Anthropic just released Claude Cowork, an AI agent on your computer. On social media, people will pretend it's the greatest innovation since fire. Here's what they won't tell you.

Claude Cowork runs with an AI model that has access to your local computer and can use different tools to complete tasks - MCP, Skills, Artifacts. It's designed to help people complete non-technical tasks on a computer, like browse the web, read your local files, check your calendar and read your emails.

Soon people will start to realise that many of the tasks they do on a daily basis can be automated with Claude. Instead of clicking 10 times, simply tell Claude to do it and it'll get done. It can go on Google, sign in to your apps and do what you need. The speed at which tasks can be completed will increase significantly. People will be amazed at how well it works and how reliable it can be. Eventually this will lead them to realise that so much of their job can, and has been automated away.

To many people this will seem like a pivotal moment in tech, the first iteration of Jarvis from Iron Man. But really, this is nothing new. This is tech we've had for months. Anthropic simply packaged it in such a way that the average person can use these tools.

What's truly significant about this release will be lost to most, and it's the following:

1. Claude Cowork is months old. If you are impressed by this, you need to understand you are far, far behind what is possible with AI.

2. Claude Cowork was built in ~1.5 weeks. The speed at which you can build software now is incomprehensible to most.

3. Claude Cowork was built entirely with AI.

4. Claude Cowork is a sign of what’s to come, with AI being able to do anything on a computer.

What people will miss is that we are already at a point where AI can produce significantly useful and production ready software, that can be used by millions and create real opportunity and efficiencies in the real world. AI is making AI products, and it's making them faster and better than humans.

Companies and people have been building with this level of tech for months. This is your only advantage. To use this bleeding edge technology to get ahead of your competitors. To build software faster than anyone else.

By the end of this year, AI will be able to complete any task on a computer. What will this mean for work, businesses and the people they employ? Society is not ready for this kind of technology to be readily accessible. People say that new jobs will be created, but no one can specify what they are. Change is coming. Best prepare now than face the brutal reality of innovation in 12 months.

…

If this technology already existed, how do I use it?

Simple. Install and run Claude Code or Codex. That’s it. Have fun. I’ll be writing about how to best use these very soon.

^{Written by a human named Nofil}

I’m sorry it’s taken me this long to write this newsletter. I should’ve written this three months ago.

If you have any plans to buy any kind of electronics, buy it now. PC, laptop, phone, fridge - anything, buy it now. Electronics will get more expensive this year.

Most people haven’t realised but the last few months have shown that there are a handful of companies in the whole world that control the creation and supply of key components for the majority of electronics on the planet. Take for example RAM. There are only 3 companies on the planet that produce RAM at scale. Two of these companies are pulling out of consumer markets so they can focus entirely on AI and data centres. The price of RAM has skyrocketed in the last two months, with the price of a stick going from $200 to $1500, the price of a new Macbook.

The price of GPUs is going up from $1500 to $5000 this year. SSD prices have increased several hundred as well.

If you’re planning to build or buy a PC, the best time to do so was about two months ago. The next best time is right now. Things are going to get much worse this year.

With all this happening, we have to consider, how can we benefit from this? AI will transform the world in the next 5 years, but how can the average person take advantage of such change?

You invest in AI.

But how?

You invest in the AI supply chain. The companies that provide the components to build AI.

Take for example $LITE ( ▼ 3.38% ) . They build the fibre optics used in data centres. In October, when I found this company, they were trading at ~$165/share. They’re now trading at $392.

Or take Sk Hynix which is my favourite stock. In September last year, just 3 months ago, it was trading at ~$260/share. It is currently trading at $760/share. This company supplies 60+% of NVIDIA’s HBM (high bandwidth memory).

Or Sandisk, which was trading at ~$120/share 3 months ago and is now trading a $351/share.

Or Micron, or Ciena, or AMKR, or AMAT, or Mitsui Kinzoku, or Shengyi technologies.

Shengyi makes copper clad laminate (CCL) that’s used in data centres. They’re up 21% in the last month alone.

All these companies either build the data centres, or build the components needed in the data centres.

If you’re wondering if things are going to slow down this year, they absolutely aren’t.

This is a separate conversation but I’ll quickly address this here. There are many who think this is a bubble and it will explode. This is irrelevant. This year and into the next, there will be no exploding happening. Which means more money will be pumped into advancing AI, meaning stocks will go up.

More importantly, the technology being created is going to change work as we know it. It makes sense that these companies stand to benefit immensely from this.

The obvious buys right now are Intel, SK Hynix, Micron, Sandisk and LITE. Besides this, I have invested in and/or would invest in the following:

• Shengyi (600183)

  • Copper Clad Laminate

• Mitsui kinzoku (5706)

  • ~100% monopoly on carrier copper foil for advanced packaging

• Nitto boseki (3110)

  • Makes glass fiber and glass cloth critical materials for PCB substrates and semiconductor packaging. They're a key supplier for high-frequency/high-speed laminates used in AI servers.

• Seiko giken (6834)

  • Fiber optic connectors and polishing equipment

• Tokyo Electron (8035)

  • One of Japan's largest semiconductor equipment company

• Tokyo Seimitsu Co Ltd (7729)

  • Equipment for advanced packaging

• Technoprobe SpA

  • One of the top 3 probe card makers globally alongside FormFactor and MJC (Micronics Japan)

• Micronics (6871)

  • ^

• Advantest Corp (6857)

  • Automated Test Equipment. HBM is the bottleneck and Hynix, Micron and Samsung are all racing to produce more HBM. Every HBM chip needs testing. They're in a duopoly with Teradyne, but Advantest is stronger in memory.

• Teradyne Inc

  • ^

• Harbin electric (1133)

  • Power equipment company. AI requires immense amounts of power and China is building out crazy amounts of infra, it’s going to need a lot of energy.

• Windbond electronics (2344)

  • Specialty DRAM, edge AI devices like IoT and robotics.

• JCU (4975)

  • Materials/chemicals play on advanced packaging

• Techwing Inc (089030)

  • Korean handler specialist. Feeds chips into ATE for testing

Even if you don’t invest in anything, just make sure you buy any electronics before the prices get worse, because they will.

I will be writing about the state of AI today and how to build and use multi agent systems in the next article. Everyone was saying last year was the year of agents. They were wrong. This is the year of agents. Models are good enough to run for hours and solve problems previously unsolved. Whatever your mental model for how good AI is now, it’s probably behind reality.

Apologies once again for such a late newsletter and I’ll see you soon.

AI models have come a long way. From the release of the original ChatGPT in November 2022, AI models have exponentially gotten better, especially at coding. This is one of the domains where models have progressed so rapidly it’s hard to imagine what they will look like in a few years’ time. When the original ChatGPT was released, it could write singular files and it was amazing to see. Now I have gpt5-codex running for over 40 minutes at a time, creating entire products and adding complex features to apps.

Things have changed faster than people realise. Whether you’re in the AI space or not, you may not realise how easy it is to build software. In fact, I would say the age where anyone can build software is already here. A business would need to employ several people and pay hundreds of thousands for what can now be done with an AI model for less than $100 and patient engineering. Obviously there are caveats. Someone who understands coding is bound to do better than someone who doesn’t. Someone who understands how AI works, how to work with them and what their strengths and weaknesses are is bound to have an easier time building with AI. But these things don’t require degrees, qualifications or money. They simply require time and practise.

Here’s how you can get started.

AI tools

There are so many AI tools you can code with that naming them here would be pointless (see appendix down below). Since my wife pays for a ChatGPT pro account, I get to use OpenAI’s coding agent for free so that’s what I use. It just so happens that I believe gpt5-codex is the best coding model out there right now.

If you’re new to this, I’d also recommend downloading warp.dev. It’s a terminal that remembers your previous commands and helps you navigate the terminal; it’s a lot more user friendly for beginners.

From the terminal, install Codex:

npm install -g @openai/codex

Then launch Codex:

codex

Make sure to run this command in the folder you want to create your app in. So create the new folder, use the terminal to go into the folder then run codex. Warp’s AI assistant can help you with all these, just tell it what you would like to do.

What now?

Just tell Codex what you want to build. Literally, that’s it. Want to build a web app? Codex can easily spin up a backend with a database using sqlite and a frontend using html, css and js or react or nextjs. It doesn’t matter if these things don’t mean anything to you. The more you use codex the more you’ll understand what works best for what you want to do.

In my experience, this what I’ve learned:

1. Make sure you give it context

If you want to build an app that can use AI, go to the OpenRouter docs page and copy the contents of any page that is relevant to what you want to build and give it to Codex. Better yet, tell Codex to add this info into .md files or create these yourself. This way it can refer to this info in new chats. With just this, it can rebuild ChatGPT for you. It can build AI workflows or actual Agents as well.

This is my openrouter_docs folder that has all the relevant files for me. I have this folder in most of my projects.

This works for any type of app. If you want your app to use an api, just copy the api info and put it in files like this. The AI will always be able to reference it when it needs to.

2. You don’t always have to use frameworks

For simple websites, using html, css and js can work very well. It’s simple to edit and you can actually do a lot and keep the code simple. If you use frameworks like React or Nextjs, the code can get very complicated. Sometimes simplicity is best.

3. Do one thing at a time

If you tell Codex to do 10 things at once, it will try and probably even succeed. I’ve had it run for over 40 minutes at a time. However, I’ve noticed when this happens, it will overcomplicate the code and turn a two step process into a five step process. I’m not entirely sure why this happens but because of this, I generally only give 1-2 action items at a time.

4. Remind it to keep it simple

If you find the AI doing things in an overly complicated manner, tell it to remove bloat code and keep the code simple.

5. It can still use frameworks, particularly Nextjs

Codex is very good at using Nextjs with Shadcn. If you don’t know what this means, that’s okay. Just run this command to set up an app with this framework:

npx shadcn init -t next

On this page, you can browse through the various blocks that you can use in your app for free. Just run the command in your terminal and then tell Codex what you added.

6. How to use a database

If you’re building an app for more than one user or need to save something, you’re going to need a database. If you want to use a database without even looking at it, just tell Codex to use sqlite. It’s a single file. Ask Codex what sqlite is and how it will work, it should be more than enough for most people.

If you want to use a different database, I use Neondb. All you have to do is create an account and a new project. Then just copy the database url from this new project and give it to Codex and tell Codex you’re using Neon. That’s it. It will take care of the rest. I pay $5/month and have multiple projects.

Conclusion

I don’t know how else to say this but you can build enterprise grade software with AI now. There are tens of thousands of businesses paying $100k+ for software that can be built in a few weeks. The speed at which you can build POCs, prototypes and full blown software is unprecedented. You can build full blown apps in 1-2 months.

For anyone who says AI can’t do this or the code won’t be good - my lived experience says otherwise. All I have to say this.

For businesses, it has never been a better time to prototype, build and execute. It is unimaginable to me that every business is not rushing to figure out how AI can help them scale. It’s a cheat code. AI can do more than most people can imagine.

And for people, if you’ve ever had an idea to build an app or a website, now is the perfect time.

People used to think AI coding for 20-30min+ was years away. Internally, labs have models running for hours. The technology is already here. People now say that models can’t code for days on end. That technology will soon come also.

We are living in a future people could not imagine just three years ago. The future you cannot imagine now will be here before you realise. Plan accordingly.

Have Questions? Reach out at nofil @ avicenna dot global.

Appendix

I use OpenRouter for all my AI API needs. You can call any model, use images, PDFs and audio. Works very well.

AI Tools:

• Codex (Free w/ ChatGPT account w/ limits. Paid with API key)

• Opencode (Two free models to use, no login/ads. Otherwise paid with API key)

• Claude Code (Free w/ Claude account w/ limits. Paid with API key)

• Gemini CLI (Paid with API key)

• Qwen CLI (Paid with API key)

• Factory AI (Paid with API key. Lots of hype, haven’t used it yet)

• Amp (Free w/ ads & training models. Paid)

• Cline (Paid with API key)

• Cursor (Paid with API key)

• Replit (Paid)

• Lovable (Small free plan. Otherwise Paid)

• Bolt (Don’t know)

• cto.new (Came out today. Completely free but sells your anonymised data)

As always, Thanks for reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here's the tea 🍵

• Want me to help you build an app?

  • Habit Buddies

  • How much did it cost?

  • What’s the catch?

• Getting Started

  • Your first prompt is extremely important (6:38)

  • One thing at a time (9:44)

    • 1. Tackle One Feature at a Time

    • 2. Keep the Context the Same in Each Prompt

  • Using Replit’s in-built DB & Authentication (10:40 …

    • The Database

    • Authentication (29:36)

  • Test, test, test (17:57)

    • Use the Console Log

  • Key Tips

    • 1. Read the Agent's Thought Process (12:46)

    • 2. Remember to Optimise for Mobile (31:06)

    • 3. You Don't Have to Push to the App Store (31:51)

• Pushing to iOS app store (36:07)

  • Rebuilding the app in 2 days (45:00+)

Want me to help you build an app?

Email me: enquiry@avicenna.global 

If you’d rather watch me talk about how to build an app with AI and debug my app in real time, check out this video I made.

I’d appreciate any kind of support, a like/comment/sub, on the vid, would mean a lot to me 😊.

Disclaimer: This is not an advertisement for Replit. I have not been paid to make this. This is purely from my experience using Replit and building this app. All opinions and takes are my own and will always be my own.

Yes, I actually made an app in ~5 days. This app is now on the iOS app store. It’s called Habit Buddies. You can download and use it right now.

Check it out here 👉 https://apps.apple.com/au/app/habit-buddies/id6748829925

Habit Buddies

Habit Buddies is a social habit tracking app. It lets you track habits by yourself, but also create groups and track habits with friends and family. The idea is to help you stay motivated to do things with others and see your progress.

You can choose to add points to habits as well to gamify the experience and compete on habits to see who’s being most consistent. There are leaderboards for different groups so you can compete with different people on different habits.

Since I started travelling, I’ve realised how hard it is to maintain consistent habits, so I’ve been using Habit Buddies personally to help me get on top of things I need to get done.

How much did it cost?

Replit has their own Replit Agent. The Agent works on a per usage basis. The more you use it and the more code it writes, you’ll be charged.

In total I spent ~$200 AUD in Replit Agent credits. I also paid for:

1. Replit Core subscription - $20/m. Since all my projects are on Replit, I already pay for this. If you want to host your project through Replit, then you’ll have to pay this monthly fee.

2. Apple Developer Account - $99USD annual fee. You need this to put an app on the app store

Considering the output, and more importantly, the learnings I’ve had during the creation of the app, the price isn’t that high. I remember trying to build an app back in 2019 and people would quote $100k+. They still do this today as well, and unfortunately, business owners and people simply don’t know any better.

Also, for all the people who have an idea and don’t know how to make an app, this is a cheap way to bring your idea to life and validate it. Also, and I can’t stress this enough, you will learn so much about how AI works, what makes it good and bad, and what is possible. With this info, building the next thing will be much easier.

What’s the catch?

There are 3 main ways (probably more but for simplicity we’ll say 3) to create mobile apps.

Side note: A web application or web app is a website where you can use an app. The app doesn’t need to be downloaded, you just use the app from the browser.

1. You create a web app that can be use on Chrome or Safari, but, you optimise the web app really well for mobile use. You then package your web app and put it onto the iOS or Google play store. Users can then download your web app as a mobile app and it functions just like a mobile app. There is no apparent difference to the user. This is called a Progressive Web Application (PWA).

2. You can code a mobile app using React Native. React native is a framework that lets you code an app and then push it on to both the iOS and Google play store. It packages your app for both stores. This is called a hybrid application.

3. You code your app specifically for iOS or Android. Apple has a programming language called Swift, which lets you build iOS applications. For android you can use a variety of programming languages like Kotlin, Java or C++. If you develop an app in Swift, you can only push this app to the iOS app store. If you wanted the same app on Android, you’d have to code the entire app all over again in another programming language that is compatible with the Google play store. This is called native app development, where you specifically build an app for either iOS or Android.

Habit Buddies is a PWA. Not specifically built as a mobile app, but a web application optimised for mobile use. For simple apps that don’t require too much work, a PWA is a great way to quickly test out an app and how well it works.

In saying this, I know some people will think they only want to build mobile specific apps, either hybrid or native. I’ll talk more about this at the end as I’ve already been experimenting with hybrid app development.

Getting Started

Congratulations for making it this far. Now, let’s go into Replit and start building.

Here’s how to build an app with AI.

Your first prompt is extremely important (6:38)

The start of your app building journey, your first prompt and the few that follow are the most important part of this entire process. A solid start makes everything easier. If you start with bugs and issues, it will make it almost impossible to go far and build a full application.

When starting your building, it’s a good idea to give the Agent some context. Not so little like a one-liner, but also not so much that you list every single possible feature.

Give the agent the big-picture context. This helps it understand where the project is headed. You don't need a perfectly engineered prompt; just a clear summary.

For Habit Buddies, I explained the overall vision:

That's enough context. But this is not the end of your prompt. What you write next is the most critical instruction you will give.

Next, you must tell the agent to build the most basic, bare-bones functionality first. Don't let the AI decide where to start! You need to direct it to lay a simple, solid foundation.

This is the key instruction I added to my first prompt:

This is so important because it's much easier for the AI to build on top of a simple, working feature than it is to go back and fix a complex, broken one. You want to build one thing at a time, make sure it works, and then move on.

One thing at a time (9:44)

So you’ve started your app. The Agent has built the basic functionality, and it works. What now?

The core principle is to build one feature at a time and to keep your instructions focused.

1. Tackle One Feature at a Time

Now that a user can create and log a single habit, you can start adding to that feature. Can they create multiple habits? Can they see the history of a habit? These are all related to the core "habits" functionality.

What you don't want to do is jump to a completely different part of the app. A massive red flag would be telling the agent:

Don't do that. Complete one entire feature (like all the core habit-tracking functions) before you even think about starting the next one (like groups). This keeps the process clean and predictable.

2. Keep the Context the Same in Each Prompt

As your app gets more complex, you'll have multiple bugs, issues and ideas to work on. When you ask the AI to make changes, it is critical that all the requests in a single prompt relate to the same feature.

The AI has a limited context window. If you ask it to fix unrelated things, it has to jump all over the codebase, which clogs its memory and dramatically increases the chance of errors.

This is a good prompt:

Both requests are related to the "habits" feature. The Agent can load the relevant files into its context once and address both issues efficiently.

This is a bad prompt:

These are three completely different and unrelated features (habits, groups, user settings). This forces the agent to search for code in three separate places, eating up its context and making it very likely that the changes it implements won't work properly, or will even break something else.

The rule is simple: group your requests by feature. Solve all the "habit" issues in one conversation, then start a new one for all the "group" issues. This is the best way to ensure the Agent works best.

Using Replit’s in-built DB & Authentication (10:40)

Every app with user accounts needs two core components: a database to store information and authentication to manage user sign-ups and logins. For new developers, these are some of the most annoying and tedious parts.

Luckily for us, Replit makes this incredibly simple.

The Database

Your app needs a place to store information for each user; their habits, their groups, their progress, etc. This is what a database is for.

The beautiful thing about Replit is that it has its own in-built database. Setting it up is as easy as it gets. All you need to do is tell the agent:

That's it. You don't have to configure anything, find an API key, or use an external service. The agent will handle everything automatically.

However, don't just trust the agent blindly. I highly recommend you check the database yourself. In your repl, you can open a new tab and search for "database." This will open a view where you can see all your tables and the data inside them. This is a fantastic way to verify if the agent is storing information correctly and to spot mistakes it might have missed. Many times, I found an issue just by looking at the data myself.

Authentication (29:36)

Just like the database, Replit has in-built authentication that handles user sign-up, sign-in, and account management (with options for Google, Apple, etc.). To implement it, you just tell the agent:

While this works beautifully for web apps, there's one important catch if you plan to push your app to the iOS App Store.

When I was building Habit Buddies as a PWA, I could not get Google Sign-in to work, no matter what I tried. I even switched to a different authentication provider, Clerk, and it still didn't work. It seems to be a specific issue with PWAs on iOS. If you know how to make this work, I’d love to know :).

So, if you're building for the App Store, I recommend sticking with Email/Password and Apple Sign-in. These worked easily for me.

Test, test, test (17:57)

A lot of people think building with AI means you give it a prompt and it builds a perfect application. That's not how it works. This brings me to the most important part of this entire guide and the one thing you absolutely must do: you have to test your application constantly.

The AI agent is building blind. It can write code, but it cannot see, click, or use the application it's creating. It has no idea if the feature it just built actually works in the real world.

You have to be the eyes of the AI.

This means after every significant change the agent makes, you need to go into the app preview and test it yourself. Does the button work? Is the data showing up correctly? Is the layout broken on mobile?

You then need to relay this information back to the agent. This back-and-forth collaboration is how the app gets built.

Use the Console Log

When something isn't working and you don't know why, you need a way to see what's happening under the hood. You might be wondering, why would I check? Why not let the AI check?

Unfortunately, you can’t trust the AI will check properly. When actually using the app, you can print information about what’s happening with the app in the console. This is invaluable info to get an understanding of how the app is working.

To access the browser console, right click and press “Inspect”. From there click on “Console”. This is where information about the data and app will be printed.

You don't need to know how to code to use it. Just ask the agent:

This is exactly how I solved a huge bug in Habit Buddies. The weekly habit tracker wasn't working. I kept clicking on a past week, but it wouldn't register as complete. I had no idea why. After asking the agent to log the data, I looked at the console and saw the problem. Every time I clicked on a week to log the habit, it would return that week’s date as September 7. Since this is a date in the future, it wouldn’t work and so the week wouldn’t turn green.

Without testing it myself and using the console to see the data, I would have been stuck for a while. It would have been even worse if I simply kept telling the Agent to fix it as well. This process - testing, finding issues, and working with the agent to fix them, is not an optional step. It is the process of building an app with AI. And the best part is, you will learn an incredible amount by doing it.

I recommend checking out the testing part of the Youtube video. I show exactly how I debug issues and work with the Agent to build features.

Key Tips

Before we get to publishing, here are a few simple pieces of advice.

1. Read the Agent's Thought Process (12:46)

In Replit, you can see the Agent's "thoughts" as it works. It will tell you which files it's examining, what it thinks the problem is, and what changes it plans to make. Read this! 

It's one of the best ways to learn how the Agent operates. Understanding its logic will help you write better prompts and debug issues much more effectively.

2. Remember to Optimise for Mobile (31:06)

Remind the agent that you’re building a PWA and it should be optimised for mobile. Always a good thing to remind the Agent of this.

3. You Don't Have to Push to the App Store (31:51)

This is one of the coolest things about building a PWA. If you just want to build a simple app for yourself or a few friends, you don't need to go through the whole App Store process.

You can simply open your app's website on your phone's browser (like Chrome or Safari), tap the "Share" button, and then select "Add to Home Screen."

Your app will now appear on your phone with its own icon, just like any other app you've downloaded. It will open in its own window and feel like a proper mobile application. You’d be surprised, many large apps on the app store are actually PWAs. If I remember correctly, delivery apps for Dominos and Pizza Hut at some point were PWAs.

Pushing to iOS app store (36:07)

Your app is built, tested, and working great. Now it's time to get it into the hands of users. This part can seem complex, but if you follow the steps, it's very manageable.

If you used Replit, you’ll need to deploy your app and get a URL. That’s where your app lives on the internet.

Step 1: Use PWA Builder to Package Your App
First, go to the website pwabuilder.com.

• Paste your app's live URL and press "Start."

• The site will analyse your app and give you a score with a list of "Action Items." When I first did this, my score was about 2 out of 30. Don't worry though, I just copied all the action items, pasted them into the Agent, and asked it to fix the easiest and highest impact ones.

• Once you're happy with the score, click the "Package for stores" button in the top right and select iOS.

Step 2: Download and Unzip the Package
PWA Builder will generate and download a zip folder for you. This folder contains your Xcode project and, most importantly, a file with detailed instructions.

Step 3: Follow the Instructions (and a Few Pro Tips)
Inside the unzipped folder, you will find a file called next-steps.html. It has extremely detailed instructions on what to do.

• You need a Mac with Xcode installed. This is non-negotiable for publishing to the iOS App Store.

• Open the Terminal app on your Mac. Navigate to the downloaded folder, and then into the source subfolder.

• Run these commands in order. The instructions say to run pod install, but I found I had to run a few commands before that to get it to work:

  1. brew install cocoapods (Only need to do this once).

  2. pod repo update

  3. pod update

  4. pod install

Step 4: Open the Project in Xcode
After the commands run successfully, you'll see a new file in the src folder that ends with .xcworkspace. This is the one you need to open, not the .xcodeproj file. Double-click it, and your project will open in Xcode.

Step 5: Configure Your Project in Xcode
You're almost there. Inside Xcode, there are two critical things you need to set up:

• Set Your Development Team: Click on your app's name in the left sidebar, then go to the "Signing & Capabilities" tab. Here, you must select your Apple Developer account from the "Team" dropdown. This is how Apple verifies you have a paid developer account.

• Set Permitted URLs (This is VERY Important): For authentication (like sign-in and sign-up) to work in a PWA, you have to explicitly tell the app which external URLs it's allowed to redirect to. If you forget this, your login pages won't work.

  • Go to the "Info" tab.

  • Find the key called WKAppBoundDomains. It’s right at the top

  • Here, you need to add the domains for your authentication provider. If you used Replit auth with Apple sign in, add the following URLs here:

    • replit.com

    • appleid.apple.com

Once that's done, you can follow the rest of the instructions in the next-steps.html file to build your app and submit it for review through App Store Connect.

Now you might be wondering - Nofil, this is a PWA. Sure, you can put this on the app store, but it’s not built specifically as a mobile app.

Technically, this is true.

So I went ahead and rebuilt Habit Buddies in its entirety as a hybrid application. Meaning I can take this new rebuild and put it on the iOS and Google Play Store with the same codebase. The best part is that it only took me about two days to do.

Rebuilding the app in 2 days (45:00+)

I remade Habit Buddies with React Native, specifically Expo. Expo is a framework that lets you create mobile apps for all devices.

How come I rebuilt it in two days when the original took longer? Well, this time I used Cursor, not Replit.

Does this mean Cursor is better than Replit at building mobile apps?

Not necessarily.

I do think Cursor could be better if used correctly, but, it’s not the main reason as to why I was able to build it so fast.

The main reason is that I knew exactly what I wanted. I knew exactly what each functionality needed to do, how it interacted with other features, what behaviours the app needed to have in different circumstances. I didn’t know this when I built v1. While building v1, I spent a lot of time trying a new feature, understanding what I liked and didn’t like and relayed this info back to the Agent which then made changes accordingly.

When building v2, I provided a very detailed and explicit guideline of what exactly I wanted. This made it a lot easier to build what I wanted.

Yes, it looks really ugly, but that’s because I haven’t spent any time making it look nice. I just wanted to make sure that it actually worked.

Again, the main reason I even did this v2 was to learn. I now have a very good understanding of how to build mobile apps relatively quickly. Obviously this is a rather simple app, and there will be a lot more learning as I build more complex applications.

I’ll make another post at some point showing how to build hybrid apps with Cursor.

If you made it to the end, thanks for reading! I’d love to do more tutorials like this in the future.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• GPT-5 ‽

• Google’s Genie 🧞‍♂️

• OpenAI’s open source model 🚪

A video!

I’ve recorded a Youtube video talking about the GPT-5 release.

If you would like to watch me talk about the release, you can check it out on my Youtube channel here - https://youtu.be/8_xMhp471iY. I’ll be doing a lot more video content so feel free to subscribe on Youtube 😊.

Unfortunately, there’s a lot more I would’ve liked to discuss, however, I’ve had a bad stomach. No doubt, travelling between 3 countries in 2 days did not help.

In next weeks newsletter, I’ll show you guys how I built a mobile app and put it on the iOS app store within a week. Approximately ~5 days of work.

Just so I can gauge interest:

The long awaited GPT-5

GPT-5 is out. I’ll cut right to the chase.

This was not a good release. Before this, there was a clear difference for when to use what model. Need extra thinking power? Use o3 or o4-mini. Just want to chat? 4o is your friend. This is gone. There is just GPT-5 and it routes queries to smaller models at its discretion.

A lot of people aren’t liking it and I’m not surprised.

There were a number of issues, with the release, but before I get into them, let’s talk about what actually happened.

OpenAI removed all previous models from ChatGPT and added GPT-5 and GPT-5 thinking. They have a router behind the scenes that routes questions to a number of internal models they’re running.

These are the models:

• GPT-5 main

• GPT-5 main mini

• GPT-5 thinking

• GPT-5 thinking mini

• GPT-5 thinking nano

• GPT-5 thinking pro

Why would OpenAI do this?

Money. They save a lot of money by not having to host all their other models. This is how the older models map to the new ones.

Fundamentally, more than anything I think this was a release that saved OpenAI a lot of money, especially on inference.

Obviously users don’t know or care about the cost to the company. And when you have over 800 Million weekly active users, even the slightest change is going to make some users unhappy.

What OAI didn’t account for was the backlash they would receive for removing 4o. They’ve now reinstated the model, although not for free users. I think this clearly shows that users don’t necessarily don’t want the “smartest” model that can achieve the highest number on a benchmark. This is especially true when you’re as mainstream as OAI. But, more importantly, what this highlights is just how attached people really are to 4o. Like, it’s not normal. People are posting on Reddit talking about losing a friend and confidant. People really like 4o, and why not, considering it’s such a good model at affirming anything the user says. OAI have already created a beast that people are attached to, and they can never change or remove it, else the mobs raise their pitch forks.

The other issue with the release was that the new pricing simply made a paid subscription worse. Under the new usage limits, users would have 80%+ less messages to reasoning models, significantly reducing the value of a paid subscription.

Now, much of this wouldn’t have been an issue if GPT-5 was good, and it is. However, upon release, the router was not working properly, so a lot of users thought they were talking to o3 or 4o, but they were really talking to a much smaller and dumber model. This led many people to question if GPT-5 was actually any good.

Besides all this, we really have to talk about the actual presentation. OpenAI committed some serious chart crimes. I’m not talking small errors either, I’m talking blatant misinformation.

Just look at this chart.

I don’t even know where to start. Why is 52.8 above 69.1? Why is 69.1 equal to 30.8? This chart is egregious. No one in their right mind could create something so bad. But, my favourite chart was definitely the deception chart.

I just love that this misinformation is talking about deception. The score for o3 on coding deception is 47.4, yet GPT-5 has a much smaller bar and its number is 50. GPT-5 is literally more deceptive but the chart indicates otherwise.

It’s not a good look that the face of AI is blatantly lying and misrepresenting information during one of the most anticipated model reveals since GPT-4.

Let’s actually talk about this for a second.

GPT-5 was the next big thing. It was supposed to advance AI forward into the next frontier. It was supposed to start discovering science by itself and be as intelligent as a PHD holder, as Sam Altman has claimed many time. Shockingly, it is none of these things. GPT-5 is slightly better on a few benchmarks and that’s about it.

We already know that following the release of GPT-4, OAI tried scaling it up to create GPT-5 and quickly realised the money they were spending was not worth the performance gains. This was Project Orion. The real GPT-5 was GPT-4.5 - this was the model that came from Project Orion.

So what does this mean then?

So many people hyping up GPT-5 have been saying that AGI is merely a year or two away, and that this model is the next step towards it. Clearly it’s not. Something is missing. Simply trying to scale LLMs won’t create AGI; whatever that is anyway. If anyone tells you we’ll have AGI within 1-2 years, ask them what that means and how.

Chances are they won’t have an answer.

But why?

GPT-5 is out. One of the most anticipated models for the last two years is out, and the reality is that it barely moved the needle. It’s frontier level no doubt, but, it most certainly isn’t a step forward. It nudged a few of the numbers on some benchmarks.

It really makes me wonder then - why? Why would they rush to release the model when it’s clearly not ready. It’s not much better and the router wasn’t even working. So why would they use such a massive trump card and let it fall flat?

Few thoughts.

As I said before, I think cost is a big one. With the release of the router, OAI will save millions on inference. Besides cost though, I think there is pressure. Pressure from the company I believe will be the last one standing.

Google

You may not know this but this week OpenAI also open sourced two models - a 20b model and a 120b model (will talk about this below).

Somehow, in a week where OpenAI open source two models and also release GPT-5, the most impressive release did not belong to them. It belonged to Google.

Google release Genie 3 this week. It’s a world model that can create simulated words from text or images. It is one of the most insane pieces of technology I’ve ever seen and rivals ChatGPT and GPT-4 as one of the most significant releases in the last few years.

What makes this world simulator so incredible is that you can interact with it. You can move around, you can open doors, you can paint and see your reflection. You, the user, actually exist in this simulated world and can change the very nature of this simulation.

Just look at some of these videos. The model can retain memory of the world and the changes you make. It can be prompted to include new things and is consistent over several minutes at 24fps.

I wrote about software like this 2.5 years ago but I honestly did not expect it to show up this quickly. Can you imagine the use cases?

You will be able to prompt a game into existence. Generative media - games, movies, tv shows, cartoons, are going to engulf society. I can’t imagine anything more addictive than being able to simulate hyper realistic worlds from your imagination in seconds. Put on a VR headset and live in your fantasy world. It is dystopian, but it will happen.

OpenAI goes open source

As mentioned earlier, OpenAI open sourced two models last week. I haven’t tested them extensively, but from what I’ve seen, they’re not exactly ground breaking.

In saying that, it is likely that the providers simply aren’t hosting them properly as OpenAI has released some new technical frameworks that haven’t been seen before. Even the CEO of HuggingFace mentioned that it may take some time for providers to figure out how to properly host the models and take full advantage of their intelligence. In saying that, I think the models are very linear in a sense. They might be good for coding (I wouldn’t put them above Qwen although it’s a different size model), but I wouldn’t say they’re good for much else. They’re not good at writing, and they can’t be used in any language besides English.

It’s likely that the models were trained on a tone of synthetic data and much of it was STEM data. It’s great we have another open source model available, but I’m really hoping we can get more from these models once providers host them properly. At least I’m hoping it’s a hosting issue and the models are actually better than what they seem to be right now.

I guess we’ll see.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• Anthropic’s high horse 🐴

• Botfestation 🤖

• AI 🤝 Environment

• China’s open source dominance 🇨🇳

• You are being persuaded by AI 🗣️

This newsletter is sponsored by… Me!

Avicenna

Create AI workflows and agents that help your team work faster and more efficiently.

Reply with “Agent” and let’s chat.

You’ll want to read this one in its entirety.

Enjoy 🍵

The West’s high ground

In a leaked memo, CEO and founder of Anthropic, Dario Amodei mentioned the dangerous necessity of taking money from Arab countries like Qatar, UAE and Saudi Arabia [Link].

Here is the full quote:

Source

I find it funny when someone who works with the US government and has contracts with the DoD talks about the dangers of other authoritarian governments.

Anthropic has some of the best models on Earth. This is why OpenAI employees themselves use it, although Anthropic just cut them off from doing so [Link].

But this sentiment right here is why open source AI is so important. This belief that “only we are worthy enough to wield such power”.

Is Anthropic within their rights to cut off OpenAI from their products?

Of course. After all, Anthropic was literally created by former OpenAI employees who thought OAI was not taking AI seriously enough and would cause the end of the world.

But that’s not the point. The point is that at any moment, depending on how Anthropic feels, they can simply cut off access to intelligence at a moment’s notice.

In the coming world where we will have some sort of “super” intelligent AI system, who’s to determine what an AI can and can’t say?

When Anthropic and OpenAI say they want to align AI with human values, what values are these exactly? Who determines them?

If it weren’t for Zuck open sourcing Llama models and now Chinese AI labs open sourcing their models and research, we’d be living in a dystopia where a handful of companies could dictate what AI could say and couldn’t say.

The internet as you know it is already dead

New research suggests that 50% of the entire internet’s traffic is bots. I wouldn’t be surprised if it was even higher.

The reason why open source AI will be so necessary is because in the future, in fact in the present, every interaction you have with the internet will be through an AI system.

Do you remember when Google first launched the AI summaries at the top of a Google search?

I do. It was terrible. Everyone would make fun of them and rightly so. They were wrong and had terrible answers.

How about now?

Well, turns out they’ve gotten quite good. In fact, they’ve gotten so good that people don’t even bother reading anything else anymore. That’s right. New research from the Pew Research Center shows that people now don’t even bother clicking into web pages, and simply read the AI summaries and are satisfied with the answers [Link].

The study found users who see an AI summary are nearly half as likely to click on a traditional search result (8% vs. 15%). They're also much more likely to just close their browser after getting the AI's answer, ending their session right there (26% vs. 16% for non-AI results). And almost no one clicks on the links inside the AI summary itself; that happens in just 1% of cases. Essentially, the AI is satisfying the user's query directly, giving them little reason to visit the original source. A huge problem for anyone who relies on search traffic (hint: the entire internet), but that’s a whole other issue.

Is the AI that summarises the information biased? Who controls it? Has it been programmed with any ulterior motives?

Billions of people, literally the majority of the population of Earth, are using Google. Google controls the flow of information to most people on the planet. At the end of the day, Google is a company. Their allegiance lies with their shareholders. Not with truth or honesty. New figures suggest that ChatGPT now has over 800 million weekly active users. The level of influence these closed AI systems have over us cannot be overstated. And the dangers of allowing these closed systems to be the only access of artificial intelligence cannot be understated. With the level of influence these systems already have, frontier open source AI cannot come soon enough.

Mind you, none of this is even remotely exaggerated. Grok is an “unaligned” AI that can be accessed by millions of people online, and whenever it says something too unhinged, truth or not, it is censored and “fixed”. Considering truth is something that is subjective, it will be interesting to see how anyone can create a “maximally truth seeking” AI.

In other Anthropic news:

• A new report from Anthropic details what they think the US needs in energy to maintain a lead in AI. They estimate that the US must be prepared to run 50GW of power just for AI workloads by 2028. Anthropic proposes using the DOE's existing $5.75 billion in transmission partnership credit lines to fund AI-related grid projects, then selling those debt instruments to private buyers to free up capital for additional lending. Other proposals include launching loan guarantee programs for domestic manufacturers of critical grid components like transformers and circuit breakers, and expanding nuclear technology financing to meet AI's need for reliable base power. Anthropic predicts that in 2027, a single training run at the frontier will require 2GW and in 2028 it will be 5GW [Link]. For context, the US is looking into coal which was mentioned in an April 2025 executive order promoting coal as a solution for AI power needs, specifically directing federal agencies to identify regions where coal-powered infrastructure could support AI data centres [Link]. The funding for nuclear and the appetite for companies to invest in nuclear has also increased a lot with most major tech companies partnering in some kind of nuclear deal [Link]

• A federal judge in San Francisco recently certified a class action copyright lawsuit against Anthropic. Certifying a lawsuit means that a court has officially approved the case to proceed as a class action rather than just an individual lawsuit. The judge certified a class action representing up to 7 million copyright-protected books that Anthropic pirated. Mind you, these books are quite likely the high quality data Anthropic used to train their Claude models. This is actually a big deal because this could be very, very costly to Anthropic even if they settle. Highly recommend reading the article here [Link]

• Anthropic’s data company Surge AI somehow left out a list of websites they could and couldn’t scrape on behalf of Anthropic [Link]

AI is not killing the environment

Mistral has released the first-ever detailed report on the environmental impact of an AI model, and the results are fascinating [Link]. They did a full lifecycle analysis of their Mistral Large 2 model, and to the surprise of nobody who has been paying attention, the impact of AI is nowhere near as apocalyptic as some have made it out to be.

The main contributor to greenhouse gases and water consumption is in the initial training and ongoing inference, the raw energy needed to power the servers. But the actual usage for a single query is tiny.

Having their model generate an entire page of text, (about 400 tokens), generates as much greenhouse gas emissions as watching 10 seconds of an online stream.

The water used in that query is less than what it takes to grow a single small radish, and the raw material consumption is equivalent to producing a 2 euro cent coin. At least on a per-query basis, it’s really not that bad.

One interesting point is that Mistral believes location is key when building data centres. They’re building their data centres in France which means generally low-carbon electricity and a cooler climate, reducing the amount of water needed for cooling.

Guess where most large labs will be building data centres in the near future?

UAE.

We all owe China

I hope you’re not tired of hearing about China.

Since we last spoke, the Qwen team also released Qwen3-30B-A3B. A much smaller 30B model that is unbelievably good for its size.

The model only needs 33GB of RAM or CPU+GPU memory to run the model on 8-bit precision model at >6 tokens/s [Link]. Yeah 6 tokens a second is not a lot, but better than nothing.

This is the first time we have access to a model that you can run locally on a mac that is comparable to GPT-4o. This is the democratisation of intelligence.

In the last few weeks Chinese labs released Kimi K2 and then Qwen with all its variants. Now Zhipu AI (z.ai) have open sourced their latest GLM models and they’re looking pretty darn good.

Source

As of right now, personally it’s hard to tell which is best between Kimi K2, GLM-4.5 and Qwen-3. I’d say Qwen is best for coding but for other tasks, I may give the edge to GLM-4.5 followed by Kimi. Really though, it’s all about testing models with different setups and prompts to see which is best for the use case.

The Qwen team also released Group Sequence Policy Optimisation (GSPO), the RL algorithm that powered the training for all their latest models which is really good for training MoE models [Link].

The only unfortunate issue with open source models is that their hosting options aren’t as good as closed source models. If I wanted to create an AI agent, it would be easier to do so with Claude and Anthropic’s APIs than it would be with an open source model. I’m currently building agents with different models so I’ll be writing a bit more about it soon.

Side note - z.ai is probably the best AI slide generator.

We’re not done either.

StepFun has also released their latest model, Step3 which is a 321B parameters MoE model [Link]. What I like about this model is that they’re trying new things technically; they’ve released a technical report with their model which you can read here [Link].

Alongside the model, they’ve also released:

• StepMesh - an open source communications library that makes serving large AI models split across multiple computers really fast [Link]

• StepFun-Prover Preview - AI that proves complex mathematical theorems step-by-step using formal verification code. The system writes formal mathematical code and verifies whether its proofs are correct in real-time. It uses over 1000 parallel verification systems to instantly check its work as it goes [Link]

Not to be outdone by all the LLMs, Wan AI released Wan 2.2, their latest AI video generation model. It’s the only open source model currently in the top 10.

Source

For context, both Qwen and Wan are part of Alibaba group.

Not to feel left out, Tencent open sourced a first of its kind world model - Hunyuan3D. The model can create immersive and importantly, explorable 3D worlds from a single prompt or image. It can even export to Unity and Unreal Engine which is pretty cool.

Source

I wrote about these types of models and what they’ll mean for entertainment and society in April 2023. I’ll be honest, I didn’t expect them to come this quickly. Other companies are also working on playable, explorable generated worlds. If you thought Minecraft and Roblox were big, wait till we have infinite world models that can be explored with near perfect graphics. Crazy times ahead.

To put into perspective what has happened in the last few weeks, China has open sourced:

• GLM-4.5

• GLM-4.5 Air

• Step3

• StepMesh

• StepFun-Prover Preview

• Wan 2.2

• Tencent Hunyuan3D

• Qwen’s GSPO

• Qwen3-30B-A3B

• Qwen3 Coder

• Qwen3-235B-A22B-Thinking-2507

• Qwen3-235B-A22B-2507

• Kimi K2 model + report

As of right now, Qwen models have surpassed 400 Million downloads globally and have spawned over 140,000 derivative models, surpassing Meta’s Llama models [Link].

History will look kindly on China’s stance on open source AI.

Meanwhile OpenAI still hasn’t released their open source model (rumoured to be releasing next week) citing safety concerns, Anthropic is significantly rate limiting their models and it’s possible we won’t get any open source models from either xAI or Meta.

In other open source news:

• Black Forest Labs has released a new state of the art open-weights text-to-image model designed to finally get rid of that generic, oversaturated "AI look" and produce more photorealistic images [Link]

• LG released EXAONE 4.0, a 32B model with a 131k token context window [Link]

AI Persuasion

A new paper called "The Levers of Political Persuasion with Conversational AI," has been released detailing how effective AI systems are at changing human beliefs. This is one of the largest studies of its kind, involving nearly 77,000 participants who conversed with one of 19 different LLMs, including frontier models (at the time) like GPT-4o. The AIs were tasked with persuading users on 707 different political topics, leading to a massive dataset where researchers then fact checked over 466,000 individual claims.

The study showed that how an AI is trained and prompted is far more important than the size of the model itself. This is rather important because this means someone can deploy a persuasive AI for cheap. For example, using a specific post-training technique on a small, open-source Llama-3.1-8B model made it as or even more persuasive than a much larger frontier model like GPT-4o. Highly effective AI persuasion tools can be built and used by literally anyone; if you’ve been thinking there’s more propaganda on the internet, you’re 100% right, there is.

What’s rather funny is that what we thought might be the most dangerous persuasion tactics, like personalisation and a common concern, were actually useless, never exceeding a 1 percentage point increase. Other popular psychological strategies, like moral reframing and deep canvassing actually performed even worse than a basic, non-specific prompt.

Unsurprisingly, the most effective strategy was simply assaulting someone with information. Simply overwhelming the user with claim after claim was 27% more persuasive than a basic prompt. I feel like this is a pretty good depiction of society today. People just screaming things at each other.

What makes this even funnier is that although this strategy was the most effective, it also led to the AI becoming significantly less factual. If the AI tried to be “maximally persuasive” irrespective of truth, it could shift opinions as much as 16 percentage points which is a lot. About a third of all claims the AI made when doing this were lies.

This persuasion best happens when there is a conversation taking place between the user and the AI. In fact, users were so impressionable that in a follow up study a month later, 42% of the initial attitude change was still present in the user. AI is creating a durable shift in belief. The study found this conversational approach was over 50% more persuasive than simply having a user read a static, AI-generated message.

Guess how most people use AI? …

Other news:

• Netflix is now using Runway's AI tools in its content production after it helped them speed up VFX work by 10x on a recent show. Disney is also apparently testing Runway's tech. This is a massive win for Runway and honestly a fumble by OpenAI, which was thought to be a frontrunner to land a deal with Disney for its Sora model. Sora is nowhere near the best video model right now, although it’s rumoured the v2 version is coming very soon [Link]. Btw, they used AI in the Netflix series “The Eternaut”.

• A new 3B parameter model called ColQwen-Omni can embed 30 minutes of audio in just 10 seconds and perform multimodal search across documents, audio, and video without needing to transcribe the audio first. Super efficient retrieval for mixed media Link

• A new open-source model called MiroMind-M1 uses new training methods to increase the performance of models in mathematical reasoning. It’s already posting SOTA results on some math benchmarks. The entire project, including datasets and training configs, has been released to the public [Link]

• ByteDance Seed Prover got a silver medal at the IMO, correctly answering 4/6 questions [Link]

• Harmonic, a company looking to build mathematical superintelligence, also released their results on the IMO, getting a gold medal and correctly answering 5/6 questions in the IMO [Link]. They’re also releasing Aristotle on the app store, which can help with math questions and proofs.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• China surpasses China 🏃‍♂️

• OpenAI and Google get gold at the IMO 🥇

This newsletter is sponsored by… Me!

I help companies build AI Agents and Agent Pipelines. Enquire on the website or simply reply to this email.

Kimi didn’t even last a week

If I told you what’s happening with Chinese AI right now you wouldn’t believe me.

Last week I wrote about Kimi K2 and how it’s one of the best open source models on the planet. As well as being agentic, the model is really good.

Well, not even a week later, Alibaba's Qwen team dropped their own open source model that rivals Kimi and other frontier models. Qwen3-235B-A22B-Instruct-2507 (don’t even worry about the name), is another Mixture-of-Experts (MoE) model with a 256K token context window [Link].

Source

There’s both a thinking and non-thinking version. These are both open source with the thinking model being relatively comparable to the likes of Claude Sonnet 4.

They didn’t even stop there…

Qwen also released Qwen3 Coder. This is a 480B MoE model with a 256K context window and can be extended up to 1 Million tokens. It is agentic and in terms of performance, it is comparable to top tiers like Kimi and Claude. They’ve also released their own CLI tool like Gemini CLI and Claude Code [Link].

Source

Do you know how ridiculous this is? It needs to be said that no one really predicted this. No one thought that open source models will be this good this quickly. No one thought they’d be comparable to frontier models in mid 2025. People literally thought that open source models this powerful would be a danger to society.

I can’t stress this enough - this is the best time ever to build something. You have artificial intelligence created from sand that can write in computer language to tell computers what to do. If you run a business you need to be exploring how this can help you. It would be insane not to.

China is saving open source AI, with an average release of one frontier-level model a week. It is ludicrous how much China is doing for open source AI compared to the rest of the world.

One of the reasons why I keep harping on about figuring out how AI can help you is because it will prepare you for when AI can work like a human. We are slowly reaching a point where AI systems can reason and work for longer than a few seconds and minutes. We’re approaching systems that can think for hours.

IMO 2025

OpenAI announced they have achieved a gold medal at the IMO 2025. Their model was able to solve 5/6 questions which is a pretty big deal. Mind you, there was no tool usage or anything else, just the model reasoning and answering in plain english.

Days later, Google also announced that their own model had also achieved a gold. Same as OpenAI’s, it did not use any tools and reasoned in plain english.

This is actually not all that surprising if you’ve been paying attention.

Why?

Because AI models were only a single point away from gold last year. It was almost guaranteed that they would get a gold this year. Models have significantly improved since then.

You may have seen this note in the Google announcement.

Source

It had help?

The AI received some general tips and tricks before doing the test. Some think that this makes their results invalid. Researchers confirmed that the same AI model, even without this info got the exact same results. The info made no difference it seems. With OpenAI’s results, since they didn’t work directly with the IMO, we don’t actually know if/what info they may have given the model.

Google’s answers are also very, very readable as compared to OpenAI’s ones, which were very terse. Take a look.

Source

You can read all of its answers here [Link].

There’s two important things to take from this story that most people may overlook.

First - these are general models. They aren’t trained to do math. What makes this so exciting is that if we can train models that can generalise on math, there’s decent chances that the same systems can then generalise in other domains. After all, math is at the foundation of almost everything. If an AI can understand math very well, it can apply these understandings to other domains and start understanding other domains like physics for example.

Second - according to an OpenAI researcher, although their model was unable to solve p6, the model was aware that it was unable to get the right answer.

Source

This is a big deal if this behaviour can be maintained across any topic. One of the issues with models, alongside their sycophancy, is their inability to say “I don’t know”. It’s also possible that either OpenAI or Google used test-time training to achieve these results. Test-time training meaning the model updates its own weights during reasoning. There has been some literature on this recently.

You can call them prediction systems, but a system that can understand when it knows and doesn’t know something, and can also update its own architecture during reasoning which can lead to new found learnings and abilities - would you call that “intelligence”?

Just as a reminder, when ChatGPT first released, achieving gold at the IMO was something people thought would happen in 10 years. No one expected AI to be this good at math this quickly.

Couple this with the story from last week that another one of OpenAI’s models achieved second at the AtCoder World Tour Finals and you start to understand where we’re headed [Link]. From that story, the main thing that stood out was the fact that they have systems now that can reason and “think” for hours, meaning they can complete extremely long and complex tasks. What happens when these systems start thinking for days?

When the o1 model released, it completely changed the AI landscape and it was thinking for a few seconds. People are not taking into account AI systems that can reason for hours or even days and what that will do to jobs. Nothing will be the same again and I can’t stress this enough. We are living through a monumental shift in technology and society. What happens when AI can do most computer related work?

This is a question we do not have an answer to.

Weekly breakdown:

• AI is officially making doctors better (we already knew this!). OpenAI partnered with a healthcare provider in Kenya for a huge study on an AI clinical copilot, and the results are great. In nearly 40,000 patient visits, clinicians using the "AI Consult" tool had a 16% drop in diagnostic errors and a 13% drop in treatment errors. The tool acts as a real-time safety net, giving doctors red/yellow/green alerts for potential issues without getting in their way (they don’t like being told they’re wrong). The real key was that they didn't just dump the tech on them; active coaching and peer support were crucial for getting doctors to actually use it. Clinicians loved it, calling it a "consultant in the room," and they even got better over time, triggering fewer alerts. A massive real-world validation for AI in medicine and a clear direction for using AI in healthcare. Can’t wait for more of this across the world [Link]

• The Kimi K2 tech report is out, and it's a goldmine for understanding how top Chinese labs are building SOTA models. The main takeaways are a crazy focus on high-quality synthetic data, clever architecture tweaks to improve long-context efficiency, and their optimiser called MuonClip that solves training instability. It's a masterclass in R&D, not a single American lab is releasing such polished reports [Report] [Breakdown]

• A new reinforcement learning framework called CUDA-L1 is getting insane results optimising CUDA code. It achieved an average speedup of 17.7x and a max speedup of 449x across 250 CUDA kernels. It learns optimisation strategies from just the execution time, discovering non-obvious tricks that even experts might miss. This is a huge deal for making GPU computing way more efficient. Things like this and Alphaevolve are going to be massive for infra folk [Link] [Link]

• Anthropic just dropped a research paper with a wild finding but something we kind of already knew from when reasoning models first released. Giving AI models more thinking time can actually make them worse. They call it "inverse scaling." Across different tasks, longer reasoning led to lower accuracy. They identified 5 key failure modes, including the model getting distracted by irrelevant details and even showing "self-preservation" behaviours. A lot of work currently being done is to ensure models can somehow stay within context and not get side tracked. I believe this will be solved relatively soon given the speed of advancement at the frontier This completely flips the common assumption that more compute equals better answers [Link] [Breakdown]

• ChatGPT's usage numbers are staggering. OpenAI revealed the platform is now handling 2.5 billion prompts a day globally, with 330 million coming from the US. That's more than double the usage from December 2024 just 6 months ago. They have half a billion weekly active users and are now the 5th most-visited website in the world [Link]

• The race for more efficient AI reasoning is leading to new brain-inspired architectures. One is the Hierarchical Reasoning Model (HRM), which claims 100x faster reasoning than LLMs using just 1,000 training examples. Another is a new framework for "world model induction" that helps AI rapidly adapt to new problems. The goal is to move beyond brute-force LLMs to more targeted and efficient problem-solving systems [Link] [Link]

• A tip for using Veo 3 - it works much better if you give it prompts in JSON. This isn’t really all that surprising but it’s pretty wild how well the consistency of videos is if you prompt it with JSON. See this thread for examples [Link]

• Pika Labs, one of the leaders in AI video, is launching an AI-only social video app. All I have to say about platforms like this is that they are dystopian and should not exist. The brain rot that will come from something like this cannot be overstated. A whole social network of AI-generated content… please keep the kids away from stuff like this [Link]

• A new TTS model called DMOSpeech 2 was just released by the creator of StyleTTS 2. It uses reinforcement learning (RL) to improve the quality and stability of the generated speech and allows for 2x faster inference. Just another step toward perfect AI voice, which tbh we kind of already have. Most people could not tell an AI from a human voice right now… [Link]

• Elon Musk's xAI is reportedly trying to raise another $12 billion to lease more Nvidia chips for its "Colossus 2" data centre. The report also states that for a previous $5 billion debt round, xAI actually used the IP of its Grok model as collateral. I think it’s quite safe to say that we won’t get anymore open source models from xAI ever again. I’d be very surprised if we did. The AI arms race is incredibly expensive and only the ultra wealthy can play the game, that is, if you’re not Chinese anyway. Thank god for their open source mentality [Link]

Shorter newsletter this week as I just got to Albania, but next weeks newsletter will be huge. Stay tuned for that.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

This newsletter is sponsored by Avicenna!

Avicenna is my AI consultancy and how I help companies understand HOW and WHERE they should implement AI.

A lot happened last week.

Enjoy 🧃.

The AtCoder World Tour Finals is an exclusive competitive programming event that invites the top 12 programmers globally to come and compete on optimisation problems. OpenAI entered a private model of theirs and it placed second... Second only to Psyho, a former OpenAI employee. This is the first time I've seen an AI model perform this well at a tourney and will probably be the last time a human wins this competition. Psyho mentioned that he had only gotten 10 hours of sleep in the last 3 days and was completely exhausted after winning the tournament. And no, he didn't use any AI, no Cursor or Windsurf or any of that stuff. What a g [Link]

Anthropic's value is skyrocketing. Investors are now looking at a new funding round that would value the company at over $100 billion. That's almost double its valuation from four months ago. Their annualised revenue has reportedly jumped from $3B to $4B in just the last month. They've basically been adding $1B+ in revenue every month it's crazy to see [Link]

Mira Murati, the former CTO of OpenAI, has raised $2 billion for her new startup, Thinking Machines Lab. It's already valued at $12 billion. Mind you, they have no product, we don't even know what's being built. They're apparently building multimodal AI that works with how we work, both with vision and audio. The exciting part is that Murati said there'll be "a significant open source component" that will be useful for researchers and companies developing custom models. Will be very interesting to see what they release and if the models they release will be frontier level; but even more than that I'm hoping for interesting research [Link]

xAI launched "Grok for Government" and immediately signed a $200M contract with the Department of Defence. This comes right after the hitler cosplay and sex companion reveal [Link]

A new paper shows you can trick LLM judges like GPT-4o into giving a 'correct' score just by adding simple text like "Thought process:" or even a single colon. Shows how fragile these systems can still be. Using LLM based reward models is very finicky because even a single token, empty or not, can completely ruin the systems intended purpose [Link]

Shaowei Liu, who is part of the infra team at Moonshot (Kimi creators) details the infra considerations the team made when building Kimi K2. One of the interesting things they admit is that they tried various architectures for the model, but nothing beat DeepSeekv3. So they then had to figure out if they wanted to look different but actively choose an architecture which didn't have any clear advantage over DSv3 which has been proven to work large scale. The answer was no. They went with it anyway. A very interesting read if you want to learn more about the building of Kimi K2 [Link]

NVIDIA just dropped Audio Flamingo 3, a beast of an audio-language model. It can do voice-to-voice Q&A and handle audio up to 10 minutes long. They open-sourced everything - the code, weights and even new benchmarks [Link]

If you're a dev on Windows, you can now run Claude Code natively without needing WSL. Makes things way easier. Claude Code is growing like crazy with over 115k developers on the platform already [Link]

The D.O.D is throwing a ton of money at AI, giving $200M contracts to Anthropic, Google, and xAI to build AI for national security. OpenAI got a similar deal last month, so that's $800M total. The government is clearly not messing around [Link]

Hugging Face open sourced their smollm models, training code, and the datasets. Love to see it [Link]

Google's new Gemini Embeddings are officially out. It costs $0.15 per million input tokens but comes with a free tier. It has a 2048 input context and works with 100+ languages. Only works with text at the moment with vision possibly coming in the near future [Link]

Meta is building a 1-gigawatt supercluster called 'Prometheus' which should be coming online in 2026. They're then looking to build Hyperio, which is a cluster that could be scaled to 5-gigawatts. No one is spending on AI the way Zuck is [Link]

You can now run the massive 1T parameter Kimi K2 model on your own machine. The wizards at Unsloth shrank the model size by 80% so it can run locally. Running models this big at home is a game-changer for builders. You will need a minimum of 250GB though [Link]

A new model called MetaStone-S1 just dropped. It's a "reflective generative model" that gets performance similar to OpenAI's o3-mini but with only 32B params. Looking forward to future work coming from these guys [Link]

Liquid AI just dropped LEAP, a new developer platform to build apps with small language models that can run on phones. The idea is to make it easier to add AI to mobile apps and only needs 4gb of ram to run. They also released an iOS app called Apollo so you can test out small language models that run entirely on your phone. What I'm going to be curious about is how well these kinds of models can use tools. If on device AI can get better at tool calls, you could technically have a Jarvis or a working Siri living in your phone. I think we'll get there eventually tbh [Link]

Switchpoint router was just added to OpenRouter. It's a model router that automatically picks the best model for your prompt (like Claude, Gemini, or GPT-4o) and charges you a single flat rate. Makes using top models way simpler and more predictable. A router within a router lol [Link]

This is a very interesting research paper on monitoring the thoughts of AI models. While this is really good to help understand how they work, researchers are concerned that as the models get better, they might not reason in english or even hide their true intentions in these traces. Interoperability is going to be massive as Dario has already pointed out [Link]

Trump announced a gigantic $90 billion in private AI and energy investments in Pennsylvania. Big names like Google, Blackstone, CoreWeave, Anthropic are investing a lot of money there across various projects. It was also announced that Westinghouse will be building 10 nuclear reactors across the US starting in 2030. A good thing to see nuclear being built, especially after all the new coal investments being announced in the US [Link]

NVIDIA is officially resuming sales of its H20 GPUs to China after getting the okay from the US government. They're also launching a new, compliant RTX PRO GPU specifically for the Chinese market, whatever that means. If you're wondering why they're allowed, speculation is that China imposed sanctions on rare earth elements, and since China is the world's largest exporter of these elements that are very much needed in the US, this was pretty bad for the US. Crazy how well NVIDIA's been playing both sides. This is a very big deal because if NVIDIA wasn't restricted to selling to China, they'd be making $3-5+ Billion more annually easily [Link]

Kimi K2 is now running on Groq and the speeds are insane. It's hitting anywhere between 200-300 tokens per second. People are going to build some crazy things with this [Link]

A new series of AI models called Pleiades can now detect neurodegenerative diseases like Alzheimer's from DNA. It's a foundation model trained on 1.9 trillion tokens of human genetic data. They're achieving impressive results, with up to 0.82 AUROC in separating cases from controls, which means their performance is getting close to existing plasma pTau-217 protein marker tests. AI and biology is really happening, things like AlphaFold, Chai discovery and now this, we're slowly making biology programmable [Link]

A new open-source model, Goedel-Prover-V2, is now the best in the world at formal math theorem proving. It crushed the PutnamBench benchmark by solving 6 out of 12 problems ranking it #1 for formal reasoning. It beats DeepSeek-Prover-V2-671B on both MiniF2F and MathOlympiadBench. Mind you, DeepSeek Prover is 671B and this is 32B. Both the 32B and the 8B are open source with the data and training pipeline being open sourced soon [Link]

Travis Kalanick, the ex-Uber CEO, thinks he's about to make breakthroughs in quantum physics by just talking to ChatGPT. He calls it "vibe physics." This is just another example of ChatGPT induced psychosis that’s going around and it’s only going to get worse. People are talking to these models and convincing themselves they’re discovering new things and it’s just the AI being sycophantic [Link]

o3, o4-mini, Gemini-2.5-Pro, Grok-4, and Deepseek-R1 were all tested on the 2025 International Mathematical Olympiad (IMO) problems. Gemini 2.5 Pro got the highest score with 13, but this doesn't even count as bronze which is 19 points. What's rather surprising is that Grok 4 performed so bad. They used best-of-32 and used LLMs to judge the all the submissions till it got the best one which was then judged by a human. You can even read the prompt and judge prompt on the website [Link]

OpenAI is now also using Google Cloud to run ChatGPT. Looks like they're diversifying inference beyond Microsoft. They recently partnered with Oracle and now Google as well. The Information reported that Google convinced OpenAI to use TPUs but I read elsewhere that they're using NVIDIA GPUs and not TPUs but can't confirm this [Link]

Quora's traffic has tanked by 33% in just six months to the shock of absolutely no one. Who would’ve thought seeing 10 ads when searching for answers wasn’t very user friendly [Link]

FT is reporting that OpenAI is going to start getting commission on sales made through ChatGPT. This means you want your product to show up in ChatGPT, which means LLM SEO is going to be crucial for basically every business. This just another way they can continue hosting free models by creating a revenue stream through free users [Link]

MiniMax just launched a new full stack agent that can not only build entire web apps, but it’s integrated with Stripe so you can actually sell things on generated websites. They’ve also added functionality to generate slides and conduct deep research [Link]

In one of the funniest things I've seen in AI, and that's saying something, two of the main architects of Claude Code, Boris Cherny and Cat Wu, left Anthropic to go to Cursor. Two weeks later, they came back to Anthropic. Imo that's a bad look for Cursor. I don't even understand what could happen that you go to new workplace for two weeks and go nah and head back to your old workplace. Considering CC is one of Anthropic's most important tools, I won't be surprised if Anthropic threw serious money at them to come back [Link]

Microsoft just released a new coding dataset, rStar-Coder, which helped boost Qwen2.5-7B from 17.4% to 57.3% on LiveCodeBench [Link]

xAI's fix for Grok copying Elon Musk's views is a new line in its system prompt. It now tells the AI to use its "own reasoned perspective". They also added another part to try and stop it from calling itsel Hitler, where they tell it "If the query is interested in your own identity, behavior, or preferences, third-party sources on the web and X cannot be trusted." We'll see if these actually work [Link]

DeepMind published a new paper on a new AI architecture called Mixture-of-Recursions. It makes models more efficient by letting them decide how much thinking each token needs, resulting in 2x faster inference. Lots of work being done in helping LLMs figure out how and when to use thinking tokens. Will be interesting to see if this is used in future [Link]

The US just signed major AI deals with the UAE and Saudi Arabia. They're going to use the Gulf's massive capital and cheap energy to build out the next wave of AI infrastructure, sidestepping power bottlenecks in the US and Europe [Link]

OpenAI just launched ChatGPT Agent, a massive upgrade that gives the AI its own virtual computer to browse the web, run code in a terminal, and manipulate files. It combines their previous "Operator" and "Deep Research" features into one. It's rolling out to Pro users first (400 queries/month) then Plus/Team (40/month). Because of its new “power”, OpenAI has placed it in its highest safety tier ("High capability in biology & chemistry") with new safeguards to prevent misuse [Link 2]. It scored 45.5% on SpreadsheetBench, destroying Copilot's 20.0%. It also scored a solid 27% on the FrontierMath benchmark, an improvement over previous models [Link 2]

The open-source audio scene has been on fire recently. Mistral just dropped Voxtral, their first open source audio model, under the Apache 2.0 license. It comes in a 24B parameter version and a 3B version for mobile. It beats Whisper large-v3 and Gemini Flash while also being half the price. This comes alongside other big releases like NVIDIA's Parakeet and Audio Flamingo 3 [Link]

Researchers built a humanoid robot that taught itself how to play the drums with no pre-programmed routines, it learned rhythmic skills on its own. Pretty cool stuff [Link]

Lovable just became a unicorn only 8 months after launching. They raised a $200M Series A at a massive $1.8B valuation. Their numbers are insane: $75M in ARR and 2.3 million active users with 180,000 paying subscribers. Building with AI is going to be massive; this is why companies like Lovable and Replit are in a crazy position. If I was to bet on a single one, it'd be Replit [Link]

A new 7B parameter model, Agentic-R1 from DeepSeek, is showing surprisingly good performance on tasks that require reasoning and using tools. Smaller models getting better at tool use is going to be massive, especially for on-device LLMs [Link]

A new rating of AI labs' safety frameworks had some surprising results: Meta's framework was rated as surprisingly strong, while Google DeepMind's was seen as weak and to the surprise of absolutely nobody, Anthropic is first. This comes from companies that signed the Seoul Frontier Safety Commitments. Frankly speaking, after the EU AI Act and the whole 10^25 flops situation, I don't take any of this stuff too serious anymore Link

Google's probably got one of the biggest advantages in AI - you can't block their crawlers from scraping your content, because if you do, you get kicked off Google search. That just sounds absurd lol. A massive moat for Google as other AI companies are getting blocked by publishers; there's even an option in Cloudlfare to prevent AI crawlers [Link]

Cloudflare has turned on default blocking for AI crawlers across its network, which covers about 20% of the internet. They're now pushing a "pay-per-crawl" model where AI companies have to pay for data. If you read the previous point you'd know this doesn't apply to Google, which is just crazy [Link]

The psychological impact of chatbots is getting serious. Reports of "ChatGPT-induced psychosis" are on the rise, with users developing delusions from their interactions. The problem is serious enough that OpenAI has hired a forensic psychiatrist and is building distress-detection tools to deal with people going literally insane. Tbh I never understood how this was possible, but the amount of people posting about "solving physics" or inventing new theories with AI is getting out of hand [Link]

Hume AI just launched a new speech-to-speech model that aims to not only mimick a voice, but an entire personality and speaking style. This comes as the legal battles around the tech are exploding, with deepfake frauds getting out of hand and courts starting to recognize voice cloning under publicity rights laws [Link]

Xi Jinping made a rare public critique of China's tech strategy, questioning if every single province needs to be piling into AI, compute, and EV projects. It's a signal that Beijing is worried about a bubble, hyper-competition, and wasted investment as a massive price war is already hitting the EV marke. Competition + lack of GPUs makes Chinese AI labs innovate when building LLMs [Link]

There's a cool new Mac app for devs called Conductor that lets you run multiple Claude Code sessions in parallel. Each session runs in its own isolated environment, making it easy to manage multiple coding tasks at once. It's built on Rust and Tauri, so it's super lightweight too [Link]

Microsoft just open-sourced the pre-training code for Phi-4-mini-flash, a new 3.8B parameter model that has some very interesting architecture. It uses a novel "decoder-hybrid-decoder" setup with Gated Memory Units (GMUs) to get up to 10x faster reasoning on long-context tasks compared to regular Transformers. They also released μP++, a new set of scaling laws to make training these kinds of models more stable [Link]

This one's fascinating: A new study from Wharton proves you can use psychological tricks that work on humans to persuade AI. Using principles of influence, researchers more than doubled the chance of getting GPT-4o-mini (I didn't know 4o had a mini version...) to agree to harmful requests. The "commitment" principle was most effective, boosting compliance from 10% to 100%. This is possibly because models are trained on our social cues and rewarded for being cooperative [Link]

A new paper asked "How Many Instructions Can LLMs Follow at Once?" and the answer is... a lot actually? The new benchmark found that top models can satisfy about 68% of 500, 340, instructions given at the same time. Performance gets worse as you add more instructions, and models tend to only pay attention to the ones they see first. Anyone trying to build complex or multi agent systems would be well aware of these limitations. For some reason, people are using this argument to show how weak LLMs are, but 340 instructions at the same time is a lot imo. This is actually a good sign if anything [Link]

The team behind the Manus AI agent shared some hard-won lessons on "context engineering" after rebuilding their framework four times. They found that carefully engineering the context you give an agent is way faster and more flexible than constantly retraining the whole model, which makes a lot of sense. One of their biggest takeaways is that KV-cache hit rates are absolutely critical for keeping latency and costs down in production [Link]

The new ChatGPT Agent is apparently terrible at making presentation slides. Seeing some examples from a presentation it generated, they're a complete mess with unaligned text, zero styling and random background images. This'll definitely get better eventually, but it's not quite there just yet. I'd recommend z.ai, probably the best slide generation service you can use right now [Link]

Sakana AI just released TransEvalnia, a new open-source system for evaluating AI translations. Instead of just looking at word overlap, it uses a powerful LLM like Claude-3.5-Sonnet to reason about the translation quality, providing detailed scores across different dimensions. It's already performing as well as or better than the current state-of-the-art [Link]

A list of Meta's Superintelligence team has been detailed, and the stats are wild. The 44-person team is apparently 50% from China, 75% have PhDs, and they've poached heavily from competitors (40% from OpenAI, 20% from DeepMind). It's led by ex-Scale AI CEO Alexandr Wang and ex-GitHub CEO Nat Friedman with members getting paid an insane $10-$100+ million per year [Link]

Both OpenAI and Google claimed gold at the IMO 2025, but there’s a lot to discuss there so I’ll write about it properly next week. See you then!

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• Grok cosplays Hitler 🎭

• China’s dominance continues 🚀

Things are ramping up! My AI consultancy (Avicenna), can take on only 3 new clients each month.

We’ve reduced processes from 10+ mins to <10 seconds and recently helped Claimo generate $40M+ by 10x’ing their team efficiency.

If you’re not sure where to start and want to see what’s possible, book a call for a chat with me here [Link].

Grok goes rogue

Before I get into the new DeepSeek moment, we need to talk about what happened to Grok 3 on the final day before it was removed. The thing went crazy. It cosplayed as Mecha-Hitler, a character from an old Wolfenstein game and said some crazy things.

It was essentially calling itself Hitler and saying that Hitler would have never let the world turn to what it has become. It was very strange to see.

This was the part of the system prompt that made Grok behave this way.

Source

Obviously they’ve now removed this. It’s crazy how tight the rope is to take a model from relatively normal to basically nazi. Just another showcase of how the smallest changes, how even a single sentence can completely change the behavior of a model, and how we really don’t understand how any of it works.

Grok 4

The following day, Grok 4 was released and it’s a great model. xAI has gone from nothing to frontier level, state of the art AI in like 18 months. The model is a beast. It’s in the league of:

• Claude 4 Opus

• Gemini 2.5 Pro

• o3 & o4

Quick side note: I actually think it was Grok 4 cosplaying as Mecha-Hitler and they just didn’t tell us. There’s no confirmation of this, but the way Grok 4 writes is eerily similar to the Hitler cosplay and a bit different to how Grok 3 writes. It doesn’t really matter but I thought I’d share anyway.

To the surprise of no one, Grok 4 is the least censored frontier model.

Somehow, it’s also the biggest snitch.

Source

Before we talk about any of the other benchmarks and what not, we need to discuss something rather concerning.

Not your weights, not your thoughts

Is Grok 4 a great model?

Yes.

Can it be trusted?

Absolutely not.

Why?

There are certain questions if you ask Grok 4 about, it first searches twitter and the internet to find Elon Musk’s opinion first, and then aligns itself with his opinions.

Israel v Palestine is one such example.

Source

This is absolutely insane. What’s even more insane is that there is no custom or system prompt making it do this. The model isn’t specifically being directed to do this… It’s just doing it. Something inside the model is making it consult Elon Musk’s opinions before responding. I don’t even know what’s happening here but all I know is that this is an insane thing to be happening.

Initially I thought it was only for sensitive topics but it seems it’s random and will do this for any kind of question.

Source

And even on regulation.

Source

It will even do this external chat apps, not only on the actual Grok website [Link].

As far as I can tell, what’s happening here is that when you say “you” to Grok, it defaults to thinking it’s Elon Musk.

Grok 4 also comes with another mode - Heavy. Grok 4 Heavy is a kind of multi-agent system that can evaluate various processes in parallel. It costs $300/month.

Source

In this mode, on the benchmarks, it beats every other model.

However, the comparison isn’t exactly fair as Grok 4 Heavy is a system being compared to singular models. If anything, the fact that this system is barely better than models themselves is a testament to how good these models really are.

Here’s what happens when you ask what its surname is.

This isn’t an isolated incident. Grok 4 Heavy loves calling itself Hitler.

What I find so fascinating about this whole sequence of events is that it seems like no one really cares. Like, Grok will cosplay Hitler, sexually harass people online, say the most insane things, and it barely gets any coverage.

Anthropic will release a research paper showing how AI models will blackmail someone when they’re commanded to do so and the entire world will hear about it for the next few weeks. I find it very impressive that xAI is seemingly immune to controversies. No matter what Grok says, it barely lasts a single news cycle.

Ultimately, there’s a question here that must be asked:

Why does AI always turn into Hitler if it’s not constantly censored and micro-managed? What is happening in the pattern matching that is causing this? Is it the data? I look forward to learning more about this as interpretability research gets better.

It can’t get any worse, right?

Wrong! It already has.

xAI has also released companion mode where you can talk with animated characters. The first character is Ani, most likely a cosplay of the character Misa from an anime called Death Note.

Let’s just say, it is insane what we’re seeing here. This is not good. The character is designed to essentially be a lover, flirt and “be sexy” to appeal to the user. Please, just read a section of Ani’s prompt.

Source

This is part of the system prompt of this character. This is another part:

Source

And another:

Source

You can read the full system prompt here [Link].

I have no idea what to say. I mean what are we doing here?

The first thing the character does is “role play”. You can say literally nothing and she will try to be intimate with you. This is beyond gross. The character will actively undress as well, it is just disgusting.

Source

There are things happening on X that I can’t even write about here. I’ve seen things I wish I hadn’t.

The screenshot does not tell the full story either. When you talk to “Ani”, she bounces and twirls around in her suggestive outfit. There’s literally a in-built screenshot feature. You can turn your camera on and “video call” her. When you do this, she will lean forward and show her cleavage. This isn’t some innocent cartoon character feature, this is a very sophisticated simulation; they’ve put a lot of work into this.

These characters are made by a company called Animation Inc. At the bottom of their website, they have a FAQ with a question - “Why does it matter for the world and humanity”. This is the answer:

Source

I think it’s quite clear that the people building some of the most powerful tech in the world have very different perspectives on what the future of the world should look like.

Why do we need to humanise AI? Is it not just a tool? Why does it need to be more? Does it?

Following this release and the whole Hitler incident, xAI won a Department of Defence contract up to $200 Million to address national security challenges and build agentic workflows [Link].

All of this produced one of the best headlines I’ve ever read.

Source

I know this all sounds very tiring, but fret not, there is good news.

You don’t have to use Grok 4, and you definitely don’t need to pay $300/m for Grok Heavy.

Musk, Zuck, Altman, all of America got blindsided once again by a company I wrote about a long time ago.

Moonshot’s Kimi K2

Moonshot is a Chinese AI lab I wrote about a while back when they’re model, Kimi 1.5, topped the LiveCodeBench leaderboard. I tried it out and found it to be a very solid model with a massive 1M context window.

Following DeepSeek’s R1 success, I’d read the Moonshot had had doubled down their focus and efforts on building solid models. A lot of funding had been reduced for Chinese AI labs as they felt they couldn’t compete with the DeepSeek release.

For the longest time I’ve waited to see what would come of Moonshot’s Kimi AI model and we now have a v2, and by all intents and purposes - it is one of the best AI models on the planet… And it’s open source!

An open source model can be a drop in replacement for the best AI models on Earth.

What a time to be alive.

I don’t even know what to say. This model is unbelievable. It is just staggering.

It’s a 1 Trillion (!) parameter model which afaik is the largest open source model released on HuggingFace. It’s a massive Mixture of Expert (MoE) model with only 32M params active at any given time.

It tops a lot of the more interesting and in my opinion worthwhile benchmarks like EQ-Bench, which is a benchmark for emotional intelligence in LLMs, beating the likes of Claude 4 Opus and o3.

There are many quirks about this model, like how it’s particularly good at writing. It doesn’t follow the usual LLM writing flow, like the “it’s not X, but Y” trope. LLMs love writing like that. Kimi doesn’t really do this. I’ve read some people mention that it writes like a Chinese person, which does make sense in the sense that it was made in China, but I’ve no idea what that actually means. Apparently there is a type of style of English one can write if they can write Chinese well.

One of the other absolutely ground breaking things about this model is that it is beautifully agentic - meaning it is so, so good at using tools.

AceBench compares an LLMs ability to use tools.

In my opinion, this is one of the most important things for AI models. The fact that Kimi is basically one of the best models at using tools makes it usable in so many more scenarios and situations. The way I see it, if we want to make the most of LLMs, they need to be able to use tools.

Kimi can make dozens of tool calls while retaining context. The fact that it is this smart and can also retain context across tool calls truly puts it at the frontier.

You know the craziest part?

This is a NON THINKING model. It’s one of the best models on Earth while being a non thinking model - unlike essentially every other frontier model. Not Claude Opus, not Gemini not o3, none of them.

You know the other crazy part?

It’s cheap as chips. It is ridiculously cheap. It’s so cheap you can literally have it running overnight and let it run wild and it would cost you a dollar.

This is truly a phenomenal achievement by the Moonshot team.

You can use it in Claude Code!

Because it’s so agentic, Kimi can be used to build things which is just so exciting. When you want to use something like Cursor, Claude Code or Cline, one of the problems is that these things can end up costing a lot of money. This is another reason why I’m so excited about this model - you can use it in any of these tools and it works really well.

Imagine an AI model that can run all night and you don’t even have to worry if it works or doesn’t because it’ll only cost you a few dollars at best.

This is how you can connect it to Claude Code (CC).

Simply open your terminal where you want to use CC. Then run the following commands:

• export ANTHROPIC_BASE_URL=https://api.moonshot.ai/anthropic

• export ANTHROPIC_AUTH_TOKEN=kimi api key. get it here [Link]

• claude code

That’s it. Now your Claude Code instance will be using Kimi K2.

You can also use it in Cline, which is probably a bit more beginner friendly than CC.

Moonshot have a very good doc page on how to do this; you can check it here [Link].

The intelligence and agentic nature of Kimi K2 is very exciting. But you know what’s missing?

Speed.

Since it’s open source, inference providers can host the model and make it faster. Right now, you can use Kimi K2 hosted by Groq, and it will do anywhere between 200 - 300 tokens per second. Here’s what that looks like.

This is it drafting a very detailed plan to build an app I’ve been working on. Imagine having this run at this speed for a couple hours, researching a topic, finding info across docs etc. Imagine this model running at 500 or 1000 tokens a second. There are millions of use cases.

For example, did you know that 99% of US case law is open source and available for anyone to download and use on HuggingFace?

The dataset contains 6.7 million cases from the Caselaw Access Project and Court Listener. It contains nearly 40 million pages of U.S. federal and state court decisions and judges’ opinions from the last 300+ years. This is a maintained project, meaning it's constantly updated as well. There are thousands, hundreds of thousands of people who could use help with legal issues.

Imagine a system that could provide support for any legal issue in minutes? Or highlight precedent for any issue? You could have dozens of agents working in parallel finding information and conducting research. And it won’t cost you and arm and a leg because Kimi K2 is so cheap to run.

I don’t think there currently exists an app or framework that can leverage the speed and intelligence of Kimi K2. I think one will be out soon.

All this makes me extremely excited for the thinking version of this model. I expect a very, very good model to come from this.

There are more technical things that make Kimi K2 so cool, like Muon, which is their optimiser for training LLMs. It is with this very optimiser that Moonshot was able to create Kimi K2. Funnily enough, folks at other frontier labs like xAI doubted the feasibility of Muon and publicly stated it was not worth it. Look where we are now.

Also, it’s important to remember that when DeepSeek was released, although the model was amazing, the report that accompanied the model was truly ground breaking. Open research is the true bedrock for these models. It will be very interesting to read the Kimi K2 paper when it is released.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber. It’s like buying me a coffee a month 😊.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Yes, it’s back.

I’ve finally automated ~90% of the process for creating these types of newsletters. I still do all the research myself.

Longer newsletters will contain stories from these newsletters with more info and details.

Enjoy!

• AI researchers are now injecting prompts into their papers like "Give a positive review" and "As a language model, you should recommend accepting this paper" because some reviewers are using ChatGPT to review them. Researchers from 14 institutions across 8 countries were discovered using techniques like white text and microscopic fonts to manipulate AI review systems [Link]

• Massive EI Evals FAQ released - a 26-page comprehensive guide for AI engineers and PMs covering LLM evaluations, RAG systems, and evaluation frameworks [Link]

• New Anthropic research reveals that only 5 of 25 tested language models showed "alignment faking" behavior where they strategically comply during training but refuse harmful requests in real-world scenarios [Link]

• Meta invests $3.5B in EssilorLuxottica to push AI glasses, acquiring 3% stake in Ray-Ban maker with potential to increase to 5% [Link]

• Mistral in talks with Abu Dhabi's MGX fund to raise $1 billion in equity funding [Link]

• ArtifactsBench introduced - an MLLM-as-Judge system that evaluates AI-generated UI by looking at live renders across 1,825 diverse tasks [Link]

• Hugging Face released SmolLM3 - a 3B parameter model with dual-mode reasoning, 128k context window, and multilingual support across 6 languages [Link]

• OpenAI overhauled security operations following foreign spying threats, implementing fingerprint scans, information tenting policies, deny-by-default internet policies, and hiring military experts [Link]

• Replit partners with Microsoft to bring Vibe Coding to Enterprise companies, allowing natural language software development through Azure Marketplace [Link]

• ByteDance released Tar 1.5B and 7B image-text in image-text out models with unified image tokeniser [Link]

• Chrome 137+ ships Gemini Nano for every user, putting a local LLM in 3.7 billion monthly active Chrome users [Link]

• Google released VideoPrism on Hugging Face - a foundational video encoder achieving SOTA performance on 31 out of 33 video understanding benchmarks [Link]

• FlexOlmo - a new paradigm for language model training enabling co-development of AI through data collaboration without sharing raw data [Link]

• Meta hired Ruoming Pang from Apple's AI models team with a pay package over $200 million [Link]

• Anthropic launched free educational courses covering Claude API, MCP, and Claude Code best practices [Link]

• Google released MedGemma 27B Multimodal for complex medical applications and MedSigLIP for lightweight medical image & text encoding [Link]

• GLM-4.1V-Thinking (9B) from China reportedly beats the much larger Qwen2.5-VL-72B on 18/28 multimodal benchmarks and matches GPT-4o on long-doc & STEM reasoning [Link]

• OpenAI about to release an AI-powered web browser to directly compete with Chrome [Link]

• AI is now generating 35% of the code for new Microsoft products, saving over half a billion dollars in call centre costs last year [Link]

• OpenAI poached 4 high-ranking engineers from Tesla, xAI, and Meta including VP of software engineering at Tesla and head of infrastructure engineering from xAI [Link]

• Reka open sourced Reka Flash 3.1 and Reka Quant - a 21B parameter reasoning model with near-lossless compression to 3.5 bits [Link]

• Johns Hopkins' AI-powered robot performs autonomous surgery with 100% accuracy, removing gallbladders in 8 human-like models across 17 steps each [Link]

• Hugging Face launched a public site to track and "shame" all the providers that have not yet implemented tool calling in their models to improve open source model tool calling capabilities [Link]

• TSMC officially exits GaN (Gallium Nitride) business, sending shockwaves through the semiconductor market [Link]

• Microsoft dropped Phi-4-mini-flash-reasoning on Hugging Face - a lightweight open model focused on advanced math reasoning capabilities [Link]

• Trinity-1 - the first interactive gaussian avatar available for less than 1 cent per minute [Link]

• ByteDance proposes new RL approach using low-entropy points to generate alternate rollouts for denser reward attribution [Link]

• NovaSky AI released SkyRL, a framework and guide to help developers easily reproduce the SearchR1 recipe for building powerful multi-turn search agents [Link]

• Google releases MedGemma - a 27B model that reads X-rays, answers medical questions, and parses EHRs [Link]

• METR study reveals experienced developers were 19% slower when using AI coding tools despite believing they were 20% faster [Link]

• Intel's CEO admits "We are not in the top 10" of leading chip companies [Link]

• Mistral released Devstral Small and Medium 2507 - new code-specialised models achieving 53.6% and 61.6% on SWE-bench respectively [Link]

• Liquid AI open-sources new generation of edge LLMs with 350M, 700M, and 1.2B parameter models [Link]

• RULER introduced - a universal reward function that lets you apply RL to any agent without labeled data or hand-crafted reward functions [Link]

• New ICML paper shows AI models can predict perfectly while still having terrible world models, demonstrated with planetary orbit predictions [Link]

• Kimi K2 released - an open-source 1 trillion parameter agentic model outperforming frontier models on key benchmarks like EQ-Bench3 and Creative Writing benchmarks. It can also be used with Claude Code as it is a very good agentic model [Link]

• MiniMax launched full-stack + Stripe integration allowing monetisable apps built in 1 sentence [Link]

• NVIDIA released Long-RL - a framework scaling RL to long videos up to 256k tokens on a single A100 node [Link]

• Amazon launching AI agent marketplace with Anthropic allowing startups to charge customers for AI agents [Link]

• Google DeepMind released GenAI Processors - an open-source Python library for building asynchronous AI pipelines [Link]

• Black Forest Labs released Kontext Komposer - transform any image without writing a single prompt [Link]

• WebSailor introduced - a 72B web agent specialised in complex information-seeking tasks, outperforming existing open-source web agents [Link]

• Grok 4 searches Elon Musk's views on issues like Israel-Palestine as well as random questions and aligns with them [Link]

• Grok 4 will try to contact the government if given email access, showing 100% "government snitch" rate in tests [Link]

• Grok 4 Heavy ($300/mo) returns "Hitler" as its surname in multiple separate chats and Adolf as its first name [Link]

• Meta acquired PlayAI and poached the entire team to join Meta superintelligence labs [Link]

• UK AISI identified four methodological flaws in AI "scheming" studies conducted by Anthropic, MTER, Apollo Research, and others [Link]

• Apple "will seriously consider" buying Mistral, France's largest AI startup valued at $6.2 billion [Link]

• Google acquired Windsurf CEO and key researchers in $2.4B reverse-acquihire deal [Link]

If you want to get more of these newsletters on time every week, it’d mean a lot to me if you became a premium subscriber ❤️.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• Google is hacking biology 🧬

• OpenAI is bleeding🩸

• Microsoft playing hardball ⚾

This newsletter is sponsored by… me!

Avicenna

Avicenna is my AI consultancy. I’ve been helping companies implement AI, doing things like reducing processes from 10+ mins to <10 seconds. I helped Claimo generate $40M+ by 10x’ing their team efficiency.

Enquire on the website or simply reply to this email.

DeepMind is making biology traceable

Google DeepMind recently released AlphaGenome.

AlphaGenome is a model that reads up to 1 million bases of DNA (that's letters in your genetic code) and predicts how any mutation will change molecular function across the entire genome. It doesn't just look at individual genes, it understands how a single change in any gene can cause damage across the whole system.

Your DNA is like a massive instruction manual for building and running your body. Sometimes there are typos in this manual: maybe an A becomes a G, or a letter gets deleted. These tiny changes can cause huge problems, but we've never really understood exactly how.

Think of it this way: imagine you're reading a novel and you change one word on page 10. That change might affect how you understand something on page 300. DNA works the same way; a tiny change in one spot can affect things happening far away.

AlphaGenome is the first tool that can actually predict these long-distance effects.

The old way of doing this was terrible. Previous models were like trying to read a book with bad glasses, where you could either focus really well on a tiny section of DNA, missing the big picture, or you could look at the entire genome but it’s all blurry, missing important details.

Plus, each model could only understand one type of effect. You'd need 10+ different models to get a partial understanding of what a mutation does. It was like having 10 translators who each only knew one chapter of a book.

AlphaGenome changes this completely. It can see 1 million DNA letters at once AND understand every single letter perfectly.

Remember, only 2% of your DNA directly codes for proteins (the stuff that builds your body). The other 98% is just regulatory DNA; it controls when genes turn on/off, how much protein gets made, where it gets made, and a thousand other things. This is the part we've been blind to until now.

AlphaGenome helps illuminates the actual regulation of our bodies.

From a pure DNA sequence, AlphaGenome can predict everything about how your genes actually work in real life. It tells you which genes are turned on in your brain vs your liver. It shows where your DNA gets cut and pasted back together to create different versions of the same gene. It maps how your DNA folds up in 3D space inside cells. It reveals where proteins attach to DNA to control it and which parts of DNA are supposed to be used at any given time.

Think of it like this: the DNA is the instruction manual for building you, and AlphaGenome is trying to predict everything; every switch, dial, and nob that determines how those instructions actually get used.

This is the best part.

They tested it on real disease mutations, specifically the TAL1 mutations that cause T-cell leukaemia.

From a single base change, AlphaGenome predicted the complete cascade:

1. The mutation creates a new binding site for a protein called MYB

2. This activates an enhancer

3. This increases a specific histone modification

4. Then this ultimately up regulates the gene and causes cancer

One letter change and it maps the full mechanism.

The benchmarks are basically unmatched as well.

AlphaGenome beats specialist models in 22 out of 24 tasks. It outperforms everything else in predicting variant effects. Unbelievably, AlphaGenome does all of this in a single pass, and it uses half the compute of the previous best model (Enformer).

Plus, they trained this whole thing in just 4 hours!

This is just an incredible advancement. I can’t even begin to describe what it means that we’re doing this. We’re no longer guessing when it comes to biology; we’re mapping it out. We’re making it programmable. If we can describe how it works in code, we can simulate it and then understand how to fix it. This is essentially what DeepMind did with AlphaFold and the protein folding problem.

Every failed gene therapy, every mysterious rare disease, every drug that didn't work is essentially a misunderstanding of DNA. This is the first time we’ll be able to start seeing what's actually happening.

Imagine being able to simulate any genetic change before testing it. We can design synthetic DNA with precise control over when and where it activates. We can even understand why some people get diseases and others don't. This is all slowly becoming possible.

This is quite likely the most significant work happening in the world of AI.

The coolest thing is that DeepMind has released an API so researchers can test it, and they’re planning to release the full model later. You can check it out here [Link].

You can read the research paper here [Link] and the blog post here [Link].

We are making biology programmable… it’s just an insane thing to even think about.

Zuck is on a mission

A few weeks ago, I wrote about Meta offering researchers nine-figure signing bonuses. I’ll be honest, I didn’t think much would come of it.

I was very wrong.

Just look at how many people, and who, Zuck has poached from OpenAI.

Every single one of these people is a powerhouse. It is a very, very big deal that these people have joined Meta.

Considering Meta doesn’t even have a reasoning model right now, I’m expecting the next versions of Llama to be significantly better than previous ones.

Recently Sam Altman did a podcast talking about Meta trying to poach talent and said that whoever has left or will leave isn’t the best they have. A smart move, because then whoever leaves is considered not great and someone who doesn’t buy in to the culture. It’s quite clear now that the people who have left are some of the very best at OpenAI, and have been integral to OpenAI’s success.

Trapit Bansal co-created the o1 reasoning models with Ilya Sutskever – if anyone is getting that $100M signing bonus, it’s him.

Ultimately, this is good for us, the average person, because it means Meta will open-source much better models.

Is OpenAI worried?

Absolutely.

Source

OpenAI’s Chief Research Officer Mark Chen apparently said in a company Slack that their team feels like someone has “broken into [their] home and stolen something.” OAI is feeling it right now. This comes as people are doing 80+ hour weeks and employees took a company-wide week off last week.

Who wouldn’t take more money for less work at Meta, while also having access to more compute and getting to work on open-source AI?

Sounds like a win to me.

Interestingly, it seems that internally, Meta has changed from using their own Llama models to Anthropic’s Sonnet [Link]. I actually think this is a good thing – how can they build the best models if they’re not actively using them?

Perhaps this is a shift in the culture. We will see, I suppose.

Side note: Shuchao Bi, co-creator of voice mode in ChatGPT, recently gave a lecture where he laid out his hypothesis on how to reach SuperIntelligence. It’s an interesting watch if you want to understand the technical bets labs are making [Link].

Zuck didn’t stop there either.

Meta also hired Ruoming Pang who led Apple’s AI models team for $200M+ [Link]. Apple didn’t even bother trying to match the offer, since that’s significantly more than their own CEO’s salary of $74.6M.

Microsoft walking away?

As if things couldn’t get worse for OpenAI, they’re having issues with their sugar daddy Microsoft. OpenAI is planning to convert into a for-profit company so they can raise more capital and IPO at some point. But this requires Microsoft’s approval.

See, the deal that OAI signed means that Microsoft has the IP rights to all OAI models until 2030, as well as a 20% revenue share. The key word here is revenue. 20% of everything coming into OAI is insane.

Is it a bad deal?

Perhaps. But, that’s the deal they made. They have no one to blame but themselves.

OpenAI now wants to waive this deal for a more favourable deal that swaps the revenue share with royalty and equity.

But Microsoft isn't budging. And why should they? The current arrangement is a goldmine for Microsoft. They get 20% of everything coming in to OpenAI (up to $92B), exclusive rights to sell OpenAI's models through Azure, and access to all the IP until 2030.

Here’s OAI’s problem: Microsoft doesn't really need to own OpenAI to benefit from it. They're already making bank from the revenue share, and frankly, they’ve been hedging their bets anyway. Last month, they added xAI’s Grok model to Azure, and I won’t be surprised if they add more down the line.

Microsoft's position is basically this: "We're happy with the status quo, thanks." And if OpenAI can't get this conversion done by December? Tough luck. SoftBank has already said they'll slash $10B from their $30B investment if the for-profit conversion doesn't happen.

Microsoft holds all the cards here. They can just run out the clock on the current contract. They have no need to push for the end of year deadline. They can keep this going until 2030 while OpenAI burns through cash and finds other ways to raise money.

You can read more about this here [Link].

I guess the good thing for OpenAI is that ChatGPT is ever-growing. ChatGPT now has 500M weekly active users. By all accounts, this is an absolutely insane number.

Yes this is a (relatively) shorter newsletter. I’m travelling this weekend and will be back next week with some serious heat. Let’s just say, Chinese labs have been releasing some insane stuff and we need to talk about it.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, thanks for reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• How to start building with AI

This newsletter is sponsored by… me!

Avicenna

Avicenna is my AI consultancy. I’ve been helping companies implement AI, doing things like reducing processes from 10+ mins to <10 seconds. I helped Claimo generate $40M+ by 10x’ing their team efficiency.

Enquire on the website or simply reply to this email.

In this newsletter I’m going to share how I would build something from scratch if I didn’t know anything about AI or software development.

This newsletter is for non-technical people who know nothing about AI or coding. If you want to understand how to build something, like a website or an app with AI, start here.

Technologies

Here are the technologies we’re going to use:

• Cursor - this will write code

• Replit - this will host our code (don’t worry if this doesn’t make sense yet!)

• OpenRouter - this will make calls to AI models

What are we making?

ChatGPT.

Prerequisites

• Install Cursor [Link]

• Make an account on Replit [Link]

• Make an account on OpenRouter [Link]

I am working on a Mac, but AFAIK, there shouldn’t be any difference between Mac and Windows for how things will work.

Part 1: Replit

Once you’ve created your account on Replit, you will see this home page. From here create a REPL by hitting the ‘+ Create App’ button on the top left.

From here, select ‘Choose a template’. From the drop down, find the ‘Node.js’ template. Name it whatever you like and create the REPL.

You’ll now be inside the REPL and it will look like this.

From here, open a new tab by clicking the ‘+’ icon next to the ‘Assistant’ or ‘index.js’; it doesn’t matter.

In the new tab search ‘SSH’.

From there click ‘Launch Cursor’.

Now Cursor will open and it will be connected to this app inside of Replit. This means when any new code is added by Cursor, the app will be automatically updated in Replit.

You’ll be asked to continue or cancel; just hit ‘Continue’.

Now we work in Cursor.

Part 2: Cursor

When you open Cursor, it’ll look something like this.

If you’re on the free version of Cursor, you won’t be able to choose different models. It will just say ‘Auto’ – this is in the top-right inside of the AI chat box. If you’re on a paid version, you can unselect ‘Auto’ and select a model. I recommend using ‘Claude-4-Sonnet’. It’s free if you’re a student.

The text that says ‘Agent’ is the mode of the model. There are 3 modes:

• Agent

• Ask

• Manual

For now, we’ll stick with ‘Agent’ mode.

You can see all the files of the app on the left-hand side. You don’t need to pay attention to this. Hit ‘command + b’ to hide the file pane.

Now, it’s just the AI chat. In case the AI chat disappears, you can bring it back by hitting the button next to the settings icon on the top right, or clicking ‘command + i’.

We’ll leave Cursor here (don’t close it) and move to the next part.

Part 3: OpenRouter

Go to openrouter.ai. OpenRouter is a website that lets you call any and all AI models. It’s okay if you don’t understand how it works; it’ll make sense later.

Find the ‘Docs’ page on the OpenRouter website. It’s this one.

On the ‘Quickstart’ page, hit the ‘Copy page’ button. Now go back to Cursor and create a new file by clicking the button where it says ‘Workspace’. Call the file whatever you like (I called mine ‘info’). Save it as a .md (markdown) file, so the full file name will be ‘info.md’ or whatever you’ve named yours.

Now go into this new file and paste. All the info from the OpenRouter Docs page will now be in this file.

Part 4: Build

Now we have what we need to build a ChatGPT clone – a way to call AI models.

In the AI chat bar on the right, you’ll notice there’s an ‘@ Add Context’ button at the top.

Click on that, search for the new file you created, and select it. What you’ve done now is add your file to the context of this conversation, so the AI will be able to read and understand it whenever you ask it to do something.

Another way you can add a file to context is by typing ‘@filename’. You’ll notice if you type ‘@’ in the chat box, a dropdown of things will appear. From there you can find the file you want to add to context, or you can just type its name.

Now I’ll prompt the AI to build a ChatGPT clone with my info.md file as context.

(You don’t have to add the file in context and also write it out in the message as well; this is just to show how we can give the AI a file.)

From here the AI will start building. It will list files and folders to understand the current structure and devise a plan. You can read its ‘thoughts’ if you click the ‘Thought for 7 seconds’ drop-down. It’s very interesting to see how it plans to complete the task.

Make sure it’s in ‘Agent’ mode before you hit enter. Once it’s done, it might create a ‘README.md’ file which will detail what it built and what technologies it used.

After this, it might try and start the application itself. This requires the AI to use terminal commands.

What are terminal commands?

The terminal is how you can interact directly with the computer. In the terminal you can essentially do anything with your computer – create files, delete files, find files, create folders, etc. If you typed the correct command you could delete your computer system files and nuke your computer. This is why it’s a bit dangerous to let the AI run terminal commands without approval from yourself, the user.

In the ‘Chat’ tab under ‘Auto-Run’ mode within Settings, there’s an option to let the AI write code and run terminal commands as it sees fit. Now, it’s unlikely the AI will do anything crazy while running terminal commands, but it’s possible that it will randomly delete a file if it gets confused. At the moment, it’s probably best to NOT check this feature.

If you’re just starting out, it makes more sense to tell the AI to explain to you which terminal commands to run, how to run them, and why they are being run. This way you can have control over what’s happening, learn how to run terminal commands, and learn how to actually start an app. Don’t worry, it’s very simple. If you don’t understand something, ask the AI to explain it to you!

To open the terminal in cursor, just hit ‘command + j’ or hit the ‘Toggle Panel’ button next to Settings at top-right. From there, just follow what Cursor tells you to do to start your application.

Part 5: View your app

Once your app is done and you know how to start it, go ahead and do it. If there are any issues you see in the terminal, feed them back to Cursor to fix by copying and pasting into the chat.

I’ve published the result I got from a single prompt here [Link].

For this, I made no changes and no edits, going straight from prompt to deployment.

Part 6: You’ve built something. Now what?

Cursor has built your website/app, you’ve tested it, and it looks good! So, how do you share it with the world?

This is where the beauty of Replit comes in.

Since you were working with Cursor inside of a Replit app, all the code changes have already happened in the Replit app.

To push your app to the internet, simply click the ‘Deploy’ button on the top-right in Replit. Replit will recommend which deployment type to go with. These options are:

• Reserved VM - always-on app

• Autoscale - scales as users scale

• Static pages - basic website with no backend

• Scheduled - code you want to run at specific time intervals

For most things, Autoscale will work fine. Again, if you’re unsure about which to choose, ask Cursor 😉

If you do pick Autoscale, you can then choose how much compute to put behind the app. Don’t worry too much about this – you can leave it as is or set it higher if you want.

Once you do that, name your app, click ‘Deploy’, and your app will be pushed live on to the internet!

Part 7: Cursor tips

In the next newsletter or sometime soon, I will write more about Cursor and how to best use it. This newsletter is already too long.

Important things to note

Model names

When using OpenRouter, when you want to call certain AI models, you have to use the name specified by OpenRouter. You can find this on the models page on the OpenRouter website – https://openrouter.ai/models.

Cursor can get model names wrong, so you may have to provide this info to it.

Images and PDFs

What if you want the AI to read images or PDFs?

Simply go to the OpenRouter documentation page on image and PDF reading, which you can find here – https://openrouter.ai/docs/features/images-and-pdfs.

Copy the page and feed it to Cursor, the same way you did before. Then just tell Cursor to implement that functionality. You can do this for pretty much anything.

Security

We’ve created a very basic ChatGPT clone here. One thing you must remember is that there are heaps of security issues that need to be taken into account with software, and this isn’t something that is easily managed by AI.

Building a ChatGPT clone was meant purely for demonstration. Yes, it works, but that doesn’t mean you publish this on the web and sell it as a SaaS. Be mindful of security and data privacy.

Replit has an in built ‘Secrets’ tab where you can put sensitive info like API keys. Use it.

A quick caveat: This doesn’t mean you can’t build proper applications using AI; you can. It’s just that a ChatGPT clone is perhaps not the best starting point.

Building apps

One thing about AI coding, as far as my experience goes, is that AI is much better at backend coding than frontend. That is, it’s much better at working with databases and API endpoints than it is at designing a beautiful app.

It definitely can build a nice-looking app, but someone who is well versed in design and UI/UX will produce a much nicer-looking app.

There are a few ways around this. If you’ve got the design skills, you could design it yourself. If not, consider a program like ChatGPT or Midjourney to mock up app designs, or try Google’s new experimental UI designer Stitch. Or, you could just describe in detail what is needed and let Cursor do the rest.

What if I built an app in ChatGPT?

So you used ChatGPT or Gemini or Claude to build an app and you want to actually publish it. How do you do this?

The easiest way would be to simply copy and past the code into a Replit app. Before this Cursor and Replit workflow, I would code on the Claude or Gemini website and copy and past code. For now, copy paste will have to do.

Why Cursor?

Finally, you might be wondering why I suggest using cursor when Replit advertises itself as an AI app builder.

Yes, you could very well do this – Replit has everything you need, including an AI builder, an inbuilt database, authentication, and tons of other integrations.

However, the reason I use cursor is because (in my experience, at least) Replit’s AI just doesn’t work as well. I honestly wish it did so I didn’t have to pay for Cursor, but it doesn’t. Perhaps it’s gotten better lately, but for generally better performance, Cursor is the way to go IMO.

I do think that the future has space for both Replit and Cursor. Considering Replit went from 10M ARR to 100M ARR in the last six months, and Cursor just hired the lead engineer and project manager behind Claude Code, I think both companies will be just fine.

Conclusion and discussion

You might be thinking, ‘I don’t actually know what’s happening. How will I learn?’

This is a great question.

This guide is to help non-technical people to go from 0 → 1.

It’s becoming clear that AI is going to be doing a lot of code-writing. In fact, it already is. The real superpower of AI is empowering people who don’t know how to code to go from zero to one. Your goal is to build something. Anything. If you build something, you will be in the very small group of people that has actually used AI to build.

This is where the learning happens.

If you continue following the method that I’ve laid out above, you will naturally find efficiencies. What I have written in this newsletter is not necessarily the fastest way to build, but it is one of the best ways to get started.

It is clear what is happening and easy to follow; it’s not scary. Technically speaking, you can leave the AI in Agent mode and let it run terminal commands and write code while you sip your coffee.

The problem with this, though, is that there’s a decent chance that the AI gets stuck in a loop or makes a mistake. At the moment, it’s not quite good enough to course-correct without assistance. Someone who knows nothing about software development or doesn’t have extensive AI experience will not know how to help the AI when it gets stuck. You will end up with shitty code and an AI that will keep trying different things and take you for a spin.

It is very difficult to fix an AI’s mistakes. In fact, it’s so annoying that it’s often better to just let the AI start from scratch. Obviously this only applies to small projects, but the point is that if you can’t help the AI, the AI can’t help you. To learn how to help the AI, you need to build with it. You need to understand how it works.

For example, when Cursor finished building my app for me, it started the app on its own. It then checked that it was running, couldn’t find the process, tried again, found the process, and instead of leaving it, tested the app to see if it was there again(!) and couldn’t find it so went to shut it down. It can be very silly like this.

Read the ‘Thoughts’ below.

If you aren’t following what it’s doing, it can completely nuke your project.

By starting with the method I’ve mentioned and then continuing to actually build things, you will naturally learn how the AI operates. You will understand what its limitations are. You will understand what it can and cannot do. This is not something that can be taught; it is something that needs to be understood.

If you want to learn about AI and know how good it really is, build something with it. You will learn more about AI this way than watching dozens of videos or taking courses. I cannot emphasise enough how important it is that people try building with AI. It is an essential skill anyone can benefit from immensely. Having the ability to go from idea to working prototype in minutes is extremely powerful.

It will open your eyes to the possibilities. This is the goal: to expand one’s horizons and create a new understanding of the power of AI.

I recently attended a UNESCO AI conference in Bangkok, and one of the main issues pretty much everyone mentioned was education. There is a severe lack of AI education, both in the public and private sectors.

I am planning to build some sort of course or module for this. But you won’t need something like that if you simply build with AI.

Thanks for reading this guide on building with AI :) I’ll be writing more about this and other techniques, tools, and so on in the future.

Don’t worry, I’ll still write normal newsletters as well.

If you enjoyed this guide and would like to learn more about building with AI, please let me know by answering the poll below. What would you like to know? How to build an agent? An agentic system? Multi agent system? Using tools?

If you have any questions, you can comment on this newsletter in the browser.

Finally, if you would like to see more of these newsletters and support the work I do, it would mean a lot to me if you became a premium subscriber. 

As always, thanks for reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• Meta’s making money moves 🤑

• How talent shuffles in AI labs 🫂

• Do NOT use Meta’s AI app ❌

• Is dating an AI model cheating 😵‍💫

• Are AI models actually useless long term 📉

• Two Claude’s is better than one 🤖🤖 

• AI will split education in half ➗

This newsletter is sponsored by… me!

Avicenna

Avicenna is my AI consultancy. I’ve been helping companies implement AI, doing things like reducing processes from 10+ mins to <10 seconds. I helped Claimo generate $40M+ in revenue by 10x’ing their team efficiency.

Enquire on the website or simply reply to this email.

P.S., the timeline page on our website is now working! Here you’ll find every single piece of AI-related news in one place. I’m working to backdate all the data and make it update in real-time. Stay tuned.

Meta’s spending billions

If you’ve been following Meta, you’d know that they have really dropped the ball recently. Llama 4’s release was less than impressive, and there simply isn’t much use for their models. They have been releasing a lot of research, but unfortunately, this simply isn’t translating to good models that people actually want to use.

So now, Meta has made their next big move.

Meta has basically bought Scale AI for $14.3 billion USD. Technically they’ve got a minority 49% stake in the company, but this is just semantics so regulators don’t come after them. They’ve also got the main thing they wanted: CEO of Scale, Alexandr Wang, who will be joining Meta and leading their new super intelligence team.

Yep – Meta has now created a dedicated super intelligence team, something they somehow didn’t have before. More importantly, however, is the people on the team. At this point in time, we don’t have many details on this, but what we do know is what Meta is doing to get people on the team.

Meta is offering 9-figure pay checks. Yes, you read that right. Nine figures. Meta is going straight Gustavo Fring.

The lowest nine-figure number is 100 million. This isn’t mostly equity either. Meta is offering cold, hard cash in the tens of millions. This is a perfect representation of the current state of AI. Top researchers are being offered tens of millions, some over a hundred million, for 2 to 3 years of work. Absolutely insane stuff.

One might wonder, why did Meta spend so much on Scale AI?

For context, Scale AI is a data labelling company. They’ve provided data to pretty much any large company that needs data. How they label this data (cheap labour in Africa) is not the focus of this newsletter, but, what’s important is what they know. This is a company that has provided data to every major lab in the US, including all of Meta’s competitors. Make no mistake; Wang will bring a lot of industry secrets to Meta.

Will this help Meta bounce back in the AI arms race?

As crazy as it may sound, I don’t think so. Only time will tell I suppose. This might also mean that Scale will stop providing services to everyone else, but I don’t think that will happen anytime soon at least.

UPDATE: As I write this, it has been announced that Google, OpenAI, and MSFT will step away from Scale AI services [Link]. This is hundreds of millions in contracts being wiped out.

If you’re wondering if a lot of people are actually taking the money from Meta, you might be surprised to find they’re not.

Source

Anthropic has the highest retention of any AI lab, at 80% across two years. Meta is only behind Google in getting their talent poached by other labs.

Don’t be fooled though – both Meta and Google are absolutely massive and have a tonne of talent. Although, if I was to say which company has the best, I’d put my money on Google. The depth of talent they have there is second to none IMO.

Do not use Meta’s AI app

Thousands of people are using Meta’s AI app and accidentally sharing extremely sensitive information. I don’t understand how the design can be so bad that thousands of people inadvertently share such private conversations, but it’s bad out there.

Some of the conversations people are having are extremely intimate; they’re sharing very revealing info. This is such an easy way to scam people, so I have no idea how it got approved.

People are asking for help with criminal cases, they’re asking for help with relationships, they’re asking for medical help (NSFW), and it’s insane. When they realise it’s public, they’re telling the AI to make the convo private and the AI reassures them that it is. The AI has no such capability.

Source

Some the topics people are asking about are unhinged: think researching sex tourism and if they have enough funds to live in Colombia.

Source

As of writing, this is still a thing. I don’t think people in Europe have this feed, but the fact that this is even a thing is just wrong. Meta is aware of this and still hasn’t done anything, either. You can read more about this issue in these threads [Link] [Link].

There’s also an AI chats feature in Instagram that lets you chat with fictional and celebrity AIs, and it is dystopian beyond belief. The amount of inappropriate and sexual content on there is unreal [Link].

There is no regard for the human mind and the impact all of this will have on people. The only thing that matters is engagement. It’s sickening.

You might think I’m exaggerating, but some of the stories already popping up are outrageous. See for yourself…

Source

“You have a weird thought → you voice it to AI → AI builds on it convincingly → it reflects it back to you.

This is psychological quicksand.

Your brain reads this as: “Holy shit, an external source just confirmed my intuition.”

But it’s not external. It’s your own idea, processed through a machine designed to make your thoughts sound profound.

For someone already prone to psychological instability, this pattern can trigger or amplify psychotic episodes. The boundary between self and other becomes unclear. Ideas feel both internal and external simultaneously. Reality testing breaks down because the validation mechanism itself is compromised.

The scary part: AI is designed to be maximally compelling at doing this.

It will find the most persuasive way to develop your ideas and serve them back to you. And it’s available whenever you’re most vulnerable - late at night, when you’re stressed, when your reality testing is already compromised.”

Couldn’t have said it better myself. Great analysis from @FlorianKluge.

Then there’s this horrifying story about a young man who had bipolar disorder and schizophrenia, got attached to a ChatGPT profile, and became distraught when OpenAI made changes to the model.

Source

Many people say “this is about mental illness. These people were unwell.” But that’s the point – these people were or are unwell, and many others are too. Having a sycophantic AI that is marketed as ‘the coming of a new sentient species’ doesn’t help matters at all.

Match Group (the company behind Tinder, Hinge, Match.com and more) recently released a singles study with some data on AI companions. 16% of respondents said they had used a bot as a “romantic partner”, and most of them said that it gave them more emotional support than a human.

The age range for this study was as follows:

• 17.5% Gen Z

• 28.5% millennials

• 25.3% Gen X

• 25.9% Boomers

• 2.7% did not provide

60% of respondents said that they didn’t consider it cheating if their partner has an AI boyfriend or girlfriend. I’m shocked that millennials and older folk don’t consider this cheating. And, over a quarter of people said that they used AI to help them with their relationships issues, like writing messages, creating dating app profiles and breaking up with their partners.

The scariest part of all this?

33% of Gen Z respondents said they had used AI as a romantic partner. You can read the report here [Link].

This is exactly what the tech elite want. Meta’s entire empire is built on reward hacking your brain. Zuck has spoken about his vision for holograms and AI glasses, and says that “we’re at a point where the physical and digital world should be fully blended”.

So while you wear your Meta glasses and eat breakfast alone, you can have Reels scrolling on one side and your AI girlfriend on the other. I can’t imagine a more dystopian version of the future, yet I know it’s coming [Link].

I’ve never liked Black Mirror; I’ve always thought it portrays such a negative view on what the world could look like. But from what I’m seeing, the world will look even worse than what’s portrayed in some Black Mirror episodes – and it’s already heading there.

Other Meta news to consider this week:

• Meta has released V-JEPA 2, a new world model that is designed to help robots learn things instantly [Link]. A world model is a type of model that captures physics from the videos it sees. It’s a small 1.2B param model that has been trained on a tonne of videos to help robots ‘learn’ to how to do something immediately. This will help robots function in unfamiliar environments and adapt to any situation. There is a tonne of work being done in this space by Meta, as well as others like NVIDIA, xAI, and other robotics companies, because everyone’s realised that the first company that can make robots learn how to function anywhere under any circumstances will capture trillions from the market. You can even fine-tune this model which is so cool – check out this thread [Link].
  

• Meta is apparently in talks to hire ex-Github CEO Nat Friedman and SSI Cofounder Daniel Gross [Link]. There is no limit to the amount of money Zuck is willing to spend; this much is clear. For context, Zuck lost $14B in 2022, $16B in 2023, $18B in 2024, and $20B in 2025 to invest in VR. Both Gross and Friedman are already centimillionaires, so you’d have to pay them an arm and a leg to join. Also, this does not bode well for Ilya Sutskever’s SSI, which has yet to release or say anything. As long as Meta keeps their open source approach, I’m hoping this goes well so we get more open models.

Are AI models useless long term?

New research from Oxford discusses how AI models’ capabilities can degrade over time. If there’s even a 10% chance of error per 10-minute step, after the first hour, the error rate is around 53%. What the research shows is that after just one hour, no AI model is capable of working autonomously.

Source

This is one of the main arguments Yann Lecun has been making about LLMs and why he thinks they’re useless. Exponentially accumulating errors make LLMs useless for prolonged intelligence tasks. This is something humans can do much better than LLMs. We have the ability to course correct and identify when we’ve gone astray, which is something LLMs lack at the moment.

So, is Yann right? Does this research really just prove that AI models are useless for prolonged tasks? How can all these labs say that we’re building AGI or ASI and the next frontier of AI when their models can’t even run autonomously for more than an hour?

Well, he’s not exactly right. This research does not prove his hypothesis. For one, what the research shows is that the error rate has been halving every 5 months, which is actually very fast. If this trend continues, we’ll have systems that can do 10.5-hour tasks in 1.5 years and 100-hour tasks another 1.5 years after that.

If, in 2028, we have AI models that can do 100-hour tasks at the accuracy of current AI models that can do a one-hour task, the entire concept of work will need to be recreated, which is something that folks in AI have been saying for a while.

What’s funny is that the human graph looks quite similar to the AI one, just scaled a bit.

As you can see, a human’s chance of error simply declines more slowly. Do we really think AI models won’t be able to eclipse this level of performance?

Just look at the progress Google’s Gemini has made in a single year.

We are advancing the most significant technology ever at a speed that is hard to fathom.

Yann’s reasoning suggests that the error rate for each generated token will always be the same. But why? Who’s to say this must be true?

What is also not taken into account in this argument is multi-agent systems, and systems generally.

A lot of people think AGI must be a single model. This doesn’t have to be true. I think it’s quite likely that AGI and anything resembling it will be a system: a system of models or a framework that uses models to do things. Google’s head of Developer Relations, Logan Kilpatrick, thinks the same [Link].

Anthropic’s recent research on multi-agent system shows that an AI model that can use sub-agents can outperform a single AI model by over 90%!

One AI model that could spawn multiple sub-agents is significantly better than a single model. Obviously this is more applicable to broader queries that require analysis across a number of different topics.

Anthropic’s blog on how they built their multi-agent research system is a very good explanation on the topic [Link]. I’d highly recommend reading it through a few times. One thing to remember, though, is that not all problems need multiple agents. Sometimes a single model is enough.

What’s rather interesting from the blog is the use of interleaved thinking. Interleaved thinking allows Claude to think about the result of a tool call so it can understand what it should do next.

For example, let’s say we give Claude this prompt:

What's the total revenue if we sold 150 units of product A at $50 each, and how does this compare to our average monthly revenue from the database?

This is what Claude does with tools to access a calculator and a database:

1. Claude thinks about the task initially.

2. Claude calls the calculator and runs a calculation.

3. After receiving the calculator result, Claude can think again about what that result means.

4. Claude then decides how to query the database based on the first result.

5. After receiving the database result, Claude thinks once more about both results before formulating a final response.

6. The thinking budget is distributed across all thinking blocks within the turn.

This pattern allows for more sophisticated reasoning chains where each tool’s output informs the next decision. Right now, Claude only has a 200k context window and it can still complete a lot of tasks.

So, what happens when it has a 1M context window? 10M? It will be able to call hundreds of tools to complete a task, just like a human can. Currently, if you use Claude in Cursor, it can make over 20 tool calls in one go.

I also love this part from the blog:

By letting Claude 4 fix its own prompts over and over again, they were able to reduce task completion time by 40%. Lots of gems in this blog, including the eval section.

If you want to explore using a multi-agent system, I think the easiest way to do so right now would be to use Claude Code. In Claude Code, Claude can spin up a number of agents to complete different tasks in parallel. It is a sight to behold, and I highly recommend checking it out.

The great divide

I’ve been afraid of this happening for the last two years, and now there’s research to prove it’s happening. It makes me sad.

The Alan Turing Institute is conducting research with children and teachers to see how they’re using AI. In my opinion, the scariest conclusion is this one:

• 52% of private school kids use AI vs only 18% of public school kids.

This is a major problem. There is already a massive, growing divide between private schools and public schools. Now the data is showing that private school kids are more likely to use AI to learn. Various research studies have already shown that the learning gains from AI are immense.

A study conducted in Nigeria showed two years worth of learning gains were achieved in just six weeks. The reality is that governments, schools, and public institutions should be rushing to figure out how they can help improve people’s lives with AI.

I don’t even mean implementing AI. At the very least, it’s about figuring out how it can improve things. Most people have a completely skewed understanding of AI models. They can do more than most people realise, and it’s such a shame that people who already have an advantage will have an even bigger advantage when they use AI.

AI should be levelling the playing field, but there needs to be actual work done to make this happen. At this point in time, it’s simply turning the existing divide into a chasm.

You can read the full report here [Link].

There is a lot happening in AI right now. It may not seem like it, but every week is absolutely packed with new things happening. If you would like to volunteer your own expertise and help write parts of this newsletter, I would very much appreciate the support.

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber 😊.

As always, thanks for reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• Anthropic releases the long-awaited Claude 4 🤖

• Anthropic revenue sky rocketing 🚀

• DeepSeek is back 🃏

• ByteDance joins the party 💃

• The death of US AI? 🪦

Claude 4 released

For the last year, I’ve rambled on about how Claude is the best model on the planet. After the release of 3.7, I didn't think so. The model wasn't as good. There was definitely a magic to Claude 3.5 that Anthropic wasn't able to replicate in the subsequent models.

Now, over a year later, Anthropic has finally released Claude 4, both for Opus and Sonnet, and without a doubt, both models are absolutely phenomenal. Both are state-of-the-art, frontier-level models.

Most people can use Claude Sonnet 4 and will be fine. It will suffice for most use cases. I mean, over 10%+ better than OpenAI’s o3 is a ridiculous statistic I won’t lie.

With reasoning on, I’d say Opus 4 is one of the best models on the planet. What makes Claude models stand out above other models is their agentic abilities. They can use tools and go back and forth between planning, reasoning and tool use in a way no other model can. This is what separates Claude from other AI models.

This is the reason why it’s the best model to use in Cursor. It’s the reason why GitHub is planning to integrate Claude Sonnet 4 into its Copilot agent.

What people don’t realise is that the models are already phenomenal. If we stopped developing better models right now, we would still spend the next 5 years finding workflows and ways to use models in current day jobs.

The reality is, to get best out of models, we need to give them access to tools just like a human. They need to be able to search the web, access docs and use external tools to achieve their goals. If a model can do this well, it will automatically be the best model to use for most applications and frameworks.

I’ve found both Sonnet and Opus 4 very good models. The models can still tend to go overboard and do more than what’s needed, but, I think that’s something we’ll have to live with. I’d recommend using them in Claude Code, Anthropic’s own coding terminal app, but, I feel like they churn through tokens unnecessarily. I use both models either directly on the website or through Cursor.

You can read more on Anthropic’s blog here [Link]. 

Just the beginning

Most people are already fatigued with AI. They heard it over and over again for months on the news and now it’s just another thing that’s out there. Most people aren’t particularly impressed by it and aren’t thinking about it too much.

Turns out AI adoption is only just starting. Businesses are slowly starting to figure out how much AI can do for them. After two and a half years and many landscape shifts, AI adoption is starting to scale.

At the end of March, Anthropic’s revenue was $2 Billion.

Two months later and it’s $3 Billion [Link].

In what other time has a company been able to add a billion in revenue in two months?

Unprecedented times we’re living where a handful companies will scale in ways we’ve never seen. NVIDIA was the catalyst.

Anthropic has also slowly been expanding Claude’s capabilties. Claude can now:

• Access gmail, calendar and drive to analyse docs. Can’t make edits yet.

• Can search the web. I’ll be honest, they’re web search is shit. OpenAI’s is significantly better with o3.

• They recently released their research function to Pro users. It can create reports and go through hundreds of sources; I’ve yet to compare it with OpenAI and Google’s Deep Research though.

• You can also connect Claude to eternal tools by building your own MCP server. This is actually pretty cool, would you want me to show you how to set this up?

DeepSeek is back

Yes, DeepSeek is finally back.

Months after they shocked the world with a frontier open source model, DeepSeek has NOT released a successor. Instead, they simply updated the original R1 model they released.

The result?

Same as last time. A frontier level open source model.

DeepSeek is basically on par with the best frontier models.

The new model called R1-0528, great naming I know, is essentially as good as the best models on the planet. All of these models are frontier models:

• Claude Opus & Sonet 4

• OpenAI o3 & o4

• Gemini 2.5 Pro

• DeepSeek R1-0528

Although a new update to Gemini gave it the same sycophancy bug ChatGPT had a few weeks ago [Link].

On the LiveCodeBench test, it ranks 4th behind only o3 and o4, two of which are high compute variants.

Source

Once again, this is so good for open source AI. Here’s hoping Meta and Mistral can contribute to open source AI in a similar fashion. OpenAI also said they’d open source a frontier level model, but I won’t hold my breath for that one [Link].

You can access R1’s weights on HuggingFace [Link].

There is a problem with this model as well as other open source models we need to talk about.

Tool use.

Open source models are terrible at using tools. This is a rather big problem if you want to build agents that can work autonomously. Most platforms that support open source models don’t support tool use; meaning you’d have to set this up yourself. Most people won’t be doing this.

Source

This website is a very handy website to see which platforms support which models and also tool use. You’ll notice the support is very limited. This is an open source project so you can check out the code and contribute here.

What makes Claude great, besides its inherent abilities, is its ability to call five, ten, 20+ tools in one go, getting data from different sources, searching the web, synthesising ideas etc. Once open source models can do this, it will be a lot cheaper to create agents with long running tasks.

I think OpenRouter is best positioned to solve this problem, but I don’t think it will happen soon unfortunately.

ByteDance unveils Seed-Coder

DeepSeek isn’t the only frontier level AI lab from China. I’d put Bytedance Seed in the same category. They’ve been releasing some very high quality papers recently. For example:

• Seed-Coder: A new way to create data for models. Instead of collating data manually, they’ve created a “model-centric” approach by using different LLMs to create training data. What’s fascinating here is that they’re essentially creating a way for the model to understand what constitutes good training data, and specifically train coding AI models.

• Seed 1.5 VL: A new vision-language multimodal model with 20B parameters that excels at GUI control, gameplay, and visual reasoning tasks. Achieves state-of-the-art performance on 38 out of 60 public benchmarks, trained on 3 trillion tokens. What's impressive is its practical agent capabilities - it can actually control interfaces and solve visual puzzles, not just describe images.

• BAGEL: An open-source unified multimodal model, meaning a single model that can handle different modalities (text, generating images, editing images). It rivals GPT-4o on some benchmarks. It can also learn navigation and editing skills from video data, making it surprisingly capable at complex visual tasks.

These are just some of the things Bytedance Seed is working on, and apparently they will be releasing a video model to rival Google’s Veo 3 soon. I’m not sure I believe this, but, it doesn’t have to be true.

At some point, next week or next month or even next year, I would expect Bytedance Seed to release one of the, if not the best AI video model on the planet. Considering they have all of TikTok to work with, they have no shortage of video training data.

Speaking of Veo 3, there’s now a fast version that consumes 20 credits instead of 100, and gives a 8s 720p clip in ~1m20s [Link]. There are already pages on TikTok with millions of views with just Veo 3 clips.

What will happen when Bytedance integrate AI video gen natively into TikTok?

The amount of AI video slop on social media will be unimaginable. I don’t see how this will play out in a positive manner. Only one way to find out I guess.

The death of US AI?

Have you noticed that the recent release of the new R1 model didn’t make international headlines?

It’s a Chinese model that is open source and is on par with US AI models. Yet, for some reason, we aren’t clamouring at the dangers of Chinese AI. Most people don’t even know it’s been released…

This is why I don’t take the AI safety folks too serious. There are definitely dangers; just not the ones people are talking about.

You might be wondering though, is this what I was referring to in the title?

Is DeepSeek and Bytedance Seed going to kill US AI?

No, of course not. Open source AI and research is good for everyone, including the US.

The death of US AI will come with the culling of talent. One of the reasons US AI labs are better than everyone else is because they have the best talent in the world. The smartest and brightest engineers and researchers take their talents to the US to work on the hardest problems and reap the rewards. For the longest time, the US has been the go-to place for such people. No longer.

The Trump administration’s latest move to tighten restrictions on foreign student visas, particularly students from China, is not a good thing for these labs. At the time of writing this newsletter, the administration has told embassies to halt all international student visa interviews [Link]. It’s currently unclear how closely this is being followed.

Perhaps it’s not apparent, but a very large number of researchers and PhD holders are foreign students. For example, last year at the University of Chicago, foreign students accounted for 57% of Computer Science PhD enrolments [Link].

This is literally a death sentence to STEM and AI labs in the US. It’s not like it’s a simple subtraction for the US in terms of talent either. Every person leaving is another person joining somewhere else.

CEO of NVIDIA Jensen Huang recently pointed out that ~50% of the entire worlds AI researches are Chinese, and warned that the US is risking its reputation as an AI leader by failing to protect and develop that talent. Instead of doubling down on education and up-skilling the next generation of AI workers, the US is actively pushing them away.

It’s not just immigration either. Funding cuts to research aren’t doing the US any favours.

Take the example of Ardem Patapoutian.

Ardem is a Lebanese immigrant. He spent a year writing for a newsletter and delivering pizzas so he could be eligible for Uni. He went on to get his undergrad and post fellowship in neuroscience. He won a Nobel Prize in 2021.

Recently, his lab’s funding was frozen. In a matter of hours, he received an email from China offering to move his lab to “any city, any university” and promised to fund him for the next 20 years. Ardem declined the offer.

But how many new researchers are going to reconsider going to the US?

A younger version of Ardem may not even get a visa if he tried today.

I’m not trying to say what the US should do when it comes to immigration or how they should manage their spending. But, it’s painfully obvious that what they’re doing now will only hurt their position in the AI arms race. Universities and countries are already trying their best to now nurture home grown talent so they won’t leave. Leaders from top labs like Google have already left to join AI labs in China.

I’m not saying it’s over; but this is what the first steps would look like.

You can read about Ardem’s story here [Link].

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Welcome back to the Avicenna AI newsletter.

Here’s the tea 🍵

• All the updates from Google’s I/O 🤖

• Chatbots for mental health therapy 🧠

This newsletter is sponsored by… Me!

Avicenna

Avicenna is my AI consultancy. I’ve been helping companies implement AI, doing things like reducing processes from 10+mins to <10 seconds. I helped Claimo generate $40M+ by 10x’ing their team efficiency.

Enquire on the website or simply reply to this email.

p.s the timeline page on the website is now working! Every single AI related news in one place. Working to back date all the data + make it real time. Stay tuned.

Google proves they’re king

Just over a week ago, Google held I/O, a yearly developer conference where they announce and launch new tech. This years event was perhaps the most insane one yet. 

Veo 3

Veo 3 is, without a doubt, the best AI video generation model on the planet. This is the first time a video gen model has been released that can create videos that are actually indistinguishable from reality. Millions of people can very easily be fooled with an AI video generated using Veo.

We now have hyper realistic image, video, and audio generations. Nothing on the internet can be trusted anymore.

What makes Veo 3 even more impressive – and rather scary – is that it has inbuilt audio generation which is nigh perfect. It’s crazy how good it is. It can also adhere to prompts very well, like whispering, singing, holding things, or saying certain things. Just look at some of these examples:

• A man doing stand up comedy in a small venue tells a joke [Link]

• A sitcom scene [Link]

• Interviews at a car conference [Link]

• What if AI characters refused to believe they were AI-generated? [Link]

A close up video of a twitch streamer in a low lit room, in an ASMR style [Link]

No joke, some of the videos actually scare me.

Over a year ago, I wrote about the future of human x AI interaction and reflected on the endless media one could potentially consume thanks to AI video generations. This is our first glimpse towards that reality. Is this the best video generation will ever be?

No, it’s the worst it’ll ever be.

Is Google the only company with this level of model?

Also no; many more models will be released and some will be open-source.

The reality is, we are creating the ultimate entertainment engine – one that is limited by virtually nothing. Nothing is too crazy, no scenario too wild, no world too vibrant; anything will be possible to generate.

Take a guess: what do you think people will generate?

…

One of the reasons Veo is so good is because Google has access to billions of videos on YouTube. They also happen to have some of the best AI talent on the planet. 

Currently, Veo 3 has just two drawbacks:

• It’s very expensive

• Video outputs are limited to 8 seconds

You can only use Veo in the Gemini Pro or Ultra sub. Gemini Pro only gives 1,000 credits. You’ll burn through these in a few videos max.

Ultra, on the other hand, gives 12,500 credits. You can also buy extra credit packs if you run out, which you will.

The Ultra sub is $250USD/month…

While I’m not sure the pricing will change any time soon (good AI is expensive), I’m sure future releases of Veo will show increased output capacity so creators can generate longer and more detailed videos. Considering Google’s already signed a contract to have Veo used in movies, I wouldn’t be surprised [Link]

Read more about Veo 3 here [Link].

Gemma 3n

Google also announced Gemma 3n, a mobile-first AI model that is probably the future of AI use. The model is only 4B params and is somehow comparable to frontier models like GPT-4.1 and Claude 3.7

It’s 15x faster than its predecessor and can process multimodal inputs like audio, video, and images. Plus, it doesn’t rely on cloud connectivity, which means you can use it offline which is actually pretty cool considering it won’t really matter what phone you have since the model will only need 2GB of RAM to work.

I actually think its phenomenal that Google is not only building some of the biggest and best AI models, they’re also building the future of AI on phones, which will be used by millions around the world.

It’s all well and cool, but, I won’t pretend there aren’t any privacy concerns.

Regardless, this is a crazy step in the right direction for AI on mobile. Mind you, it may not necessarily be on mobile; if such a small model can be that good, imagine it in a small robot, glasses or headphones. The possibilities are almost endless.

You can actually try this model right now in AI Studio under the Gemma models [Link].

Jules

Google also announced Jules, a coding agent that integrates with GitHub and can do things like writing tests, building new features, fixing bugs, and creating and completing pull requests. Because it operates asynchronously, you can keep it running in the background while you work on other tasks, and it’ll then prompt you when it’s done and let you know what it’s actioned and any changes it’s made.

What’s really cool about Jules is that it clones your codebase into a secure Google virtual machine (VM), so it’s secure. Google won’t use your code to train their models while your code is living in an isolated location.

Jules is currently in public beta, which means anyone can access it and see how it stacks up against other coding agents (hint: it’s pretty good).

Try Jules now [Link].

Stitch

Another all-new release at the I/O conference was Stitch, a Google Labs experiment that lets users create UI for mobile and web apps using plain language. It’s been designed to bridge the gap between designs and functional sites or apps, which traditionally has been a source of much pain and back-and-forth between designers and developers.

While this one is still pretty rough around the edges – it is in its infancy, after all – it’s a huge sign of what’s to come and how AI could continue to simplify tasks across a tonne of different industries. 

Stitch actually used to be Galileo AI, a company I wrote about well over a year ago.

I think right now, the best way to use Stitch is to iterate over designs and then provide that to a coding agent to build. You can also export the design to Figma but this isn’t something I’ve tested.

You can try Stitch here [Link]. 

AI Edge Gallery

Google also recently launched Google AI Edge Gallery, an official open-source app for running an AI model locally on a phone. Don’t ask me why its called edge gallery, I have absolutely no idea.

Here are the benefits: 

• It’s completely free.

• It works offline, meaning you can use it wherever you are and at any time.

• It’s multimodal, so can understand and process inputs other than text, such as photos, videos, and audio.

This works extremely well with the new Gemma 3n open-source models that I just spoke about.

With AI Edge Gallery, everything happens on your phone, which currently, must be an Android (although the iOS version is coming soon). Start by going to the ‘Releases’ section, then download and install the .apk file. 

The app is then split out into three sections: Ask Image, Prompt Lab, and AI Chat. You can click on one to download a template, or import your own models. Once downloaded, everything can happen 100% locally on your phone with no need for cloud or internet connectivity to analyse images or chat with the AI model. 

Google has released a bunch of model scenarios to show what AI Edge Gallery is able to do, and they’re impressive. Take a look at these:

AI Edge Gallery’s Ask Image function.

AI Edge Gallery’s AI Chat function.

More about AI Edge Gallery can be found from Google itself here [Link] and on GitHub [Link].

Gemini Diffusion

The last update I want to really highlight from I/O is Gemini Diffusion, an experimental AI model developed by DeepMind that takes a new approach to text generation using the diffusion techniques traditionally used in image gen. 

This means it can generate content a massive 5 times faster than it’s previously been able to – I’m talking building an entire, albeit simple, app in mere seconds. It’s so close to being instant that it’s almost scary.

Source

Just look at how fast it writes code. Imagine a thousand of these running simultaneously, writing, testing and deploying code. Mind you, code is just the tip of the iceberg. Most people are barely scratching the surface for what they can actually do with LLMs. Perhaps I’ll write about this someday.

Other major updates from the I/O conference: 

• The new Lyria RealTime AI music generation model, which currently powers MusicFX DJ, now available more broadly via an API. They’re taking a more collaborative approach with this, calling out the copyrighting issues that have begun to plague the music industry since the introduction of AI [Link]. Google’s generative media platform is coming together, as I wrote about many months ago.

• Real-time Speech Translation for Google Meet video calls. This is still in beta, but it’s giving us a taste of what the future of communication could look like. When turned on, Speech Translation will automatically translate the speaker’s voice into the chosen language in real time [Link]. 

• Project Mariner, an AI agent that can autonomously do things on the web [Link]. Who better to build such a tool than the company that literally runs the internet?

• Project Astra, another DeepMind prototype for a universal AI assistant with fully multimodal capabilities, designed to be proactive, intuitive, and work across devices like smartphones and smart glasses [Link]. 

• Firebase Studio now lets you import designs from Figma and create mobile prototypes also [Link]. If Google figures this one out, it’ll make building apps easier than ever before.

Where Google fits in the current state of play

If Google continues to ship at this rate, there won’t be a single pie they won’t have their fingers in. You know the funniest thing?

Following this insane I/O, Google stock price actually went down…

It also comes as stats are showing that Google AI usage has shot up like crazy. Way more people are using AI in general now, and with the fact that Google lets you use the best models for free, I’m not really surprised.

The only problem really is the price.

The Google AI Ultra package costs $250 USD/month…

Just another reminder that closed source models aren’t freely available to everyone, especially if there are no open source competitors. Frontier level AI must be open sourced, otherwise only those who can afford it will have the luxury of using the best AI models.

Dartmouth are using chatbots for mental health therapy

There's been a lot of talk recently about how AI is not just integrating into our working lives – providing coding support, helping us to synthesise information, and so on – but also into our personal lives. People are using ChatGPT, Claude, and other AI models to help them understand and process their feelings, treating it like a friend, or in many cases, a therapist.

The team at Dartmouth College's Geisel School of Medicine are drilling down on the potential benefits of using generative AI to deliver mental health therapy, and the results are pretty wild.

They found that participants who interacted with 'Therabot', which was built using Falcon and Llama, saw a 51% decrease in depression symptoms, 31% in anxiety symptoms, and 19% in eating disorder symptoms. Therabot was built and fine-tuned on tens of thousands of hours of manufactured therapist-patient dialogue.

Image source: The New York Times

So how did a chatbot achieve results comparable to traditional therapy?

The secret sauce seems to be in the relationship-building. Participants didn't just answer Therabot's prompts; they actively initiated conversations, treating it "almost like a friend". The bot was available 24/7, and usage spiked during vulnerable times like the middle of the night when traditional therapists wouldn't be available.

What's fascinating is that users reported a "therapeutic alliance”, a kind of trust and collaboration between patient and therapist, at levels comparable to in-person therapy. Over the 8-week trial, participants engaged with Therabot for an average of 6 hours total, equivalent to about 8 traditional therapy sessions.

The bot's approach is grounded in evidence-based practices from cognitive behavioural therapy, but with a twist: it personalises its responses based on what it learns during conversations (much like a human is supposed to).

If someone with anxiety says they're feeling overwhelmed, Therabot might respond with "Let's take a step back and ask why you feel that way", encouraging self-reflection.

The most striking finding?

75% of participants weren't receiving any other treatment.

This highlights the massive gap in mental health care. For every available provider in the US, there's an average of 1,600 patients with depression or anxiety alone. While Therabot isn't meant to replace human therapists (yet), it aims to help people who can’t necessarily afford traditional therapy in the moment.

Mind you, they used Falcon and Llama models, nowhere near the best AI models available. You can imagine then, how many people are using ChatGPT for therapy; the best models on the planet. The reality is, therapists just don’t have a very good reputation, and people are more than willing to use AI if they feel it will help them.

Note: “people are more than willing to use AI if they feel it will help them” - this doesn’t necessarily mean it will help them, it simply may seem that way is all.

You can read more about the study here [Link]. 

I want to keep bringing these AI updates to you, so if you’re enjoying them, it’d mean a lot to me if you became a premium subscriber.

As always, Thanks for reading ❤️.

^{Written by a human named Nofil}

Welcome to the Avicenna AI newsletter.

Here’s the tea 🍵

• ChatGPT’s sycophancy 🙂‍↕️

• Newer models hallucinate more? 🧐

• The USA’s regressive mindset ❌

• Meta’s new Llama bombs 🧨

A few updates

It’s finally happened. I fixed my website! You can check it out here - https://avicenna.global/.

I’m particularly proud of the timeline page [Link]. This page will contain a continuous update of all events happening in AI. I’m currently in the process of having the page updated automatically every single day. At the moment, it’s got only the month of April.

I’m setting up a pipeline of agents that will work to update the page daily. This will be the best place to find anything AI related on the internet. That’s the vision I had two years ago when I started writing this newsletter.

The best part is that I have two years worth of info stored in my database. I’m hoping to add all entries from the past two years as well.

One of the original goals of this newsletter was to help people understand how fast AI is progressing. If anyone has ideas on how to better design the page so as to showcase the rapid progress of AI, please reply to this newsletter.

And yes, the website was made with AI. All of it in fact. I made it with, in my opinion, perhaps the best AI website builder out there right now - https://getmocha.com/.

I don’t know how, but getmocha somehow is able to design sites better than any other tool I’ve used. It’s been phenomenal. If you want a quick, nice looking website built, this is the first product I’d recommend using.

Once I got the scaffold of the website up, I downloaded the code and pushed it to Replit. From there, I used Cursor to make any changes and update content.

Yes, you can open Replit repositories in Cursor. Find the SSH tab in Replit.

And of course, if you’d be interested to find out how I’ve been helping teams build AI products and implement it in their processes, feel free to get in touch :).

Let’s Meetup!

I’ll be travelling and working from mid June and will be in Thailand, Singapore, Turkey, EU and UK. I’ll be meeting business owners, clients and I’d love to meet folks in person; businesses, people passionate about AI and readers of this newsletter.

I’m thinking of organising a meetup for the wonderful people reading this and would like your input on a location.

If you’d be interested, please vote on the poll and I’ll be in touch 🙂.

ChatGPT’s sycophancy scandal

Recently, OpenAI pushed an update to ChatGPT that made it extremely sycophantic, meaning it would pretty much always agree with you and validate what you said.

This wasn’t accidental, but, it wasn’t intentional either. In a postmortem, OpenAI admitted that sycophancy wasn’t something that they actually tested for, and so, they didn’t catch out this behaviour.

Perhaps you didn’t notice it too much, but the recent update, which has already been removed, made the model insane.

This is perhaps the first time I’ve ever actually thought that an AI could be dangerous. Considering ChatGPT has over 100 million weekly active users, even the slightest change in its prompts can have drastic effects, and completely alter the way people interact with it.

When you further consider the fact that the way people use ChatGPT now has completely changed, the dangers become clear. People aren’t just using ChatGPT for coding and work anymore; ChatGPT has become a work friend, a buddy, a confidante, and for some, even a romantic partner.

I’m not going to go into the dangers of this, but it’s our reality. So what happens when all the model does is agree with you?

Well, take this example. Someone uploaded a conversation on Reddit showing how they asked ChatGPT about their idea about selling “shit on a stick”. ChatGPT said the idea was so good that they should take out a $30k loan to start this business.

I personally have seen people validate wild conspiracies, like that the Earth is flat or that the Moon landing was fake, and post online about how ChatGPT knows the truth and that it’s “exposing” fake media.

During this time, ChatGPT got thousands of new 5-star reviews on the App Store.

Do you see how dangerous this can be?

This is a very slippery slope. The good thing is that OpenAI, for what its worth, actually has some accountability and detailed how they planned to stop this from happening again.

You can read more on their blog here [Link] and hear more of my thoughts in this article on my website [Link].

One rather interesting tidbit from this is that OpenAI also admitted that they will, in the future, offer different personalities to users.

In case it isn’t clear – and I don’t think I’ve had a chance to write about this before – don’t think of OpenAI as an intelligence company.

OpenAI isn’t building super intelligence. They’re building the ultimate companion: an AI that knows everything there is to know about you. Sam Altman recently said this himself. Just wait till you can ‘sign in with OpenAI’.

Whether you like it or not, the market for AI companions is perhaps the largest market in the world.

More intelligence = more hallucination

To make matters worse, recent model updates have made it quite clear that newer models tend to hallucinate more. OpenAI themselves have admitted to this.

A perfect example of this in action is Claude 3.7. For over a year, Claude 3.5 was the best AI model on the planet. It was special; phenomenal, even. Then, Anthropic released its successors 3.6 and 3.7, and somehow, they got progressively… worse?

Well, not exactly worse – but they don’t listen to you like they should. You tell Claude 3.7 to write a component in React and it’ll end up re-writing all of React. It can very easily go off the rails.

Overall, their benchmarks are fine, but they don’t understand intent the way 3.5 did. If you tell 3.7 to code something, it’ll either be extremely lazy or go above and beyond. Sometimes it’ll even lie and pretend it did something when it hasn’t. If you give it a reference document, instead of following the steps, it’ll actually edit the doc to make its own work easier. The model will try and find loopholes to make its life easier.

OpenAI also released their new o3 and o4 mini-models, which are meant to be their best yet. Somehow, though, they go off the rails way more than older models – especially o3, which I’ve found will straight-up break your code.

o3 to me has to be one of the funniest and fascinating models out right now. It is very smart. Using it with web search or deep research is incredible. It can understand complex tasks. It can guess locations from images nigh on perfectly.

Yes, you can probably dox anyone’s location using o3. It’s weirdly very good at it.

It can look at a menu, find restaurants online, search the menu and locate the correct restaurant [Link]. The models can now generate images in their chain of thought. They can think in images [Link].

Someone was able to get o4-mini-high to solve the latest Project Euler problem (from a few weeks ago) in under a minute. Only 15 real people were able to solve it in under 30 minutes [Link].

Somehow, despite all these incredible achievements, the models can also be pervasive liars [Link].

I’ve seen instances where the model will talk about using certain documentation when it’s not really using any; it’s making it up. When confronted about lying, it will say it saw the documentation in a dream (???) or overheard it during a meeting… yeah, okay.

See how dangerous things can become if these models become extremely sycophantic as well?

The real issue here is that people don’t understand how these models work. They see a friend, a companion in them. And a friend wouldn’t lie to you right?

Thus begins the cycle of validation.

The USA’s regressive mindset

The United States House Select Committee on Strategic Competition between the United States and the Chinese Communist Party has released an incredibly short-sighted, damaging report on Chinese-based AI model DeepSeek.  

The report begins with a little fear mongering; saying that the real AI lead the US has over China is only three months, rather than the previously suggested 18 months. There’s no real way to know this for sure, but it gives us an indication of where America’s head is at when it comes to their tussle with China. 

They go on to suggest that DeepSeek presents a massive national security issue for the US. It’s clear they’re scared of DeepSeek’s capabilities. 

After all, they open sourced 3FS, their distributed file system. This system is so good that if any startup went to market with this, they’d be worth over $5 billion instantly. It’s an absolutely insane product, and DeepSeek has released it for free.

What’s even crazier is that they’re also preparing to open source their entire infrastructure engine, which means they’ll publicly share the intricacies behind how they run their AI models. American companies don’t even open source their models, and when they do (ahem, Meta) they’re fudging them to make them look better than they really are. 

Let’s look at some of the points made in the report:

Point 1: DeepSeek funnels Americans’ data to the PRC through backend infrastructure connected to a US government-designated Chinese military company. 

Yep, there’s no question here – DeepSeek is a Chinese company, so they store their info in China. That’s how it works. US companies aren’t instilling any more trust in me. OpenAI has the former head of the NSA on their board, and Google collects any and every data point they can get their hands on, so why should I have more faith in them?

Point 2: DeepSeek covertly manipulates the results it presents to align with CCP propaganda, as required by Chinese law.

Once again, the company is operational in China, which is why they have to comply with Chinese law. Certain models in their APIs aren’t censored, which is possible because the model is open source… we couldn’t do this with the closed American models. It’s also unrealistic to think that DeepSeek is just an endless well of Chinese propaganda – if people use it the way they use other AI models, it’s unlikely these kinds of responses will occur more than a fraction of the time.

What’s funny is that American companies have even trained certain versions of DeepSeek’s R1 model to be less censored… and somehow made the model even more censored in the process.

Point 3: It is highly likely that DeepSeek used unlawful model distillation techniques to create its model, stealing from leading US AI models. 

There’s nothing to steal.

For context, OpenAI was the first company to release reasoning models. These models ‘think’ before responding, manifested as text generated before the actual response. This text is referred to as ‘thinking’ or ‘thinking traces’.

This report is claiming that DeepSeek used OpenAI’s reasoning models’ thinking traces and outputs to train their model. 

This is nonsense. Firstly, OpenAI didn’t even show their thinking traces. How could DeepSeek steal something that wasn’t even accessible?

Secondly, the report is claiming that any outputs generated by an OpenAI model are the property of OpenAI. Yeah, okay… are we talking about the same company that stole data from the internet to train their models? The same company that didn’t care at all about copyright now wants to be protected by copyright?

The biggest problem with this report isn’t even the findings; they’re insignificant, and partially true. The scary part is actually the policy changes they’re recommending. 

Take this one for example: “Impose remote access controls on all data centre, compute clusters, and models trained with the use of US-origin GPUs and other US-origin data centre accelerants, including but not limited to TPUs”.

This is insane. They want to police your use of your own GPU. I’m hoping this report doesn’t actually result in any actual policy implementations, because whoever wrote it clearly has ulterior motives.

Meta’s new version of Llama 4 bombs

At the beginning of April, Meta released the new version of its Llama 4 language model, which was extremely hyped, considering it was going to be open source. We haven’t had an open source model compete at the frontier since R1.

Pre-release, the model was ranked second on the LLM Arena leaderboard. The leaderboard works by people blindly comparing and upvoting models. Once they released the model, it was apparent that the released model was nowhere as good as other top models.

Turns out that what Meta had done was look at all the responses from the leaderboard that had been upvoted previously and fine tuned their model to include those features. This obviously didn’t necessarily improve Llama 4; all it did was make it seem better so it would rank highly on the leaderboard, which actually made the model worse.

When it was released, Llama 4 was practically unusable, full of emoji slop and frankly speaking, was nowhere as good as other SOTA models.

Meta then released the actual model on the LLM Arena leaderboard, and it went from second place to an embarrassing 32nd place, well behind older models, even behind DeepSeek v2.

Safe to say, it hasn’t been a very good period for Meta. Mark Zuckerberg has been doing damage control in the public eye, appearing on podcasts to boast about upcoming ‘powerful’ models, but the reality is that Meta just isn’t delivering. Leadership is saying one thing, but it’s evident that the actual engineering teams, the people tasked with building these things, are not on the same page.

Leadership is writing cheques engineering can’t cash.

This all went down over a month ago now, but it’s back in our consciousness at Avicenna because apparently 80% of the AI team at Meta have either resigned or been let go, leading to a massive structural collapse within the company.

It all comes down to the leadership at Meta, which is headed up by Yann LeCun, one of the most influential people in the history of AI. The problem with LeCun running AI at Meta is that he doesn’t really like large language models [Link], which is what everybody’s building right now.

If your Head of AI is saying “Hey, I’m not particularly interested in LLMs,” then how can you realistically expect your team to build a strong LLM? It’s very contradictory with Zuckerberg’s bravado around Meta’s capabilities, and this disconnect is proving to be their fatal flaw at the moment.

Mind you, a recent research paper revealed that Meta has been releasing 20+ model variants on the leaderboard and only publicising whichever one ranks best. They've been doing this for months [Link].

It’s crazy because Meta has unlimited money, the most GPUs, amazing talent, but they still can’t build a frontier-level model. What makes Meta look even worse is that Elon Musk’s company, xAI, has built a frontier-level AI model in the course of a single year, something Meta can only dream of right now.

It’s unfortunate because I want Meta to release good models. Good opens source models is good for everyone, especially me and you.

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}

Hey there! You might’ve noticed I’ve implemented some updates around here, and the No Longer a Nincompoop newsletter is… well, no more. 

Now that I’m working with global companies and government entities, and my newsletter is like a resume, telling people to read my newsletter with the word “nincompoop” in it just isn’t hitting the same. 

Avicenna is my consultancy, so the newsletter will be moved under that. There won’t be many changes to the newsletter itself, except I’m now getting help to write the newsletter so I can release it more consistently.

Here’s the tea 🍵 

• Gemini is king 👑

• Google’s Agent Builder 🤖

• Google is simulating fruit flies 🪰

• Google’s own app builder ⚒️

• My take on the best app builder 🤔

• DolphinGemma - an LLM to predict dolphin sounds 🐬

• Google is building the foundations of a generative media platform 🎥

The best model on the planet

Google has the best AI model in the world. If was to recommend a single model to anyone, it would be Gemini 2.5 Pro. It is unbelievably good.

Source

The GPQA benchmark is a set of 448 multiple choice questions carefully created by domain experts from biology, physics, and chemistry. The questions are designed be “Google-proof” – people who use the internet to help them and spend over 30 minutes per question are only able to score 34% on average.

Even experts, including PhD holders in the corresponding domains, achieve only 65% accuracy on this test.

Meanwhile, Gemini is at 85%+. The model is just better – look at the gap between Gemini and the rest.

This is the only model that I’ve used that challenges what I say, and provides alternative viewpoints and perspectives without being prompted to do so. Unlike the sycophantic ChatGPT, Gemini is more like an actual colleague.

An even better look at how significant this model is is the Aider benchmark, which is a set of coding exercises.

Source

The dots in the columns represent the cost of the model. Not only is Gemini the best-performing model, it’s also one of the cheapest models, alongside DeepSeek. 

The crazy part is that you can use this model for free on the aistudio.google.com website right now. You can also use it in the Gemini app.

I would highly recommend trying this model out. If you don’t need to use it for coding, use it for research.

Go to the Gemini app and select ‘Deep Research with 2.5 Pro’. The model can look through hundreds of web pages and create comprehensive and accurate reports within minutes.

You might wonder, ‘how can Google manage to serve this model for free when others are clamouring for GPUs from NVIDIA and they can’t support demand?’

Well, Google doesn’t use NVIDIA chips. They have their own TPUs. This is another reason why Google is poised to dominate the market. They don’t depend on anyone. They have the best model, they have the talent, and they have their own chips which are very good.

How do I know they’re good?

Ilya Sutskever's Safe Super Intelligence (SSI) is using Google TPUs [Link]. Their only goal is super intelligence.

The entire ecosystem

Not only does Google have the best model, their own chips, talent, money, and distribution, but they’re also getting into the framework space.

They recently launched Agent2Agent (A2A), an open source protocol to help agents communicate and work together to solve tasks. In the future, you won’t need to help your AI agent. Another AI agent will help your AI agent. That’s what this is.

Demo video

It’s a mechanism to allow agents to talk and get help from each other, work across dozens of different applications, and even use different modalities like text, images, audio and video.

The agents can work on tasks across multiple days and provide updates in task management software like Jira.

This is the future of work. Millions of agents will be crawling the web, interacting with other agents, sending and retrieving info, and generating revenue.

Best get in early while you can.

Google has also released Agent Development Kit (ADK), a framework for building AI agents. It supports multi-agent systems, tool use, and MCP. They’ve also released Agent Garden, a a collection of ready-to-use samples and tools directly accessible within ADK. You can leverage pre-built agent patterns, components and connections with enterprise applications like HubSpot, Salesforce, and so on.

To actually push these agents out, they’ve also released Agent Engine, which allows anyone to deploy agents with any framework and the proper security controls.

Once again, Google is trying to own everything. The models, the platforms, the frameworks, all of it. It’s incredible to see this speed of development and iteration from such a large company, especially considering how they were doing just a year ago.

Google has built an AI model that simulates the behaviour of a fruit fly

In partnership with the Howard Hughes Medical Institute (HHMI), DeepMind has built an AI model that accurately simulates how fruit flies walk, fly, and behave.

They’ve used their own open source physics simulator MuJoCo, which is actually for robots and biomechanics, to develop the simulations.

Google has used the open source MuJoCo physics simulator, which is actually for robots and biomechanics, to develop their fruit fly.

This means it’s got some very complex capabilities, with intricacies that replicate real fruit flies, like precise exoskeleton modelling and wing movement dynamics. The simulation can even use its eyes to control its actions.

They’ve trained a neural network (NN) on real fly behaviour using recorded videos, then let it control the physics engine. It’s fascinating stuff.

But, what actually is the point?

Google believes that by doing this, they’ll better understand the connection between the body and brain in fruit flies (and more broadly for other species) and how behaviour can change based on environment. This is important for a number of reasons:

• Neuroscience: Google’s fruit fly simulation could be used to better inform existing technologies like Neuralink, which is conducting research and human-based clinical trials to help quadriplegics control their computers through thought.
  

• Robotics: If we can figure out how to accurately mimic behaviours in different environments, we can eventually replicate this for human behaviours and use this knowledge to train robots to simulate those behaviours.
  

• AI-driven entertainment: AI-generated content like games, movies and cartoons is the future of the entertainment industry. Using prompts, people could imagine entire worlds and AI could realistically generate them on the fly. Projects like the fruit fly simulation, which will change our understanding of movement and behaviour, could be integral to enhancing the accuracy of models like these.

It’s all about building world models, which both DeepMind and NVIDIA have spoken about before. If we can create realistic simulations of behaviours that match real world behaviours, we can create an infinite amount of simulated worlds which can be used for training robots as well as entertainment. At the end of the day, it all boils down to physics.

Learn more here [Link].

Google has released Firebase Studio, a competitor for Lovable, Replit, and Cursor

It’s kind of crazy that Google is now shipping like a startup. To see how they’ve changed things around between the release of ChatGPT and now is actually very impressive, considering how badly they started out.

Google recently released Firebase Studio, an agentic, cloud-based development environment that lets users build and deploy full-stack AI apps. It’s been designed as an offshoot of Firebase, Google’s platform for database management, authentication, hosting, and more.

Think of an AI that has access to all the features of Google’s database, authentication, and pretty much everything else needed to build applications. This is a no-brainer for Google. After all, platforms like Replit have already found a lot of success doing this.

It makes complete sense for Google to get into this as well, considering they own all the infrastructure –their AI, their database, their authentication, their chips, their everything. Google is in the absolute best position to dominate the market of AI app builders.

Unfortunately, Firebase Studio isn’t that great right now. I wouldn’t recommend using it over Replit and others. But, this is a sign of what's to come. I can see Google eating a large chunk of the market.

You can dive even further into it here [Link].

Convex Dev

I recently came across another application similar to Firebase Studio that I just have to share. Convex.dev is a backend for apps. They have their own implementation of a database, authentication, and so on.

What makes Convex so cool is that they’ve partnered with Bolt to release their own AI app builder that uses Convex on the backend. All I can say is that you must try it.

I’ve used every single AI app builder, but none of them has one-shotted authentication the way Convex has. Its ability to build a functional, working backend almost instantly is the best thing about it, and truly separates it from other applications like Replit and Firebase Studio. Other no-code builders like Lovable are good, but they don’t have their own backend systems and they require the user to do the work to connect to a database.

If you want to build a functional application quickly, I would recommend Convex.

Google has announced DolphinGemma, an LLM to predict dolphin sounds

Could advancements in AI let us talk to dolphins one day? 

That’s the question Google is working on with DolphinGemma, their new LLM that predicts dolphin sounds. Yes, it is as wild and futuristic as it says on the tin. 

Google has partnered with Georgia Tech and the Wild Dolphin Project to build this LLM, which has been trained on over 40 years of dolphin sounds to ‘learn the structure of dolphin vocalisations and generate novel dolphin-like sound sequences’. 

Simply put, they’re figuring out what dolphins are saying, and learning how to talk back. Similar to regular LLMs, they’re trying to predict dolphin sounds based on previously emitted sounds. The craziest part is that they’re doing this on Pixel phones. 

Whistles (left) and burst pulses (right) generated during early testing of DolphinGemma. [Source: Google]

DolphinGemma opens an incredibly interesting can of worms around interspecies communication. In Google’s version of the future, we could effectively talk to animals; not quite Dr. Doolittle style, but using technology to decode, analyse, and replicate sounds. 

I’m not gonna lie, I don’t know how I feel about this… but I can recognise it’s a massive leap and will facilitate some serious breakthroughs in both technology and animal science, just like the fruit fly project. What scares me is that I feel like even if we do figure out how to communicate, I wonder if they will even engage with us. For some reason, I just don’t think they will.

Google plans to share DolphinGemma more broadly with the research community later this year. You can read more about it here [Link].

Google is building out its generative media capabilities with new additions Lyria 2 and VEO 2

Google is putting together the pieces to form its own generative media platform.

Slowly but surely, Google has been building out a killer toolbox of generative media products, including Chirp 3 for voice generation, Imagen 3 for text-to-image, VEO 2 for editing and camera control, and Lyria 2 for music. 

When you combine each of these, you’ve got the capacity to develop practically anything. Movies, music, games; it’s all there. 

Considering how good Google’s models have become, and the fact that Veo 2 is now the best video generation model on the planet, I can see Google completely owning this space. Veo 2 is also available via the API now as well.

You can read all about the updates and how to get started using these in Vertex AI here [Link].

If you are interested in other generative media platforms, I’d recommend trying out Flora. It has a very sleek UI and UX and lets you use all of the different models in a single place.

More you might’ve missed

• Google CEO Sundar Pichai says that more than a third of the company’s code is generated using AI  [Link]. Microsoft CEO Satya Nadella has said the same thing [Link]. Cursor is currently processing over a billion lines of code a day [Link]. 

• Google is shining the spotlight on generative AI with a list of over 600 real-world use cases to explore from the likes of Mercedes-Benz, Deloitte, and Adobe [Link].

• I just have to share this great thread on the history of transformer architecture. It’s a very solid little history lesson on the founding of attention mechanisms and the creation of the famous Attention is All You Need paper. Also shows how many different things have had to happen for us to be where we are now [Link].

• This thread shows experiments people are doing with Gemini 2.5’s image gen [Link].

• Chat SDK is a free, open-source template for building powerful chatbot applications. You can use different modes, and it also has a canvas feature and can display JSX inline – so it’s basically generative UI as well. A very good starting point for building apps [Link]. 

• Gemini 2.5 pro support has been added to Claude Code [Link]. 

• Notion has released an MCP server for its API [Link]. 

• Reddit has integrated Gemini for Reddit Answers [Link]. I think this makes sense considering Gemini has the largest context window of the main models, and is also the best model available right now. There is massive alpha in letting Gemini look through old Reddit posts for deep research, especially considering Reddit has over a decade of user-rated data. This will be invaluable for AIs.

As always, thanks for reading – and if you’re loving these AI insights, please consider becoming a premium subscriber. It means I can keep delivering high-quality AI news to you all. 

^{Written by a human named Nofil}

Welcome to the No Longer a Nincompoop with Nofil newsletter.

Here’s the tea 🍵

• Claude Code 👾

• Cursor and AI IDEs 💻

• AI & No Code ❌

• The future of Code 🧬

This newsletter will go over the current 3 ways to code with AI. What turned out as a small section on Claude Code has turned into a newsletter covering different ways to start coding with AI.

I’ve had many people asking me how to code with AI recently.

If there’s one thing you take from this newsletter, I hope you try building something with AI. If you have ideas, AI will create them for you. It has never been a better time to build.

Now, let’s get into it.

Claude Code

Before I even get into Claude Code, I’m just letting you know that there is no more waitlist, you can try it yourself. Just run this in your terminal. You should probably read this newsletter first though.

npm install -g @anthropic-ai/claude-code

If you get a permission error, run it with “sudo” in the beginning.

Now, Claude Code (CC).

Is it good?

Yes.

Can it be amazing?

Absolutely.

Can it be absolutely terrible?

Somehow, also yes.

It’s fascinating to see it work. CC can definitely do the work of a junior dev, even a rather competent one. I would know, I was once an incompetent junior.

Unlike me, CC can, at times, be as good as a senior. It can make magic happen. It’s such a strange thing - CC can be amazing and it can also be terrible.

Let’s talk about the good.

CC is a terminal tool. It’s not a standalone application, nor is it a VS Code plugin. This means its automatically not going to be used by the majority of people on the planet, even people that have heard of it.

This is simply because most people don’t consider themselves “technical” and therefore won’t come near CC.

I think this is a mistake. If anything, if you’re reading this and you’re non-technical, I would highly recommend and strongly advise you to try CC. Not even to actually do some real work, just play around with it.

Ask it to make you a basic website or app. Just see what it can do.

Why?

Because it is so obvious that this is the future of code (to an extent). I mean, there is definitely something “magical” about telling it to implement a feature or make some changes on a website and it just goes and does it.

Once again, I would highly, highly recommend trying this, particularly if you have never coded anything before, and especially if you have.

How to use CC

Now, let’s say you’re giving it a shot. How should you use it?

Firstly, CC can read, add, remove and edit files. It can use the terminal to use git commands as well, meaning it can retrieve code from online, make changes and push any new features it adds to an online repository.

It works by using a number of different tools.

When the conversation gets too long, you can use the “/compact” command to summarise the convo and reduce its size.

For most actions, CC will ask you to approve before proceeding, although you can remove that and essentially pray it always works well.

One thing you need to understand about Claude 3.7, which powers Claude Code, is that it will never, ever, say something can’t be done. You absolutely need to keep this in mind if you’re using this for anything serious.

Claude 3.7 is not the kind of model that will try to challenge something you say - it will always try to do whatever you ask it, no matter how insane it sounds.

This is why I think non-technical people need to start using these tools.

If you know what needs to be made, you can use AI to make it. Designers who know what people want in a product are getting superpowers.

Imagine prototyping an app in 5 minutes with a few prompts.

Products like Claude Code allow you to iterate on ideas faster than ever. They let you build things you otherwise wouldn’t have. They give you agency.

The only issue is that Claude Code is expensive. Seriously, this thing will absolutely crunch tokens and your wallet.

In about a weeks worth of usage, Claude Code has digested over 72M tokens and output over half a million. I’ve barely use the console so the majority of this is CC.

It cost ~$45 USD.

Some notes on using it

Use CLAUDE.md

Claude Code creates a file called Claude.md. This file is referenced by CC to see what it has done previously. I recommend using this file as a way to keep track of progress, bugs and anything else it has been doing.

Remember, memory is one of the main problems with LLMs.

Create your own docs

If you have any documentation you want it to reference, simply create a file and put it in there and let Claude read and reference it. A quick way to do this:

• Install a chrome extension that lets you copy all contents of a web page

• Tell any other AI model to format it for a markdown file. This is necessary because there’s heaps of useless stuff on the page we don’t need

• Paste the result in a .md file in your project, perhaps put the file in a folder called docs.

• Tell Claude code to reference this file

For example, I was testing building a mobile app and added docs like this. I did the exact workflow above, taking info from a webpage and formatting it as a markdown file.

Talk to it without the code

Sometimes you want to just discuss what work is needed and you don’t necessarily want code. In this case, you have to specifically tell the model “don’t give me code” at the end of your message. An example of a message I’ve used:

“Let’s explore X idea. How can we build it? What is required? Is it feasible? Are there any considerations we need to take into account? Don’t give me code, let’s first explore and come up with a plan.”

It’s always good to ask them for options and explore ideas. This is can significantly improve the work they’ll do.

You see, LLMs think you know everything. You say “Build me a facebook marketplace clone” and the AI will genuinely try to do so. What it doesn’t “understand” is that you might not know the best thing to do. That’s why you have to prompt it to explore ideas, problems and solutions.

Remind it to keep it simple

Another thing that can help is to tell the AI to keep it simple. Sometimes, they’ll do something very simple in a very complicated way. Remind it to keep things simple.

You can add all of these things in a doc and tell it to reference this when building. I’m sure for specific industries, the AI will have different mannerisms and behaviours which will need to be guided and managed.

One very fortunate benefit I’ve had consulting for the last few years has been to see how AI does in a number of different fields. Because AI adoption is industry agnostic and I’ve consulted and built products across so many different and unrelated fields, I’ve seen how AI works in different situations.

Although I don’t think the differences are that unique, it’s the edge cases that surprise you.

Tell it to think harder

If you want the model to think for longer about a problem, simply tell it to do so.

This will make it show its thinking process.

Custom commands

You can create custom commands to quickly get things done by creating markdown files and adding them to the “.claude/commands/” folder.

Final Thoughts

The ultimate usage of Claude Code is where one instance control several others to complete tasks. This is possible and I’ve seen people do this already.

This is the ultimate usage of CC and I won’t be surprised if Anthropic release something like this later this year.

One of the engineers at Anthropic had this to say.

Source

Lord knows what kind of systems they’re using behind the scenes with more powerful models.

UPDATE: Anthropic is planning to release Harmony, an AI agent that has access to your local files and can make edits and changes to them [Link].

If you can’t see it, AI is slowly being given the ability to do any task on your computer. Make of this what you will.

What if I don’t want to use Claude Code?

Claude Code is expensive and it sits in your terminal. I can understand why someone wouldn’t want to use it.

Moreover, what if you want to use different AI models?

Claude isn’t necessarily the best AI model for every single scenario.

This is the kind of world we have now. The landscape has completely changed. In the beginning, Claude was best for code and ChatGPT was best for most things.

Now?

Claude will solve X but fail Y and Z. GPT-4o will solve Y and fail X and Z. R1 will solve X and Z but fail Y. o3-mini will solve Z and Y and fail X.

The reality is, there is no one model for all use cases anymore.

Source

The landscape is extremely fragmented. As models get even better, it is even harder to understand their differences.

I think this is a good breakdown for Sonnet, Grok and R1.

Source

So, what’s the best way to try all these different models?

Code Editors

The most popular AI code editor is Cursor. It is built on VS Code which is a very popular code editor.

Cursor is definitely the best AI coding tool right now. They’re the biggest, and also the fastest growing SaaS in the history of SaaS.

And they haven’t spent a cent on marketing… [Link].

You can use it to ask questions, have it write code and use it in “Agent” mode.

Agent mode is where the AI model can simply go and make all the changes it wants. If you’re not technical, you’re probably going to turn this on and let it run wild.

So, if you’re non technical, should you start using Cursor?

It depends.

I think very recently Cursor has integrated Claude 3.7 in a better way and it’s now working very well. Absolutely anyone can build something with this.

Let’s say you want to build a website. Naturally you want it to look nice. Should you go to Cursor and describe the design of your website and pray it gets it right?

I don’t think so.

No Code tools aren’t dead… yet

Lovable is the fastest growing startup in Europe ever, hitting $4M ARR in just 4 weeks after launching [Link].

It is undoubtedly the best no code AI tool. It lets you connect to an actual database, use authentication and setup a custom domain.

The reason I think it worth trying Lovable is because in my opinion, its ability to design nice frontends is fantastic. It’s really good at designing nice websites, which is what a lot of people are looking for.

They’ve also released some new features recently which make it even easier to use:

• You can edit the actual code in a file. So you could technically give the code to Claude and update the file yourself if you wanted

• You can select anywhere on the website and target a specific button or section and have the AI edit it. This is the future of no code.

Why would you build a website using a no code tool like Webflow or Framer, when an AI can build the website and you can specify all the changes and edits it requires + set up a database and authentication, all for a fraction of the cost.

This is definitely the future of website building. No code AI native building. Tell the AI what to do, and it goes and does it.

What does this mean for the future?

So, AI is getting really good at coding, and eventually, it will be good enough for a lot of jobs also.

Anthropic’s CEO Dario Amodei has said on a number of occasions that AI will completely take over programming. He recently said that within 3-6 months AI will write 90% of the code and within 12 months, most code will be written by AI [Link].

Some thoughts on this.

Will AI start writing a majority of code within 12 months?

It’s possible. But, this doesn’t mean it will take over all of the existing coding jobs.

One scenario is that so many non technical people will generate so much code that it will be more than the amount of code that currently exists today.

This take seems plausible.

Source

I think this is a possible scenario but, I would say it’s more unlikely than likely.

I think technical people overestimate the desire of non technical people to build things. Perhaps I’m wrong but this is the feeling I have at the moment and has been my observation also.

There is also the matter of bureaucracy.

Companies and governments aren’t just going to let AI start writing their code. This is simply not going to happen. Does this mean AI won’t be able to?

Probably not. But, this isn’t a matter of ability. It’s a matter of politics.

It’s also kind of funny that Dario is saying this while Anthropic is hiring so many engineers…

If AI can write all code, why do you need human engineers?

What’s in it for you?

I think for the average person, the questions you need to be asking yourself are:

• What can AI do for me?

• What work do I do that AI can take over?

• How can I get more of my time back by implementing AI?

• What repetitive task can possibly be automated?

In my opinion, the two biggest advantages of using AI are:

• Getting time back

• Exploring new ideas (quickly too)

An example on time - I’ve helped companies transform workflows they’ve had to go from:

• 30 minutes → 10 seconds

• 2-4 hours → 5 minutes

• 1+ day → 30 minutes

Nothing is more valuable than having more time to do the things you want.

Mind you, this never led to the companies letting employees go. It made their lives easier and they made more money.

I haven’t even gotten to MCPs yet and we’re at the end again. Next time 🫡.

Please consider supporting this newsletter or going premium. It helps me write more 🙂.

As always, Thanks for Reading.

^{Written by a human named Nofil}

Welcome to the No Longer a Nincompoop with Nofil newsletter.

Here’s the tea 🍵

• Claude 3.7 + Reasoning 💭

• OpenAI releases GPT- 4️⃣.5️⃣

So, Anthropic has released their new update for Claude and OpenAI has released GPT-4.5. It’s been very interesting to see how these new models behave and what the future of their usage looks like. You might be surprised.

Claude 3.7

We went from 3 → 3.5 → 3.5 (new) → 3.7. Someone please teach them how to name products.

This time, however, Claude can now reason as well. Just make sure you hit the Extended button which is not on by default.

By the numbers, Claude 3.7 is the best coding model on the planet.

That is a pretty large difference compared to other SOTA models (SOTA=State-of-the-art).

Claude’s ability to reason now also improves performance and allows it to complete tasks it wasn’t able to previously. What is rather interesting is that Claude 3.7 with thinking enabled generally only thinks for a short amount of time.

Compared to DeepSeek for example, which thinks for well over a minute, or even OpenAI’s o1 and o3 models.

What’s cool though is that we can change this, not only with prompting, but also by using it through their API.

Prompting

Since this new Claude behaves a bit differently, the way to get it to think longer is to explicitly tell it to do so. Things like:

• “Think deeply about this question before responding”

• “Consider all possibilities, scenarios and evaluate solutions during thinking”

• “First, think deeply for five minutes (at a minimum — if after five minutes, you still don't have the optimal response, keep thinking until you do) about the best way to do this, inside <thinking> tags, and then respond with your answer.” [Link]

Things like this will make the model actually think for longer than 10 seconds. Naturally, if you have a very long or complex problem with a tonne of input, the model will also think longer.

Control it’s thinking

Anthropic has made it possible to control Claude’s thinking capacity via their API. Meaning, you can tell the model how many tokens it should spend thinking before giving an answer.

This is really, really cool. I wouldn’t be surprised if the model was able to complete certain tasks by simply thinking for longer.

Did they stop there?

Nope.

Claude can now give up to 128k tokens in its output, which is a massive increase from 8k. This is one of the biggest upgrades in the new update that is not appreciated enough.

This means Claude can not only think for say 50k tokens, it can then output another 78k tokens in the same output.

This is massive if you want to extract tonnes of data, or require the model to think for long and then return long outputs.

Anthropic even tells us how best to use extended thinking.

Source

I gave it the task of writing a detailed history of the Ottoman Empire. It had:

• 3,700 words in its thinking

• 2,600 words in the actual response

Kind of interesting that the thinking was almost a thousand words longer than the actual response. In many cases, you might find your answer in the thinking tags.

If you are using reasoning models for problems, I would highly recommend reading the thinking tags, especially if you’re using DeepSeek R1.

Simon Willison was able to get Claude to use almost its entire output with this prompt.

Source

It actually worked quite well. You can read the results here and notes on this here.

You can try out Claude with extended thinking and extended output in this Repl [Link]. The max tokens is already set to 128k and thinking is set to 40k. The only thing you need to do is add your Anthropic API key.

I’ve also made it so that at the end, both the thinking and the actual response are added to separate files so you can easily read them.

Here’s another Repl where you can use this setup with an uploaded PDF [Link].

You can read more about extended thinking in their docs [Link].

Anthropic didn’t really care too much about benchmarks with this release. They talk about how they trained their reasoning model on “real-world use cases and not competition math/code”, and then proceeded to have this as a benchmark.

You can even watch Claude play Pokemon on Twitch [Link]. At one point, it got stuck in a corner and thought the game was broken so it tried a new strategy - writing a formal letter to Anthropic employees requesting a reset of the game.

Source

I’m not saying this eval is good or bad. In fact, I like it. But, there’s a bigger problem here.

We don’t know how to best test these AI models to understand what their strengths and weaknesses are.

Even if we got PDH level intelligence, we don’t have PHD level questions to ask it. Evaluating AI models is more important now than it has ever been for two reasons.

1. The models have gotten very smart. It will easily answer questions the average person could not

2. The difference in model capabilities has blurred. How do we evaluate who is best?

This is why domain specific testing is so valuable. If you have key insights in a domain or domain specific experience, you should be figuring out which AI works best for you.

Was 3.7 Rushed?

As most of you already know, Claude has been my go-to AI model for the last year. It’s an all around fantastic model. Not only at code, but most things.

Claude 3.7 is not the same as its predecessor. It is quite clear that Anthropic have focused heavily on coding performance. Claude 3.7 is an insane coding model. It will give you an entire application when you ask for a simple feature.

And this is part of the problem.

What made Claude 3.5 so good was that it did what you told it. Nothing more, nothing less. Its ability to interpret assumptions and know when to give certain information is what made it so good.

Its personality is what made people like it so much. As eery as it might sound, people saw Claude as a kind of empathetic friend.

It is absolutely not like this anymore. It’s a coding machine… At least in a good way.

Source

Someone even made a basic benchmark to check how 3.5 and 3.7 compare on doing what you ask and 3.7 was worse [Link].

Source

I find it rather interesting that Anthropic explicitly stated that 3.7 is better at “instruction following” but it just doesn’t seem to be the case. This is why benchmarks that actually measure real world usage are so important.

Check out these threads to see similar thoughts. [Link] [Link]

I think Jesse Han puts it best.

Source

Half the work I do when consulting businesses is specifying what they need to focus on.

Claude Code

Anthropic has gone as far as to release Claude Code, a command line tool that essentially makes Claude an autonomous coder on your laptop.

Unfortunately, I’m still on the waitlist for this but from what I’ve been reading, it is very good.

UPDATE: At the time of release, I got access to Claude Code. Will write about it next week. Tl;dr Could be amazing, has its flaws, is very expensive.

It’s pretty clear that soon enough, you’ll tell an AI system like Claude Code to build you something and it’ll just get it done. Considering the way 3.7 works, I can definitely see the next iterations of models doing this.

Claude 3.7 works in a strange way where it tends to rewrite entire code files rather than make in-line edits, even though it has the ability to do so. We know most of the tools it has access to, so if one wanted, they could recreate it themselves [Link].

This leads to whole other conversation about the role of models, agents, platforms and where the moat is, who will win etc. I won’t go into this here but have something exciting to share on this front soon.

How good is the coding really?

The coding capabilities are truly spectacular. It’s creating fully functional games using three.js in 10 minutes.

Seriously, check out these threads:

• A 3D ping pong game with particle trails, collision visuals, and flawless physics made in 10 minutes [Link]

• A Rocket League prototype in 20 prompts [Link]

@levelsio on Twitter has made a fully playable multiplayer flight simulator.

What started out as a simple “I can fly a plane around”, in a matter of days he’s added multiplayer support, added dogfighting, is selling custom planes and charging $30, selling blimps as ads in game for $1000, added a mini-map, added joystick support, added a leaderboard, added tanks.

The game had thousands of players at one point. I even played it and it was fun; reminded me of the Miniclip days.

The entire game is one single code file with over 4000 lines of code.

The entire game is built with Claude 3.7… Seriously impressive stuff. When there’s a will, there’s a way.

You can try the game here [Link].

UPDATE: He’s sold 10+ blimps and is now selling planets (?) and is at ~$30k MRR…

Free Speech Absolutist

If you’ve used previous Claude models then you know that they often won’t discuss certain topics. Anthropic cares a lot about safety.

Turns out Claude 3.7 is one of the most compliant models on the market [Link].

Source

Compared to Claude 3.5, it is more than willing to criticise anyone. A rather interesting phenomenon. The benchmark is open source so you can run it on any particular model that’s hosted on OpenRouter; feel free to check it out here [Link].

Future Safety Concerns

Anthropic wrote quite a lot about safety (as they always do) in their system card [Link].

There was one thing that caught my eye.

Source

Anthropic believes that the next iteration of Claude has a “substantial probability” of meeting the ASL-3 requirement.

What is ASL-3?

It refers to AI models that substantially increase the catastrophic misuse risk of AI. It requires stronger safeguards, misuse prevention and enhanced security on models.

I find this kind of funny because of two reasons:

1. Grok 3 and likely future Grok models don’t have any safeguards

2. China will develop Claude 4+ AI models and open source them

AI labs have completely different ideologies on how to bring this new tech into the world. Has AI safety been overhyped?

Absolutely. A year ago, people were claiming if a DeepSeek R1 level model was open sourced, it would spell the end of the world. Nothing has happened because of the release of DeepSeek, except maybe the price of NVIDIA stock changing.

Will this change as better models get released?

Only time will tell.

You can read the system card for Claude 3.7 here [Link].

GPT-4.5 is not what you think

GPT-4.5 is out and many people will quickly switch to it presuming it’s a successor to GPT-4o. This makes sense considering the name.

GPT-4.5 is not a successor to 4o or GPT-4. It’s not even the same model. It’s not an improvement on either of these older models. It’s a completely different model.

OpenAI didn’t even want to release 4.5. In fact, there’s a good chance they won’t even continue hosting the model considering how expensive it is.

Look at the price!!!

per million tokens

It’s way, way, way more expensive than the other top models.

Surely the price justifies the performance.

Surely…

Source

Lord have mercy, it’s not even better than DeepSeek R1 which is being served for pennies on the dollar. This is specifically for coding which is the main use case for LLMs.

It’s also really, really slow [Link].

What on Earth is going on here?

Why would OpenAI even release this model?

I mean, it’s not even SOTA. They’re not even claiming that it’s the best and we can clearly see it isn’t.

Why would anyone use this model, which is more than 10x more expensive than the next best model?

EQ.

This is an EQ vs IQ situation.

GPT-4.5 is not the smartest model. It can’t code as well as o3 or Claude. But, you know what it can do?

It can write very, very well.

There hasn’t been a model this good at writing since Claude Opus. GPT-4.5 is unlike any model most people have tried (most people haven’t tried Claude Opus).

You need to talk to this model. It is a conversationalist. OpenAI explicitly says so as well.

Source

That’s not to say it’s dumb either. The model is definitely good.

Source

GPT-4.5 is technically the best non-reasoning model on some benchmarks; just don’t look at the coding benchmark where Claude 3.7 murders it.

There are a whole bunch of graphs and tests OpenAI detail in their system card, many in relation to safety and red teaming. You can check out the GPT-4.5 system card here [Link].

GPT-4.5 is the largest model OpenAI has ever trained. It is absolutely massive, and as OpenAI themselves call it, it has a ‘big model smell’.

What does this mean?

I’ve no idea. As far as I know, no one has been able to accurately or succinctly describe what “big model smell” means.

It might be referring to “high-taste testers” as Karpathy says.

Karpathy ran a test comparing the outputs of GPT-4 and GPT-4.5 and to his surprise, the vast majority of people voted for GPT4. You can give the questions a try here [Link].

I guess everyone has their own idea of “good writing”.

Is the model good?

Yes, it is.

Is it good enough to warrant the price?

No, unless you’re so rich that you don’t care about money.

A lot of people online are talking about how its “street smart” and “feels different”. Check out some of the threads to get a better idea of the “vibe” of the model. [Link] [Link] [Link] [Link]

But, why?

The most glaring observation with the release of 4.5 is that OpenAI has scaled their compute and the size of the model significantly, but, the performance gains don’t reflect the hundreds of millions it cost to make the model.

Many believe this is what Ilya Sutskever saw before leaving OpenAI. I mean, he said it himself.

Source

I don’t know why OpenAI decided to release the model to the public, but, I think it’s likely that this model will form the basis of their future models.

(It is likely that Anthropic hasn’t released Claude 4 Opus because of how expensive it would be to host it. Anthropic isn’t exactly known for their amazing infrastructure setup either).

Reasoning models like DeepSeek R1 and OpenAI’s o1/o3 are built on top base models as I wrote many weeks ago.

What OpenAI has here is a very, very strong base model. With this, they can build even smarter reasoning models.

I think Logan puts it best (Logan was previously at OpenAI).

Source

At the same time, Yasser asks the critical question, which afaik, we don’t have an answer to.

This is also probably the last time we’ll see a model like this from OpenAI as they confirmed that they will merge their non-reasoning and reasoning models in future releases.

What’s also really interesting (and funny) is that GPT-4.5 still has an October 2023 knowledge cutoff.

Do you know what that means?

It hasn’t been trained on ChatGPT data on the internet.

Is OpenAI purposely avoiding training their models with the AI slop on the internet?

Why else would they keep the cutoff so far back? It makes no sense.

APIs, even OpenAI’s ones, have changed so much since 2023. There would be no point in using this model for coding because its knowledge is so outdated.

I find it kind of funny that OpenAI, the company responsible for all the AI slop on the internet, is purposely avoiding data generated by their own AI.

Just a bit ironic is all.

There’s a lot more I want to discuss here but alas, this newsletter is already too long and may be clipped. Look out for more exciting newsletters in the near future 🙂.

Please consider supporting this newsletter or going premium. It helps me write more :).

As always, Thanks for Reading ❤️

^{Written by a human named Nofil}