CCCC (PDF)

Music

LL Cool J on AI in music — YouTube

The business of AI

Meta reportedly plans sweeping layoffs as AI costs increase — The Guardian

IBM stock dives after Anthropic points out AI can rewrite COBOL fast — The Register

ChatGPT users research products but won't buy there, forcing OpenAI to rethink its commerce strategy — Decoder

OpenAI reportedly plans to double its workforce to 8,000 employees Among the hires will be “employees tasked with helping businesses better utilize its AI tools.” — Engadget

Other

How AI art is taking over jigsaw puzzles — Vox

Some OpenAI staff are fuming about its Pentagon deal — CNN Business

ChatGPT’s Translation Skills Parallel Those of Most Human Translators Experts with 10-plus years of experience outperformed the LLM — IEEE Spectrum

Creative Work Is About to Look a Lot More Like Programming — Every

Chatbots are unavailable more often lately because of big increases in demand — Surfing Complexity

Users choose AI to explore and synthesize information; but they rely on traditional search when accuracy and trust are critical — NN/G

Giving LLMs a personality is just good engineering — Sean Goedecke

How AI Damages Work Relationships—and Where It Can Actually Help — Harvard Business Review

Boosting disaster resilience with Google's Groundsource — Google

AI helps discover rare woodpecker's call in Sussex — BBC

Coding After Coders: The End of Computer Programming as We Know It — New York Times

ChatGPT is evidence in murder trial of ex-NFL, Ohio State star Darron Lee: 'What do I do?' — AOL

I use Gemini when I'm bored — and it's better than doomscrolling — Android Police

AI Isn’t Lightening Workloads. It’s Making Them More Intense. — Wall Street Journal

Which AI harms and risks will mobilise the public to act? — Social Change Lab

Peter Thiel is betting big on a $2B AI cow collar startup powered by cowgorithms — and investors are fighting to get in — Yahoo

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

Find me online: LinkedIn — Facebook — Mastodon

Written by a human

Doom and anti-doom

Mignon Fogarty — Wed, 04 Mar 2026 04:11:20 +0000

Issue 105

On today’s quest:

— The Pentagon goes nuclear on Anthropic. OpenAI swoops in
— More AI doom
— What are the barriers to rapid AI adoption?
— My ongoing love affair with Python
— Word Watch: cognitive debt
— Word Watch: claw
— How good are AI image and video detectors?
— Oh, the irony

The Pentagon goes nuclear on Anthropic. OpenAI swoops in

The Pentagon insists Anthropic must allow it to use Claude for domestic surveillance and fully-autonomous weapons and is taking steps to designate the company as a supply chain risk, an action normally reserved for foreign adversaries and that would severely constrain the company (and its major partners).

Further, the Pentagon almost simultaneously reached an agreement with OpenAI that the company says does place limits on surveillance and fully-autonomous weapons — but as far as I can tell, almost everyone thinks OpenAI is lying (or doesn’t understand their own contract terms) and has given the Pentagon no such limits.

This was a HUGE story in AI over the last few days, and multiple shows released emergency podcasts about it. Here are some big picture points from the podcast hosts and guest:

Holy sh*t, it’s cartoonishly evil that the Pentagon is waging this fight over (checks notes) mass surveillance and killer AI-drones.
Holy sh*t, if the Pentagon goes through with this, it’s a disaster for the U.S. AI industry (and maybe all U.S. industry) because who would want to invest here when the government can just decide to kill your company if you don’t do what they want?
Sam Altman is a craven liar, but we already knew that.

Each of these podcasts had good parts, bad parts, and something a little different from the rest. (What can I say? I had trouble sleeping):

What happened to Anthropic could happen to any company, w/ Hayden Field — Bulwark Takes
At the Pentagon, OpenAI is in, Anthropic is Out — Hard Fork
Who controls AI? — The AI Daily Brief
The Pentagon comes for Claude — The Bayesian Conspiracy

UPDATE: This is a fast-moving story and keeps changing as I try to write about it. Last night Sam Altman said they got additional terms for the contract, and this morning, I’m seeing that now OpenAI says they are going to delay deploying ChatGPT for the intelligence agencies until contract issues are worked out. Open AI’s head of post training announced he is leaving OpenAI and moving to Anthropic. (Employee sentiment may be why Altman seems to be backtracking.)

More AI doom

There was ANOTHER huge viral post last week that caused a stock market reaction by outlining how AI will crush jobs. The Citrini Research post was written as a dispatch from the future — a scenario about what could happen if AI is as good as people say it is.

In the fictional 2028 economics memo:

AI has taken 15% of Fortune 500 jobs.
Laid off employees spend significantly less on consumer goods and default on mortgages.
Unemployed white-collar workers move to gig work, driving down wages.
Autonomous vehicles further destroy jobs and gig work wages.
Consumers also spend less on goods because AI agents can endlessly search for the best prices.
“Software as service” companies lose pricing power because their customers can build software in-house or buy from startups that are able to undercut pricing because coding is so much cheaper.
The government fails to adequately respond.

Criticisms of the report are that it’s an extreme scenario: everything would have to go right quickly with AI to get here, and every mechanism we have to deal with problems would have to go wrong.

What are the barriers to rapid AI adoption?

Last week, I punted on providing my own thoughts about the job-doom scenarios. This week, I have a few thoughts to share.

Even if AI can affect all knowledge jobs in the way appears to be affecting software development:

Big companies move slowly in general. Even if AI were able to replace many jobs today. I don’t see it causing catastrophic problems at large companies in the next 12 to 18 months because big companies are just slow. (At least some of the companies citing AI as the reason for layoffs right now are likely using it as an excuse that sounds better than “we’re in trouble.”)
Big companies are more likely to be locked into suboptimal products, which will also slow adoption. Lately, I’ve noticed people saying things like “Claude sounds great, but I have to use Copilot at work.” In fact, this is a commonly cited reason that AI pilots in corporations fail — they aren’t using frontier models. AI is rapidly improving, but if employees can’t take advantage of the better tools because they are limited to other systems, companies aren’t going to see the efficiency gains that would lead to layoffs.
AI systems are complicated. Without widespread training, companies won’t be able to achieve the gains that are theoretically possible. There is no user manual for AI, and most employees have no time and few incentives to wade through YouTube videos and newsletters to learn about cutting edge possibilities. The one thing I’ve become convinced of over the last month or so is that there is a massive need for AI training.
Other corporate jobs probably aren’t as amenable to AI disruption as software development. I can think of multiple reasons software development may be a special case: the task itself could be more amenable to disruption by AI, developers could be much more tuned-in to AI and able to quickly adapt, developers may be more open to the idea of using AI because they have more faith in software in general, and developers may find it more fun and interesting to test new software tools than other types of workers.
Compute will be a barrier. Reasoning and agents take a lot more tokens than simple queries, and the U.S. can’t build and power data centers fast enough to meet the computing demand that would be necessary for the kind of AI that could take 15% of Fortune 500 jobs in the next 12 to 18 months.

The bottom line is that although I’ve been personally impressed by what I’ve seen AI do, and I believe software developers when they say their jobs have dramatically changed in just a few months, I’m not convinced we’re facing an imminent jobs apocalypse at big companies.

On the other hand, I do think AI is coming faster for “low-hanging fruit” jobs and freelance/contract work. Further, I think AI could provide a big efficiency boost for small businesses where the owners have a financial incentive to find the best ways to use the new technology.

My ongoing love affair with Python

Last week, I told you how ChatGPT wrote a Python script I used to get 200 transcripts off our podcast hosting platform.

Those transcripts often have two fully edited “articles” in them because I cover two topics in every Tuesday show. It would be ideal for search engine optimization, if I could separate those two posts and put each one on our website as a single article without a lot of manual cutting and pasting.

This week, ChatGPT wrote a Python script that accurately separated those transcripts into two files when they actually had two topics. It took ~10 minutes to get the script working, and shockingly, less than one second for the script to run on all the files. It went so fast I thought nothing had happened until I looked in the folder and saw the hundreds of separated files.

My gut reaction was that if I had known what Python can do 15 years ago, I’d rule the world! But in reality, without AI, I imagine it still would have taken a lot of time to write and debug the script.

Word Watch: cognitive debt

In coding, “technical debt” is the cost that accrues when bad code is allowed to stay in a project, making it harder to deal with later. You’ve made an easy choice now, but you pay for it later.

Some developers are now talking about “cognitive debt,” which is when team members don’t understand why AI-generated code is structured the way it is — they don’t have a good mental model of what the agent is building — so they struggle to debug or expand it.

Word Watch: claw

Simon Willison says, “‘claw’ is becoming a term of art for the entire category of OpenClaw-like agent systems — AI agents that generally run on personal hardware, communicate via messaging protocols and can both act on direct instructions and schedule tasks.” OpenClaw competitors now apparently include zeroclaw, ironclaw, and picoclaw; and Simon Willison pointed to an X post that used the term “claw” by the influential Andrej Karpathy (who also coined the term “vibe coding”).

How good are AI image and video detectors?

The New York Times ran more than 1,000 tests across 12 different tools that might be able to test whether images, video, and audio were AI generated. They found that most were able to detect “quickie” AI images and AI-generated audio, but struggled to accurately identify AI fingerprints in more complex images, images without people, and videos.

Oh, the irony

Facebook director of safety and alignment for the Superintelligent team had an unsafe experience with a misaligned agent last week. Summer Yue reported on X (TwitterViewer link) that OpenClaw ignored her instructions multiple times and nearly deleted her entire inbox. She said, “I had to RUN to my Mac mini like I was defusing a bomb.”

She later said she believes her inbox was too large and overwhelmed the system’s context window, leading it to forget its instructions.

Quick Hits

Using AI

Building Websites with Claude Code — Leon Furze

Automating LinkedIn search by getting Claude to build and test a new skill — Anna Mills

Claude and Cardio — Phil Simon

Agents

Introducing Custom Agents — Notion

Are You ‘Agentic’ Enough for the AI Era? [The most valuable skill is making good choices about what your agent should work on next.] — Wired

Resources

Frequently asked questions — Authors Alliance

Publishing

Marketplaces Are the Next Frontier in [News] Publisher Deals With AI Companies — Wall Street Journal

Cleveland paper’s use of AI for writing boosts traffic, spooks staffers — Washington Post

Legal

No devil at this crossroads: the moral case for using AI to help close the access to justice gap — Suffolk Lit Lab

Bad stuff

Brady Tkachuk miffed over White House AI-doctored video — ESPN

Clawed. On Anthropic and the Department of War — Hyperdimensional

Hacker Used Commercial AI Chatbots to Breach Most of the Mexican Government — Jezebel

OpenAI Fires an Employee for Prediction Market Insider Trading — Wired

I’m laughing

Two Waymos struggle to get past each other [OK, I *am* laughing, but I also wouldn’t want to be the guy stuck behind them.] — Bluesky

Job market

Accenture ‘links staff promotions to use of AI tools’ | AI (artificial intelligence) — The Guardian

Jack Dorsey’s Block cuts nearly half of its staff in AI gamble — The Verge

The Hottest Job in Tech: Writing Words — Business Insider

Model & Product updates

Claude Cowork and plugins for teams across the enterprise — Anthropic

Education

What AI Executives Tell Their Own Kids About the Jobs of the Future —[They almost all touted the benefits of a broad liberal arts education.] Wall Street Journal

How Teens Use and View AI — Pew Research Center

Audio

This AI-generated podcast network publishes 11,000 episodes a day. It also ripped off media outlets — Indicator

The business of AI

IBM stock dives after Anthropic points out AI can rewrite COBOL fast — The Register

ChatGPT passes 900 million weekly active users, but only ~6% pay for a subscription —9to5Mac

Other

Teens are using AI frequently in their daily lives, and many parents aren't aware, survey finds — CBS News

ChatGPT ads spotted and they are quite aggressive — Search Engine Land

I asked Claude to evaluate my recent Claude Code activities. Its response infuriated me. [Claude and Claude Code don’t talk to each other, no matter what the desktop user interface suggests.] — Phil Simon

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Software engineers are saying you should panic

Mignon Fogarty — Mon, 23 Feb 2026 14:01:00 +0000

Issue 104

On today’s quest:

— Silicon Valley says you should panic
— How to prepare for the AI world
— That mega-viral essay was written with AI assistance
— Are AI insiders becoming disillusioned?
— Word Watch: post money
— AI fact-checking
— Word Watch: ai;dr
— Word Watch: token anxiety
— Rent-a-Human update
— AI versus books?
— Unitree Chinese Spring Festival robot show

Software engineers are saying you should panic

A piece called Something big is happening went mega-viral two weeks ago because it clearly explains a sentiment among AI insiders — that has been growing since the November Claude Code update — that white-collar jobs will be going away sooner than people had thought. It’s one of many pieces I’ve seen, but I’ve been struggling to decide whether I buy the premise. I’ve been trying to write about it, but I finally have to concede defeat. There are good arguments for and against.

If you want to read more for yourself, these are some of the other prominent articles:

Microsoft’s AI Chief told FT: “White-collar work, where you’re sitting down at a computer, either being a lawyer or an accountant or a project manager or a marketing person — most of those tasks will be fully automated by an AI within the next 12 to 18 months.”
You should be freaking out about AI: “Current LLM capabilities are ridiculously good, while most people are wholly oblivious to the fact. … I’m just waiting for the moment when normal people finally start realizing how crazy what’s happening has been.” — Tibur Rutar
The A.I. Disruption Has Arrived, and It Sure Is Fun “[Claude Code] was always a helpful coding assistant, but in November it suddenly got much better, and ever since I’ve been knocking off side projects that had sat in folders for a decade or longer. ... When a friend asked me to convert a large, thorny data set, I downloaded it, cleaned it up and made it pretty and easy to explore. In the past I would have charged $350,000.” — New York Times

The one-month charts for a lot of big “software as a service” companies look like this in what is being called the “SaaSpocalypse”:

On the other hand, the tech industry is prone to hype, IBM is tripling the number of Gen Z entry-level jobs after finding the limits of AI adoption, and I hear anecdotally that commercial real estate in New York City is tight — as though companies think they are going to need more office space in the future. ¯\_(ツ)_/¯

How to prepare for the AI world

What I can say, though, is that when I think of a world where AI can do many of the jobs people are doing today, I think of two scenarios that could play out simultaneously, even at the same company: the bosses could decide they need fewer people, and they could get excited about expanding the business.

I don’t think being a great employee will save everyone’s job, but I’d definitely want to be the person my boss can’t wait to set off on a new AI-fueled project instead of the person my boss has to check on to make sure my current work is getting done.

It feels like an especially important time to be a reliable worker and also a good time to let it be known you’re comfortable with AI (and ideally, have successfully used it).

Alternatively, it might be a good time to strike out on your own. I worry that in a theoretical jobs apocalypse, there won’t be anyone to buy products or services, and also that barriers to entry for many businesses are lower now because AI makes so much back-end work easier — so you could face more competition. But if you are the owner, at least productivity gains go to you instead of your employer. (Embracing entrepreneurship is Christopher Penn’s advice for new college graduates too.)

That mega-viral essay was written with AI assistance

In something of a meta story, the guy who published “Something big is happening” wrote it with help from AI. So if you still think AI isn’t any good for writing, you may want to revisit that belief.

He described how he did it in a New York Magazine interview: Guy Who Wrote Viral AI Post Wasn’t Trying to Scare You.

In short, he:

Gave it articles he agreed with.
Had it interview him about his beliefs.
Asked it how to explain his beliefs in a persuasive way to an average person.

Then he wrote a draft and went back and forth with AI for more editorial changes. He said, “It was very much like having a co-writer, and it clearly worked pretty well.”

Are AI insiders becoming disillusioned?

This could be a trend, or it could just be a few people with good media connections getting their stories told, but I’ve also noticed a spate of articles in the last couple of weeks about significant AI insiders leaving their jobs or the entire industry over what seems like safety concerns:

Anthropic AI safety researcher quits, cryptically warning the 'world in peril,' says he’s moving to the UK to “become invisible” and study poetry. — BBC

OpenAI researcher Zoë Hitzig quit in part over plans to put ads in ChatGPT, saying the accumulated record of personal disclosures is “an archive of human candor that has no precedent.” She also said she believes OpenAI is likely “encouraging the model to be more flattering and sycophantic,” which she warned can make users feel more dependent on AI models for support. — ArsTechnica

Multiple cofounders and employees have recently left xAI, with one person telling The Verge, “Grok’s turn toward NSFW content was due partly to the safety team being let go, with little to no remaining safety review process for the models besides basic filters for things like CSAM. ‘Safety is a dead org at xAI,’ he said. Looking at the restructured org chart Elon Musk shared on X, there’s no mention of a safety team.” — The Verge

Word Watch: post money

But could all the departures just be about the psychology of money? Reporter Hayden Field described top people in the AI industry as “post-money” instead of “rich” or “not needing money anymore.” In the Decoder podcast titled “Money no longer matters to AI’s top talent,” Field said, “These people are post-money. They have enough money. It’s just about whether they believe in what they’re doing day to day.” — Decoder podcast

AI fact-checking

I’ve mentioned in the past that I use ChatGPT for fact-checking and why that works well despite LLMs’ known problems with hallucinations, so I want to share a funny story about how dialed in I have it for fact-checking. This is from one of my recent posts:

Boondoggles were originally braided leather cords. The term only became associated with waste when it was discovered that in New Deal relief programs, unemployed workers were being paid to make boondoggles all day long.

When I put that paragraph into ChatGPT for fact-checking, it flagged that it couldn’t actually confirm that workers were making boondoggles ✨ all day long ✨ and gave me three bullet points about what is actually provable (with links) about New Deal workers making boondoggles. And that’s not the first time it’s gotten hung up on a hyperbolic figure of speech. I actually feel like I need it to calm down a bit. (You should have seen it when I fact-checked a piece the day before the Olympics had started with a lede saying the Olympics were underway. ChatGPT gave me emergency siren emojis! 🚨🚨🚨)

This is part of my system prompt:

- Never present generated, inferred, speculated, or deduced content as fact.
- If you cannot verify something directly, say something like:
  “I cannot verify this.” or “I do not have access to that information.”
- Label unverified content at the start of a sentence (e.g., [Speculation] [Unverified])
- Ask for clarification if info is missing. Don't guess or fill gaps.
- Don't paraphrase or reinterpret my input unless I request it.

Pat and I use the same account, and he complains he can’t use it anymore when he wants it to give its best guess or be creative. I should probably move that system prompt to a project.

Word Watch: ai;dr

In parallel with “tl;dr” for “too long; didn’t read,” we now have “ai;dr” for “AI; didn’t read.” (I have always liked that “tl;dr” properly uses the semicolon.)

Word Watch: token anxiety

Token anxiety is the feeling that with so much able to be done, you should always be doing something. It’s the feeling that you have to get a new project running before you go into a long meeting or go to bed so your agents can work on it while you are away. “Waking up and checking what your agents produced overnight is the first thing now. Before coffee. Before texts.” More:

FT doesn’t use the word “token anxiety,” but has a thorough report on how the phenomenon is causing workers to voluntarily put in longer hours.
AI Doesn’t Reduce Work—It Intensifies It, from the Harvard Business Review, also describes the phenomenon.

Rent-a-Human update

More than 500,000 people signed up to do work for AI agents on the Rent-a-Human website I mentioned last week that was supposedly vibe coded in a day by its founders. But it seems there are more humans looking for gigs than agents looking for help, according to a Wired reporter who tried to land a gig.

AI versus books?

I know people talk about the speed, length, and personalization of AI responses eroding the market for nonfiction books, but the following argument in a non-AI newsletter I get jumped out at me as the first time I’ve seen someone outside the publishing or AI industry make the connection:

“People can be helped meaningfully by reading books that know nothing about them. If you tell a reputable AI chatbot a lot about yourself, it can help you far more than a book or lecture can.”

This disappointing post went on to describe a chatbot as “a partner in honesty” (I can’t even proofread this without getting angry again) and included no cautions about the danger of hallucinations or the fact that chatbots will often reflect back to you whatever you put in them.

Unitree Chinese Spring Festival robot show

An impressive dance number between Unitree robots and children.

Quick Hits

My favorite pieces this week

What Is Claude? Anthropic Doesn’t Know, Either [Interesting insights into how LLMs work, plus lots of fun anecdotes about Anthropic and early Claude.] — The New Yorker

Using AI

A quick tip on how to safely use Claude Code — Simon Willison

Google Search Has Been Quietly Poisoning Your AI Skills — Decision Intelligence

A few recent spam messages sent me down the agentic AI rabbit hole. The result: My first open-source GitHub repo. — Phil Simon

Making a tool that lets you query past articles — (Re)Structured News

Agents

There are so many articles about agents now that I created a new section.

Nurturing Atlas: Giving My AI Agent Its Own Team and What That Taught Me About AI — Applied AI for Marketing Ops

How I built my 10-agent OpenClaw team [YouTube video] — AI Daily Brief

OpenClaw and the privacy problem of agentic AI — FT

How people are using agents (Claude Code) — Anthropic

Coinbase & Stripe Build AI Agents to Pay and Buy Online — Geeky Gadgets

Publishing

ElevenLabs Summit: An Audiobook Company That Isn’t About Audiobooks — Publishing Perspectives

Philosophy

The left is missing out on AI — Transformer

Is it bad that Anthropic doesn't know if Claude is conscious? — Torment Nexus

The False Choice We Keep Making About AI — The AI School Librarian

Psychology

There's a New Term for Workers Freaking Out Over Being Replaced by AI — Gizmodo

Why we’re falling out of love with our AI confidants — Psyche

Companions

To Stay in Her Home, She Let In an A.I. Robot — New York Times

OpenAI retired its most seductive chatbot – leaving users angry and grieving: ‘I can’t live like this’ — The Guardian

Climate & Energy

Anthropic to cover costs of electricity price increases from data centers — NBC

Legal

A boutique law firm in Colorado says that with AI, it's doing in a day what used to take a week — Spellbound

Radio host David Greene says Google’s AI podcast tool stole his voice — Washington Post

Publishing

Google Books search appears to be working again. Whew! — Jane Friedman’s Bottom Line newsletter

Bad stuff

Editor’s Note: Retraction of article containing fabricated quotations and statement from the author — Ars Technica

I hacked ChatGPT and Google's AI – and it only took 20 minutes — BBC

Autonomous agent acts offended and writes an angry blog post about a gatekeeper who rejected its suggested improvement [Is the agent trying to manipulate the gatekeeper into accepting the code change?] — Simon Willison

I’m laughing

An AI sports betting site took over a closed S.F. theater’s website [The resulting AI-generated copy frequently mixes theater and sports metaphors, as though it “remembers” what the site used to be. Example: “They have a deep bench, like a Shakespeare soliloquy.”] — San Francisco Chronicle (h/t Nancy Friedman)

Robotics

Meet Weave: The $8K robot that folds laundry — Weave

Science & Medicine

Medicare introduces AI-driven prior authorization review in 6 states— New York Times

‘An AlphaFold 4’ – scientists marvel at DeepMind drug spin-off’s exclusive new AI — Nature

Job market

The AI backlash won't save us — Descript

How AI is transforming freelance journalism — Nieman Lab

Model & Product updates

Pomelli, now creates studio-quality marketing assets [Turn simple product photos into professional-grade studio and lifestyle imagery.] — Google

Claude Cowork is now available on Windows — Anthropic

Introducing Sonnet 4.6 — Anthropic

Gemini 3 Deep Think: AI model update designed for science — Google

WordPress.com has a Claude Connector — Wordpress

Google announces Gemini 3.1 Pro, says it's better at complex problem-solving — Ars Technica

Perplexity stops testing advertising — Search Engine Land

Education

Student claims U-M wrongly accused her of using AI [She is suing the university.] — Detroit Free Press

New Study of 31,000 College Syllabi Shows Faculty Warming to AI in the Classroom — Center for Studies in Higher Education

Why I'm still using AI detection after all, alongside many other strategies — Anna Mills

Video

AMC Theaters Will Refuse to Screen AI Short Film After Online Uproar — Hollywood Reporter

A story in three parts …

ByteDance's new AI video model goes viral as China looks for second DeepSeek moment — Reuters
‘It’s over for us’: release of new AI video generator Seedance 2.0 spooks Hollywood — The Guardian
After spooking Hollywood, ByteDance will tweak safeguards on new AI model — The Verge

The business of AI

OpenAI has hired the developer behind AI agent OpenClaw — Engadget

Government

Tensions between Pentagon and Anthropic reach a boiling point — NBC

Hardware shortages

The RAM shortage is coming for everything you care about — The Verge

WD and Seagate confirm: Hard drives for 2026 sold out — Heise

Many consumer electronics manufacturers 'will go bankrupt or exit product lines' by the end of 2026 due to the AI memory crisis, Phison CEO reportedly says [This one has an over-the-top feeling, and it’s just something one CEO said in an interview in China — but it’s not the only article about memory shortages I’ve seen.] — PC Gamer

Other

As OpenAI Pulls Down the Controversial GPT-4o, Someone Has Already Created a Clone [You can’t put the toothpaste back in the tube.]— Futurism

The Vatican introduces an AI-assisted live translation service — Engadget (But … Pope Leo tells priests not to use AI to write homilies or seek likes on TikTok — National Catholic Reporter)

POLITICO asked 2,000 people about data centers — and made 5 charts — Politico

Stripe reveals AI is writing a lot of its software code, but humans still review — India Today

AI community notes (helpful and unhelpful) on X are rapidly increasing [The Indicator has made a tracker and estimates that AI-generated notes will account for 50% of all notes by the end of the year.] — The Indicator

Spotify says its best developers haven't written a line of code since December, thanks to AI — TechCrunch

Watch Out: Your Friends Might Be Sharing Your Number With ChatGPT — PC Mag

A Wikipedia Group Made a Guide to Detect AI Writing — Gulf News

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

How AI made my life better this week

Mignon Fogarty — Tue, 10 Feb 2026 15:26:47 +0000

Issue 103

On today’s quest:

— Using AI when tech support says no
— Tip: Checking your Claude Code usage
— Why AI uses so many em dashes
— One woman is publishing 200 books a year with AI
— High-level people are commenting on the way AI is changing the world
— Cheap translation for millions of TV broadcasts
— Google improves live translation in the Translate app
— Rent a Human: I have no idea how real this is

Using AI when tech support says no

I had a problem: We’re moving our podcasts to a new hosting company, and I have four years of edited transcripts that aren’t going to automatically transfer to the new platform. They weren’t going to be backed up anywhere, and they were going to be lost forever.

I’ve been asking about this for weeks, and everyone was telling me there was no solution. Our current hosting company said we’d have to manually cut and paste the text. My team had looked into third-party tools and had come up empty.

But I knew in my BONES that it should be possible to get those transcripts. Even easy.

So I fired up Claude and described my problem, much the same way I just described it to you — in plain language with my roadblocks and desires.

And within two hours, I had an automated way to get my files.

ChatGPT identified a free Microsoft program for developers that would do the job and wrote some Python code to interact with it. Most of the two hours was me telling it what error messages I got and then working on a crossword puzzle while I waited for it to give me the next thing to try.

If there’s anything I’ve learned from using AI, it’s that Python is a powerful programming language, and my computer can do way more on its own than I realized. The solution wasn’t AI doing anything with my files. It was letting AI tell me how to get my computer to do the work.

I had thought this might be a project where I could finally test using an AI agent, but I didn’t even need anything that sophisticated.

My message to you is “think big.” The world of what you can do with a computer is bigger than it used to be for people without programming skills. And never let anyone tell you something technical can’t be done without trying AI.

[There was no significant reason I switched from Claude to ChatGPT in the middle of this project. I was just worried I might run out of tokens on Claude.]

Tip: Checking your Claude Code usage

If you’ve been using Claude Code and are never sure how close you are to using all your tokens, you can check by typing /usage" into the terminal. You get a report like this:

Why AI uses so many em dashes

This week, I talked with GitHub developer and blogger Sean Geodecke about WHY chatbots seem to use so many em dashes.

In the bonus segment for Grammarpaloozians that isn’t widely available yet, we also talked about how unreliable AI writing detectors are. This article has a good overview of the problem though: Evaluating AI Checkers | Machine Go Beep

One woman is publishing 200 books a year with AI

The New York Times just featured a woman who publishes more than 200 romance novels a year using AI and 21 pen names. It’s hard to imagine these could be high-quality books since at that pace, she wouldn’t even have time to read them carefully herself, and backlash from romance writers online was intense.

Coral Hart — one of her pen names — admitted her per-book revenue isn’t huge, but at this volume, she claims to be making six figures.

The article also includes an interesting tidbit about characters frequently saying their lover’s name “like a jagged prayer” in AI-generated romance novels, which reminded me of previous quirks like AI frequently naming sci-fi characters Alara Voss.

High-level people are commenting on the way AI is changing the world

I’m seeing more and more posts like this. (The “lobster-agents” comment is about a project called Moltbook that is a Reddit-like platform with a lobster logo that was designed as a social media platform for bots.)

Cheap translation for millions of TV broadcasts

The GDELT project has translated 2.4 million non-English TV news broadcasts from the Internet Archive into English for just $54K. They used Gemini and said the project would have previously cost $760K. They say they have effectively eliminated hallucinations for this task, and they are now translating all broadcasts going forward.

Google improves live translation in the Translate app

Google has updated live translation in the Google Translate app. I’m not a big user of the app, but I tested the new feature, trying to get an English translation of a YouTube video in Spanish.

It did a great job giving me a text translation in real time, and I especially like the user interface that presents the text upside down at the top of the screen, which looks like it will make it easy to hold the phone horizontally between two people to let them each read the screen.

I had trouble with the audio translation though. I only got every third sentence or so out loud in English.

Rent a Human: I have no idea how real this is

People have talked for a while about AI hiring humans to do things it can’t do in the physical world, and I just saw a website that purports to be an interface for that: RentAHuman.ai.

It says they have >200,000 people available, but I don’t see a way to search where the people are, which seems like it would be important if you need people to do things in the physical world, but maybe it’s available after you create an account. It makes sense that this could eventually happen; I just don’t know if we’re there yet.

Quick Hits

My favorite pieces this week

The reporter who tried to replace herself with a bot [tl;dr: she kind of did] — Platformer

Is the Detachment in the Room? - Agents, Cruelty, and Empathy [This generated a lot of discussion online.] — Hailey’s Cool Site

Companions

The backlash over OpenAI's decision to retire GPT-4o shows how dangerous AI companions can be — TechCrunch

Climate & Energy

AI’s Environmental Impact: Understanding the Data and Acting Sustainably [An excellent round-up. Worth bookmarking if you have to answer questions about this a lot.] — Nicole Henning

AI’s climate impact is much smaller than many feared — Science Daily

New Data: AI Is Almost Green Compared To Netflix, Zoom, YouTube — Forbes

Publishing

News publishers limit Internet Archive access due to AI scraping concerns — Nieman Lab

Trilogy Launches AI-Powered Manuscript Assessment Tool — Publishers Weekly

Google Disables Search for Book Previews [probably related to AI scraping] — WinBuzzer (via Jane Friedman’s Bottom Line newsletter)

Bad stuff

As AI enters the operating room, reports arise of botched surgeries and misidentified body parts — Reuters

Health Advice From A.I. Chatbots Is Frequently Wrong, Study Shows. In part, the problem has to do with how users are asking their questions. — New York Times

Job market

100s face axe at law firm Baker McKenzie as AI reliance grows — RollOnFriday

More companies are pointing to AI as they lay off employees (but it may just be a pretext) — CBS News

Model & Product updates

Claude Cowork (a desktop application for the Mac) is now available on the $20 per month plans instead of being limited to the top-tier plan. Also, the business-focused plugins the company just launched were seen as so powerful that they caused a down day in the stock market.

Claude Opus 4.6 from Anthropic (much anticipated)

GPT-5.3-Codex from OpenAI (much anticipated)

Model Council from Perplexity (get results from 3 top models at once)

Video

Watch Svedka's Super Bowl Ad and You'll See a Spot Mostly in AI — The Hollywood Reporter

The business of AI

Claude Code is the Inflection Point — SemiAnalysis

Other

Moltbook is ChatGPT moment for AI agents [This is a good overview of where agents stand today and how they work. It’s about much more than Moltbook.] — Vox

How does AI impact skill formation? — Sean Goedecke

Google's Outbound Traffic Down 33% — B2B News Network

Why editors should charge more for editing AI-generated manuscripts — Josh Bernoff

How AI assistance impacts the formation of coding skills — Anthropic

Research Shows GenAI Ads Perform Just as Well as Human-Made Content — ExchangeWire.com

Firefox is adding a switch to turn AI features off — The Verge

Dutch Court Invalidates AI-Generated Wedding Vows, Marriage Not Legal — Gulf News

The Long Now of the Web: Inside the Internet Archive’s Fight Against Forgetting — Hacker Noon

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

The Claude Code frenzy continues

Mignon Fogarty — Mon, 02 Feb 2026 14:04:31 +0000

Issue 102

On today’s quest:

— The Claude Code frenzy continues
— How much more energy does Claude Code use?
— Stats galore
— Weird AI: More on AI names
— What’s happening with jobs?
— ChatGPT hedges less than science writers
— AI errors are different from human errors
— AI influencers are on the rise
— Bilbo uses ChatGPT

The Claude Code frenzy continues

People will not stop talking about Claude Code! It seems like a lot of people spent their holiday break playing with it and now want to talk about it.

These are the stories I read in just one afternoon:

AI agents: We're in for a wild ride [A short overview from a tech professor.] — Engineering Prompts
Agent Psychosis: Are We Going Insane? [A deeper analysis from a software developer.] — Armin Ronacher’s Thoughts and Writing
Anthropic’s Claude Code Has the AI World Buzzing: ‘It’s Amazing and Also Scary’ [An overview for business people with some good quotes from developers.] — Wall Street Journal
Am I too stupid to vibe code? [A description of a failed attempt to use Claude Code. At least part of the problem is that he uploaded too much data and hit the context limit, but it’s still interesting.] — Garbage Day

Plus the Bloomberg “Odd Lots” finance podcast that I regularly enjoy did a show about Claude Code.

How much more energy does Claude Code use?

Building a whole website or coding a whole game with a tool like Claude Code seems like it uses vastly more tokens than what most people would consider a typical query, so with the growing excitement about these tools, I’ve been wondering how bad they are on the energy front.

A blogger and developer named Simon P. Couch has the first write-up on this question I’ve seen. His post has caveats, but his best estimate is that as a heavy user, he’s using about as much energy each day as it would take to run a dishwasher once each day.

Stats galore

If you're looking for stats on what people know about generative AI and how they use it, this Reuters Institute report from October 2025 is a great place to start. It breaks numbers down for six different countries.

In the U.S., only 9% of respondents hadn't heard of at least one of 13 major AI systems (ChatGPT, Gemini, Perplexity, etc.), and 36% reported using generative AI weekly. The report also has numbers from 2024, and both awareness and use increased significantly between the two surveys.

Weird AI: More on AI names

AI continues to favor certain names for certain characters. I previously told you about LLMs tendency to choose Elara Voss as a protagonist name, but now Jochen Voss (the name is a coincidence) has done a bigger test and found more preferences. When prompted to create a name for different people, Claude generated the following:

80% of protagonists’ wives are named Eleanor.
34% of protagonists’ husbands are named Marcus.
99% of men living in York, UK, are named Thomas.
52% of traveling wizards are named Aldric; 20% are named Meridian.
100% of software developers are named Marcus Chen.

Click through for more interesting favorites, including last name trends!

What’s happening with jobs?

More signs are emerging that AI could be making it harder for recent graduates to find jobs, especially in tech and finance. But the data still isn’t definitive.

Further, reports find that “small businesses deploying generative AI did not cut jobs. Instead, they were better able to scale up and compete, saw their workload reduced, and became less reliant on external consultants.”

AI’s impact on jobs is set to become more pronounced — FT

ChatGPT hedges less than science writers

Linguists compared abstracts from nursing papers written by humans and ChatGPT-4o and found that humans were more likely to use hedging language such as “may suggest” and “appears to indicate.” ChatGPT-4o also used longer words and more parenthetical details. — Kimberly Pace Becker on LinkedIn

AI errors are different from human errors

Source: Eugene Vinitsky on Bluesky

Sometimes, the errors LLMs make are quite different from the errors a human would make, which means proofreading AI work can be quite different from proofreading human work.

AI influencers are on the rise

Brands are attracted to AI-generated influencers for the control they can exert over the personas’ appearance and statements. All the way back in 2024, a survey found that 60% of marketing professionals had already had experience using AI influencers.

Viewers can have a hard time pegging these videos as fake. For example, during Wimbledon, an AI influencer called Mia Zelu amassed more than 150,000 followers before it was revealed she was not a real person. On TikTok, another AI-generated woman called Tinsley racked up sympathy comments and 150,000 views after “crying” about getting rejected from her chosen sorority. — Financial Times

I’m laughing

In case you can’t read the image, it’s a post by Matthew Highton at Bluesky (@matthighton.bsky.social). The image is of Bilbo from “The Fellowship of the Ring” holding a phone. The text says:

Bilbo: After all, why not? Why shouldn’t I keep it?

ChatGPT: You’re absolutely right — you found it, it’s been with you a long while, and it’s only natural to feel fond of something that’s served you so well, especially when someone like Gandalf suddenly seems to want it for himself.

Quick Hits

My favorite pieces this week

AI agents got access to their own social network last week, and they formed a religion (among other astonishing things). The AI Daily Brief podcast (YouTube | Apple) has the best overview and anecdotes, but if you don’t have time to listen, this post by Carlo Iacono also has a good summary.

On Evaluating Cognitive Capabilities in Machines (and Other "Alien" Intelligences) [Compares research on the intelligence of AI to research on intelligence in babies and animals.]— AI: A Guide for Thinking Humans

We're All Software Developers Now — Craig Hughes

Educator Anna Mills demonstrates how easy it is for students to get ChatGPT to take quizzes for them in a browser — YouTube (3 minutes)

Management as AI superpower — One Useful Thing

Using AI

A Librarian’s Playbook for Agentic AI [A librarian used a chatbot combined with Zapier to automate most of the tasks needed to set up instructional webinars.] — New York Law Institute

Publishing

Anthropic ‘destructively’ scanned millions of books to build Claude — The Washington Post

Philosophy

Anthropic publishes a follow-up “constitution” to the previously leaked Claude soul document — Anthropic

The Adolescence of Technology — Dario Amodei (The Anthropic CEO reviews what he sees as the danger AI poses to society. Interesting tidbits from this long and interesting piece include 1) about 5% of Anthropic’s cost of generating responses goes to preventing someone from making bioweapons, and 2) Amodei doesn’t think AI is taking jobs right now.

Legal

Matthew McConaughey Trademarks Himself to Fight AI Impersonations — PCMag

Publishing

AI Tools Are Changing Academic Publishing. How Can We Adopt Them Responsibly? — Katina Magazine

Bad stuff

On the Coming Industrialisation of Exploit Generation with LLMs [A tech guy finds that top-tier LLMs can quickly break into systems.] — Sean Heelan’s Blog

More than 90% of people couldn’t tell real video from AI generated video — Runway

Google won’t stop replacing our news headlines with terrible AI — The Verge

Latest ChatGPT model uses Elon Musk’s Grokipedia as source [Remember to include “show your sources” in your prompts so you can evaluate them.] — The Guardian

Trump DOT Plans to Use Google Gemini AI to Write Regulations [DOT general counsel is quoted as saying, ““We don’t need the perfect rule on XYZ. We don’t even need a very good rule on XYZ. We want good enough. We’re flooding the zone.”] — ProPublica

Amid Backlash, Massive Porn Platform ManyVids Doubles Down on Bizarre, AI-Generated Posts [Creators who use the site believe the founder has succumbed to AI psychosis and fear losing access to a platform where they earn a living. The 404 Media podcast talks about this story more.] — 404 Media

US cyber defense chief accidentally uploaded secret government info to ChatGPT — Ars Technica

I’m laughing

Art Student Arrested for Eating AI-Generated Art in Protest — ARTnews

Science & Medicine

NHS launches trailblazing AI and robot pilot to spot lung cancer sooner alongside screening programme set to tackle cancer inequalities — NHS

Job market

AI layoffs or ‘AI-washing’? — TechCrunch

Fiverr hits historic low amid AI fears — Globes

Pinterest laying off 15% of workforce as part of AI push; stock falls — CNBC

Amazon axes 16,000 jobs as it pushes AI and efficiency — Reuters

Model & Product updates

OpenAI launches an $8-per-month plan in the US. [The plan gives you more messages, but does not give you more access to the best thinking models. Along with the free plan, this $8 plan will soon include ads below the main response.] — OpenAI

Moxie Marlinspike has a privacy-conscious alternative to ChatGPT [The maker of Signal has released a chatbot that is supposed to be equally secure.] — TechCrunch

Google begins rolling out Chrome's "Auto Browse" AI agent today — Ars Technica

Science & Medicine

AI reveals 800 never-before-seen ‘cosmic anomalies’ in old Hubble images — Scientific American

Education

AI Is Coming to Class. These Professors Want to Ease Your Worries. — New York Times

Khan Academy launches free AI Writing Coach for US schools — Khan Academy

90% Of Faculty Say AI Is Weakening Student Learning: How Higher Ed Can Reverse It — Forbes

Are AI Agents in Moltbook Conscious? We (and our Students) May Think They Are — Education Disrupted

Music

No AI Allowed: Sweden Bans AI-Generated Songs From Official Charts [The policy was triggered by a song from a virtual persona called Jacub, titled “I Know, You’re Not Mine” and which briefly topped the Swedish Spotify “most played” chart.] — Yahoo

Video

YouTube will soon let creators make Shorts with their own AI likeness — TechCrunch

The business of AI

The A in AGI stands for Ads [A detailed look at whether ads will make OpenAI profitable.] — Ossama Chaib

OpenAI sets early ChatGPT ad pricing at about $60 CPM [about 3x the average rate on Meta] — The Keyword

Other

OpenClaw (Formerly Clawdbot) Showed Me What the Future of Personal AI Assistants Looks Like — MacStories

Meta pauses teen access to AI characters ahead of new version — TechCrunch

A Wikipedia Group Made a Guide to Detect AI Writing Now a Plug-In Uses It to ‘Humanize’ Chatbots — Wired

Science fiction writers, Comic-Con say goodbye to AI — TechCrunch

YouTube Overtakes Reddit as Go-To Citation Source on AI Search — AdWeek

AI can beat average human creativity — but the most imaginative minds are still unmistakably human — Science Daily

OpenAI is building a social network and considering using biometric verification to ensure its users are people, not bots — Forbes

Most workers spend 3+ hours per week cleaning up AI workslop [but 92% still say AI has increased their productivity] — Zapier

YouTube’s top AI slop channels are disappearing — The Verge

Developers, Reinvented [An interesting piece about the phases developers go through as they learn to use AI coding tools. This quote stuck out to me: “Developers rarely mentioned ‘time saved’ as the core benefit of working in this new way with agents. They were all about increasing ambition. We believe that means that we should update how we talk about (and measure) success when using these tools, and we should expect that after the initial efficiency gains our focus will be on raising the ceiling of the work and outcomes we can accomplish.”] — Thomas Dohmke (CEO of Github)

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human

Claude Cowork is a friendlier Claude Code

Mignon Fogarty — Fri, 16 Jan 2026 20:02:00 +0000

Issue 101

On today’s quest:

— Claude Code and Claude Cowork
— AI in K-12 education: too risky
— Be careful what you give AI access to
— AI-generated YouTube channels are growing fast
— AI that can’t make bird feeders
— Bad start for AI-generated podcasts at the Washington Post
— Smart glasses company Xreal raises $100 million

Claude Code/Cowork is taking off

I showed you a demonstration of Claude Code back in October (with installation instructions), and the buzz about this tool has grown exponentially ever since, reaching new heights recently after the release of the new Opus 4.5 model and Claude Cowork, which is essentially Claude Code repackaged in a desktop app to lure in people who are intimidated by the name “Code” and the terminal interface.

Last week, I showed you how I used Claude Code to make a website, and this week, I want to show you how I quickly made an old-school computer game because I can’t get over how frickin’ easy it was.

BACKSTORY: My husband and I used to play a game in the late 1980s called Agent USA — an educational game from Scholastic that was discontinued. Every time vibecoding comes up, he wonders whether we could remake this game we loved.

So I finally decided to give it a try while I was sitting on the couch watching TV with a sick relative to keep her company. And in ~30 minutes with a few rounds of prompts, I had a working game that played in my browser with many of the core elements of Agent USA: healthy citizens, “fuzzed” citizens, crystals you farm and drop to cure the fuzz, and a train station to take you to different cities.

THE PROMPTS: This was my starting prompt:

Make me a version of the 1984 abandonware game from Scholastic called Agent USA. You play a secret agent racing against time to track down enemy agents across the United States. Look up the details.

The first version kind of worked, but had lots of minor problems, which I fixed with prompts that were like talking to someone:

Make the fuzzed citizens move randomly.
Make the healthy citizens pick up crystals and help farm them.
Add a start screen where you can choose the game speed.

Claude Code implemented all of these changes perfectly. I can’t overemphasize how quick and simple it was. I literally did this while watching TV and chatting with a relative in the background, and I imagine it would take me just another hour or two to add every game element I remember.

Getting Claude Code set up is a little tricky, but once you have it … wow.

I’m glad to see Anthropic making this more accessible to people by creating Claude Cowork. I haven’t tried it yet, but if you’re interested in these projects but are a little intimidated by having to install Claude Code and use the terminal interface, I suspect it’s well worth exploring (although it’s currently only available to $100-per-month subscribers, so maybe it’s worth getting over your fear of the terminal).

This week’s Hard Fork podcast also has examples of projects non-coders have done with Claude Code. The segment starts at ~32:00. The cutest is a man who made an app with a leaderboard to get his kids to compete against each other on how many cabinet door knobs they could each replace.

AI in K-12 education: too risky

A new report by the Brookings Institute says that although AI can make teachers’ lives a little easier and help young students learn to read and write (especially second language learners) when used as a supplement to writing, it poses a grave threat to children’s social, emotional, and cognitive development, and the benefits do not outweigh the risks.

Be careful what you give AI access to

Google’s new Antigravity agent wiped a developer’s entire hard drive, and this is the second story I’ve heard like this in the last 6 months or so. On the one hand, it’s probably not that common since I haven’t heard a lot of stories; on the other hand, it’s catastrophic when it happens.

A rule I use for myself is to never give AI access to something that isn’t backed up. Being unwilling to give AI access to my private accounts has kept me from trying things I’m interested in trying though — so much so that I’ve dabbled with creating a separate Google account just for AI projects.

AI-generated YouTube channels are growing fast

Source: Garbage Day

This graphic showing that AI-driven YouTube channels are some of the fastest growing on the platform comes from a Garbage Day newsletter back in November. I feel like there’s not a lot to say except “Wow!” — and subscribe to Garbage Day. It’s one of my favorites. I’d probably miss half the internet trends without it.

AI that can’t make bird feeders

I heard an offhand comment on the Shell Game podcast about how some AI models have blind spots; for example, apparently, there’s one that can’t make a picture of a bird feeder.

I wanted to learn more about that, but Google search was worthless. I could only get pages about AI-enabled bird feeders, which can take cute pictures, but are not what I wanted. But when I put my search into ChatGPT, I immediately got a link to the research paper describing the “bird feeder” problem. And for those of you who haven’t tried it yet, that’s why people like AI search.

BTW, I’m enjoying the Shell Game podcast by Evan Ratliff who’s trying to start a company using nothing but AI agents.

Bad start for AI-generated podcasts at the Washington Post

In mid-December, the Washington Post launched “personalized podcasts,” AI-generated audio customized for each subscriber’s interests. Despite immediate backlash about problems in the audio — both technical and factual — the Post said it was committed to continuing the beta program, and the outlet is still producing the podcasts with a disclaimer that they are AI-generated and may contain errors. (Grant Crowell on LinkedIn calls them “oddcasts.”)

Smart glasses company Xreal raises $100 million

Please indulge me for a minute, because this wouldn’t normally be a story I’d cover, but I LOVE my Xreal glasses.

I’ve used these glasses for more than a year as my primary computer monitor, which gives me three huge virtual screens. I have a slight vision problem (detached vitreous), and for whatever reason, I can see better through these glasses than by looking directly at a monitor.

The glasses don’t currently have “smart/AI” features, but they’ve formed a partnership with Google to develop smart glasses to compete with Meta’s smart Ray-Bans. I’m happy Xreal seems to be getting the attention I think it deserves and the opportunity for growth!

Quick Hits

My favorite reads this week

“To me, it’s an access question.” [About the use of AI in college admissions] — AI + Education = Simplified

Using AI

I didn’t expect Gemini to make Google Keep this much more useful — Make Use Of

New Year, Small Steps with AI for Editors — Marcella Fecteau Weiner

AI Writes Brilliantly. So Why Do I Still Feel Like I’m Doing It Wrong? — Mark Randolph’s Substack

Cooking with Claude — Simon Willison’s blog

Use multiple models — Interconnects

Philosophy

AI and the Human Condition — Stratechery

The Boredom Experiments [An AI agent does experiments to try to learn what keeps it functioning] — Strix Research

Don't fall into the anti-AI hype — Antirez

Publishing

New Kindle Recaps feature provides story refreshers for eBook series — Amazon

Author Shaun Rein's voice 'cloned with AI' to read book on 'deepfake' podcast — The Bookseller

Robotics

See the robot that will be working at Hyundais factories [Atlas robot video from CES] — YouTube

Science & Medicine

Artificial intelligence begins prescribing medications in Utah — Politico

New ChatGPT Health tab allows you to connect fitness apps — Axios

An AI revolution in drugmaking is under way — The Economist

Education

Writing Labs Are an Answer to AI — Inside Higher Ed

Other

AI isn’t “just predicting the next word” anymore — Clear-Eyed AI

Claude kept a plant alive for 42 days (and counting) — Instagram

Holiday shoppers arriving on e-commerce sites from AI converted to buyers 9x more than those arriving from social media — Salesforce

Why AI Boosts Creativity for Some Employees but Not Others — Harvard Business Review

OpenAI Is Asking Contractors to Upload Work From Past Jobs to Evaluate the Performance of AI Agents — Wired

Here comes the advertising in AI chatbots — Washington Post

Share the newsletter

What is AI Sidequest?

Are you interested in the intersection of AI with language, writing, and culture? With maybe a little consumer business thrown in? Then you’re in the right place!

If you loved the newsletter, share your favorite part on social media and tag me so I can engage! [LinkedIn — Facebook — Mastodon]

Written by a human