Good Morning! OpenAI just dropped two open-weight models and they’re surprisingly capable, even on edge devices. NASA, meanwhile, is handing over the future of space stations to the private sector, shifting from landlord to renter. And if you’re working in JavaScript, JSON.stringify just got a massive speed boost in V8, no changes needed on your end.

— Forrest Knight & Meghanadh Vasireddy

OpenAI Releases Two New Open-Weight Language Models

OpenAI just released two open-weight language models, gpt-oss-120b and gpt-oss-20b, targeting developers who want more control, flexibility, and cost efficiency. Unlike the usual closed models, these are fully downloadable, customizable, and Apache 2.0 licensed.

Their goal with these models is to focus on reasoning models. gpt-oss-120b runs well on 80 GB GPUs and matches o4-mini on core benchmarks. gpt-oss-20b is optimized for lower-latency or local setups which includes edge devices with just 16 GB of memory and shows similar performance to o3-mini.

Technical highlights:

• Models are available on Hugging Face, quantized in MXFP4

• gpt-oss-120b supports production-scale tasks; gpt-oss-20b is edge- and latency-friendly

• Compatible with OpenAI’s Responses API and agentic workflows

• Includes chain-of-thought support and safety optimizations

• Partners: Azure, AWS, Vercel, llama.cpp, Fireworks, LM Studio, and more

Microsoft is also rolling out GPU-optimized versions of gpt-oss-20b for Windows.

OpenAI says the goal is to support a healthier open model ecosystem. These models don’t replace their hosted offerings, but they should giving developers more options for building, fine-tuning, and deploying on their own terms.

Read More Here

Time to change compliance forever.

We’re thrilled to announce our $32M Series A at a $300M valuation, led by Insight Partners!

Delve is shaping the future of GRC with an AI-native approach that cuts busywork and saves teams hundreds of hours. Startups like Lovable, Bland, and Browser trust our AI to get compliant—fast.

To celebrate, we’re giving back with 3 limited-time offers:

• $15,000 referral bonus if you refer a founding engineer we hire

• $2,000 off compliance setup for new customers – claim here

• A custom Delve doormat for anyone who reposts + comments on our LinkedIn post (while supplies last!)

Thank you for your support—this is just the beginning.

👉️ Get started with Delve

Book your demo now

NASA Lays Out New Rules for the Post-ISS Era

The end of the International Space Station is no longer a distant idea. Around 2030, NASA plans to deorbit the ISS and let it burn up over the Pacific. That means it’s officially time for the private sector to step up, and NASA is changing how it approaches commercial space station development.

This week, NASA released a new directive outlining how it will work with companies building private orbital platforms, like Vast, Axiom, and Blue Origin, to replace the ISS.

But there’s a catch: There is an immense budget pressure, and the agency is moving away from long-term cost-plus contracts in favor of milestone-driven, fixed-price agreements.

What’s New:

• NASA will not operate future stations, it will rent time on them.

• Private companies must handle station development, ops, and sustainment.

• Funding will be focused on commercial viability and low-Earth orbit services.

• NASA wants these platforms to support science, astronauts, and international partnerships — but they won’t be the landlord anymore.

For devs, researchers, and space hardware startups, this is a pretty big shift. There will be more API consumer than infrastructure maintainer.

Read More Here

JSON.stringify Just Got a Major Speed Boost in V8

If you touch JavaScript in any way, you’re using JSON.stringify whether you're serializing data to send over the wire, stash in localStorage, or log in dev tools. And now, thanks to a major revamp in V8 13.8 (shipping with Chrome 138), it’s more than 2x faster.

What’s New:

• Introduced a side-effect-free fast path that skips unnecessary checks for plain objects

• Split implementation for one-byte vs two-byte strings to avoid branching

• SIMD and SWAR-based string scanning for escaping ", \, etc.

• A new Dragonbox double-to-string conversion replaces Grisu3 for number serialization

• Switched to a segmented buffer system, avoiding costly reallocations

• Added an "express lane" optimization using hidden class metadata for ultra-fast key serialization

The improvements apply automatically to simple, common cases (plain objects, no custom .toJSON, no pretty-printing). If you’re serializing structured data for APIs, config, or storage, you’ll likely benefit without changing a thing.

Read More Here

🔥 More Notes

Jack Dongarra on the Future of Supercomputing: The next wave of supercomputers will blend CPUs, GPUs, quantum chips, neuromorphic hardware, and even optical processing. And AI will play a more transformative role than the internet itself, but he cautions quantum computing still needs serious maturation.

Autodesk CEO: Interdisciplinary Thinking Beats Coding: Autodesk CEO Andrew Anagnost argues coding alone won’t cut it in the AI era. What’s becoming critical is systems-level, cross-disciplinary problem-solving and knowing how to harness AI agents to collaborate, not just write code. He sees software teams shifting to more integrative roles, where broad thinking outranks pure tech.

Meta Throws $250M at a 24-Year-Old AI Whiz: Meta just handed $250 million compensation package to 24-year-old AI researcher Matt Deitke, doubling an already massive offer.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! A sleep-deprived human just out-coded an OpenAI model in a 10-hour brain-melting marathon—yes, really. Meanwhile, Intel pulled the plug on Clear Linux, breaking the hearts of performance nerds everywhere. And Python 3.14 just dropped its first release candidate, finally giving us GIL-free multithreading and a JIT, because dreams do come true.

— Forrest Knight & Meghanadh Vasireddy

Human Beats AI at Coding—But Barely

Context: In what feels like a modern retelling of the John Henry legend, Polish coder Przemysław Dębiak ("Psyho") just beat OpenAI’s custom model in the AtCoder World Finals Heuristic Contest. The 10-hour grind involved solving a brutally complex optimization problem—think NP-hard, no exact answers, just iterative improvements until your brain melts. OpenAI entered a beast of a model, similar to o3, into this head-to-head showdown. And guess what?

The human won. Barely.
Dębiak scored ~1.81 trillion points and OpenAI scored ~1.65 trillion. That’s a 9.5% edge—tiny, but enough. Also, this was not some comfy for Dębiak, he was sleep-deprived, coding for days across multiple contests. His response? “I’m barely alive.”

Why this matters:

• AI tools like Copilot and Cursor are basically standard now, GitHub says 90% of devs use them.

• In just one year, coding benchmarks saw AI jump from solving 4.4% of problems to 71.7%.

• Yet, even with all that, human creativity still edged out algorithmic efficiency.

Read More Here

Clear Linux is Dead And It Sucks More Than You Think

Welp, it’s official. Intel just nuked Clear Linux, their in-house, performance-tuned distro. If you've ever benchmarked it, you know it was a beast—squeezing out jaw-dropping perf numbers on both Intel and AMD hardware. But in a very 2025 move, Intel dropped support out of nowhere. No more updates. No more security patches. GitHub's now in read-only mode. Time to migrate.

RIP to the distro that made your Ryzen CPU feel like it hit the gym. Intel’s statement was all “thanks for the memories” while waving the budget axe. They're still contributing upstream, apparently, and working with other distros. But let’s be real—Clear Linux was their sandbox for aggressive kernel tweaks, LTO, PGO, and everything perf nerds live for. Now that sandbox is gone.

Takeaways:

• Clear Linux = officially sunsetted, effective immediately.

• No more updates or security patches—seriously, move off it.

• Intel says they’ll keep supporting Linux... just not this Linux.

Honestly, this hurts. Sure, most of us weren’t daily driving Clear, but it was a living example of what Linux could be with the right tuning. And now, between layoffs and driver maintainers walking out, it’s hard not to wonder where Intel’s Linux strategy is really headed.

Read More Here

Python 3.14 RC1 Is Here, Time to Test Your Code

Context: Python 3.14’s first release candidate (3.14.0rc1) has officially dropped, putting the language one step closer to its October launch. The ABI is now frozen for the entire 3.14 series, meaning if you’re building wheels against RC1, they should work just fine when the stable release hits. For now, only bug fixes will make it in.

Free-threaded CPython (PEP 779) is finally real. The GIL (Global Interpreter Lock) is effectively removed for many workloads, enabling true parallelism. Yep, your multi-core CPUs can finally stretch their legs without multiprocessing hacks.

Beyond the GILocalypse, 3.14 packs some seriously cool upgrades:

• JIT compiler (experimental): macOS and Windows builds get an opt-in JIT that could speed up many workloads.

• Multiple interpreters (PEP 734): Isolation and concurrency within a single process—think cleaner, safer parallel tasks.

• T-strings (PEP 750): f-strings on steroids for custom string processing.

• Syntax glow-up: except/except* without brackets, REPL syntax highlighting, and colorized output for unittest and friends.

There’s also a new compression.zstd module, faster UUID generation (up to 40%!), a formally verified HMAC implementation, and a zero-overhead debugger interface (PEP 768).

RC2 is planned for August 26, and the final release drops October 7. If you maintain Python packages, now’s the time to test and publish 3.14-compatible wheels.

🔥 More Notes

• NVIDIA announced this week that they are bringing their CUDA software to RISC-V processors

• Google launches OSS Rebuild tool to improve trust in open source packages

• Apple unveils second Osaka store ahead of its opening

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Google I/O 2025 just unleashed a tidal wave of AI features—from the snappy new Gemini 2.5 Flash Preview to agentic Colab and Jules the async coding sidekick—so you can prototype and ship at warp speed. A fresh Lawrence Berkeley Lab report warns that by 2028 AI workloads could gulp 165–326 TWh annually, enough to power nearly a quarter of U.S. homes. And MIT Technology Review’s “Power Hungry” deep dive shows how a single query can burn hundreds of joules once you factor in servers, cooling, and carbon intensity—courtesy of MIT.

— Forrest Knight & Meghanadh Vasireddy

Google I/O 2025’s Fresh AI Models & Agentic Tools

Context: Google I/O 2025 was a full-on AI bonanza for devs. Mat Velloso (VP of AI Developer Platform at Google DeepMind) reminded us that “developers are the architects of the future,” and then unleashed a slew of model and tooling upgrades to supercharge your workflows. Whether you’re building chatbots, vision apps, or full-stack AI services, there’s something here to speed up your next proof-of-concept.

What’s new:

• Gemini 2.5 Flash Preview: Faster, leaner coding and reasoning, with thought summaries to help you manage token costs.

• Gemma 3n & PaliGemma: Lightweight open multimodal models for on-device audio/text/image/video, plus a vision-language champ for captioning and VQA.

• Agentic Power-Ups: A sneak-peek at agentic Colab that auto-fixes your notebooks, plus Jules—the async coding sidekick that spins up VMs, clones repos, and preps PRs.

• Workflow Boosters: Firebase Studio now auto-provisions backends from Figma, and Stitch can whip up UI designs and front-end code from plain English or images.

Dive into the previews today in Google AI Studio and Vertex AI—Flash goes GA in early June, Pro follows soon after. Grab the keys, build faster, and let these tools handle the heavy lifting so you can focus on the fun stuff.

MIT’s Uncovering the Carbon Footprint of Every Query

Context: You’ve probably noticed AI is gobbling up cycles and data-center juice like never before. But what’s the real cost of that chat prompt, image render, or video snippet?

What’s new: MIT Technology Review’s latest “Power Hungry” analysis (May 20, 2025) crunches the numbers from expert interviews, open-source benchmarks, and grid data to map AI’s carbon footprint down to the joule:

• Per-query footprints:

  • Text LLMs: ~114 J for Llama 8B → ~6 700 J for Llama 405B

  • Stable Diffusion (1 024×1 024): ~2 300 J

  • Open-source video (5 s @16 fps): ~3.4 MJ

They forecast that by 2028, AI could gulp 165–326 TWh annually—enough to power 22% of US homes—driving a data-center boom that breaks past decade-long efficiency gains. Since most centers rely on fossil-heavy grids, regional carbon intensity swings wildly from ~70 gCO₂/kWh in sunny afternoons to >300 g at night or in coal-dependent areas.

Takeaway: A single AI call feels negligible, but at billions of daily hits, the cumulative toll is massive. The piece urges closed-box AI providers to open their energy books and for smarter grid planning to keep our AI ambitions sustainable.

Sponsored

Snyk Launch 2025 | Secure the Future of AI Application Development

AI is definitely speeding up app development and letting us build bigger, more complex things. However, security needs to evolve just as quickly, or we risk new vulnerabilities. Balancing innovation with protection is a growing challenge.

Snyk is holding an event, Snyk Launch 2025, on May 28th. They'll be sharing their latest innovations to help development teams tackle this security gap, especially with AI in the mix.

Here's a quick look at the agenda:

• Peter McKay (CEO): Discussing security risks in AI enterprises & leadership strategies.

• Danny Allan (CTO): A deep dive into Snyk's new AI security technology.

• Manoj Nair (CIO): Talking about expanding secure AI through ecosystem partnerships.

• Live Demos: Showcasing their AI-driven security tools in action.

If you're working on keeping software secure in this fast-paced AI world, this could be insightful.

Register for Snyk Launch: https://snyk.plug.dev/tDSUQbj

🔥 More Notes

• World’s First Petahertz-Speed Phototransistor in Ambient Conditions: Researchers built a phototransistor that manipulates electrons with light pulses under real-world conditions at petahertz speeds—breaking the trillion-cycles-per-second barrier and opening doors for ultrafast optoelectronics.

• Microsoft Simplifies AI Training on Companies’ Own Data: At Build 2025, Microsoft unveiled “Copilot Tuning,” a low-code tool that lets non-engineers fine-tune AI models on their own datasets, plus multi-agent orchestration features to streamline enterprise AI development.

• Russian GRU Cyber Actors Targeting Western Logistics & Tech Firms: A joint CISA/FBI/NSA advisory warns that Russian GRU hackers are actively exploiting supply-chain networks and tech infrastructures—urging firms to harden endpoints and monitor for IOCs.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Firefox just flipped the switch on its hg.mozilla.org Mercurial servers and officially moved the Firefox source to GitHub. Meanwhile, Stack Overflow is staring down a 64% year-over-year drop in Q&A traffic and is rolling out a full rebrand—AI assistants are swiping our questions and they are attempting to fight it. And if you thought Figma’s slick Dev Mode was “included,” brace yourself—teams need extra seats per project to actually use it, and that little fine-print sting is leaving freelancers feeling scammed.

— Forrest Knight & Meghanadh Vasireddy

Firefox’s source code is now officially on GitHub!

Context: Firefox has been living on hg.mozilla.org under Mercurial since forever—complete with its own commands, hgtags, and a separate CI setup outside most devs’ day-to-day GH workflows.

What’s new:

• The canonical Firefox repo is now at https://github.com/mozilla-firefox/firefox

• Legacy Mercurial servers will stay in sync for now; CI, bug tracking, and code review pipelines aren’t moving (yet)

• Contributions are smoother—no extra accounts required if you already have a GitHub ID

If you’ve forked linux-kernel mirrors or poked at GNOME or LibreOffice on GitHub, this should feel like second nature. No major workflow changes, just fewer infra headaches. So grab your PR hat, skim the new GitHub Actions configs, and let’s start rolling those Gecko patches. Happy hacking!

Stack Overflow is gearing up for a facelift.

Context: Stack Overflow’s core Q&A action has taken a nosedive—April 2025 saw question-and-answer volume plummet over 64% compared to last year, and it’s down more than 90% since its 2020 heyday. AI chatbots, in-IDE helpers, and Copilot extensions are straight-up stealing our rep points and pageviews.

Stack Exchange is hitting refresh with a full rebrand that breaks the mold—they’re no longer “just Q&A.” Here’s the skinny on what they’re launching:

• Three new pillars—Q&A stays, then add “Community” for open discussions and “Careers” to hook you up with jobs

They’ve already tinkered with AI Answer and Question Assistants, revamped the jobs board (thanks, Indeed), and dropped GitHub Copilot, Slack, and VS Code extensions to keep you tethered. Basically, if you’re knee-deep in devops pipelines, language internals, or framework wars, expect more AI in the future rather than old SO that we were all familiar with.

Dev Mode or Dev Scam? Figma’s Hidden Seat Trap

Context: You’ve probably seen Figma’s slick Dev Mode launch pitch—inspect CSS, grab tokens, spin up VS Code extensions, and generally feel like a front-end wizard. But here’s the kicker: if you work with multiple clients or freelancers, that “Dev Mode included” badge hides a team-seat trap. Pay your $15/month, sure—but only for one team. Want to peek into your client’s file? You need them to pony up too. Cue the “Figma Dev Mode is a scam” Reddit thread, where devs are fuming over hidden dark patterns and double-dipping pricing.

Figma’s doubling down on Dev Mode as a premium feature, but community backlash is real. Highlights from the trenches:

• Developers report needing extra paid seats per project team, losing comments/annotations when importing into free alternatives (Pixso, Penpot)

So, if you’re knee-deep in webpack configs, SCSS mixins, or debugging that one CSS grid bug, weigh your options: stick with Figma and budget for every collaborator, or export to an “inspect-for-free” challenger. Either way, keep your Git branches tidy and your wallet guarded—because even the coolest tooling can sting when the fine print bites.

🔥 More Notes

• Potential U.S. semiconductor manufacturing boom faces policy headwinds: Major investments under the CHIPS Act are fueling a domestic chip revival, but proposed tariffs and trade probes from the previous administration threaten to slow or derail new fabs in Texas and California.

• Tennessee to require every high schooler to take computer science: Starting with the class of 2028, all Tennessee students must complete at least one CS course—part of a push to boost tech literacy and meet future workforce needs. Other states are watching closely.

• Vision-language AI models can’t reliably handle “no” and “not”: New MIT research shows that common multimodal models stumble on queries containing negation words, raising concerns for high-stakes applications like medical diagnosis or legal text interpretation.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Discord just pulled off a massive upgrade to their search system, finally making it strong enough to handle trillions (yep, trillions) of messages without falling over. Meanwhile, the crypto world took a hit when the official XRP NPM package got hacked to steal private keys — so if you use it, go patch yesterday. And over at Cape Canaveral, Amazon is gearing up (weather permitting) to launch the first wave of Project Kuiper satellites to take on SpaceX’s Starlink empire.

— Forrest Knight & Meghanadh Vasireddy

How Discord Scales Search for Trillions of Messages

Context: Back in 2017, Discord's search engine was already handling billions of messages using Elasticsearch with smart sharding (by server or DM). It worked—until it didn’t. Redis queues melted under load, Elasticsearch nodes threw tantrums, and massive servers (aka "Big Freaking Guilds") hit size limits. Fun times.

In classic Discord fashion, they tore it all down and rebuilt smarter. Highlights:

• Kubernetes + ECK: They moved Elasticsearch onto Kubernetes for easier scaling, upgrades, and zoning resilience.

• “Cell” Architecture: Instead of giant clusters, they now run tons of smaller ones (better fault tolerance, faster queries).

• PubSub Queues: Goodbye Redis. Hello guaranteed message delivery, even if Elasticsearch has a bad day.

• Smart Bulk Indexing: Now messages are grouped by cluster/index before bulk upload, minimizing retry nightmares.

• Cross-DM Search: DMs are now sharded by user, letting you finally search all your DMs without crying.

• BFG Handling: Big guilds that post like maniacs get custom multi-shard indexes to avoid blowing up Lucene limits.

Results:

• Doubled indexing throughput 🚀

• Median search latency dropped from 500ms to under 100ms

• 40 clusters, thousands of indices, and rolling updates with zero downtime

Read More Here

Major XRP Supply Chain Attack — What You Need to Know

Context: If you thought open-source dependencies were safe… uh, think again. This week, Aikido’s threat feed spotted something shady: the official xrpl.js library for XRP (you know, the cryptocurrency) got compromised on NPM. Attackers sneaked a backdoor into five versions to quietly steal private keys. Yeah — private keys. Wallets = exposed. Fun!

Turns out, someone likely hacked a Ripple dev’s NPM account, pushed poisoned versions, and almost pulled off one of the biggest crypto supply chain attacks we’ve seen this year.

Here’s the damage report:

• Affected versions: 4.2.1, 4.2.2, 4.2.3, 4.2.4, 2.14.2

• Backdoor: Secretly exfiltrated keys to 0x9c[.]xyz

• Attack method: Malicious checkValidityOfSeed calls buried in wallet functions

• Patch versions: 4.2.5 and 2.14.3 — upgrade now

• CVE: CVE-2025-32965 (CVSS 9.3 — critical)

If you had any of those versions, assume your keys are compromised, rotate immediately, and move funds out of impacted wallets. Don't wait.

And a PSA: just because a package is “official” doesn’t mean you can skip checking your dependencies. 🛡️

Amazon’s Kuiper Launch – Will They Beat the Weather This Time?

Context: Amazon is finally kicking off Project Kuiper — their $10 billion answer to SpaceX’s Starlink — by launching their first 27 internet satellites into low Earth orbit. After a weather fail earlier this month, they’re back at Cape Canaveral, crossing fingers for a Monday night (April 28) liftoff aboard a ULA Atlas V rocket. Forecast? 75% chance it’s go-time, but Florida thunderstorms are always lurking.

Here’s the quick rundown on why this matters:

• First major Kuiper deployment: 27 satellites today, 3,200+ planned in total

• Rocket: Atlas V with maxed-out boosters for the heavy lift

• Backup launch window: Tuesday, with 90% better weather odds

• Orbit goals: Initial 280 miles up → then cruising at 392 miles

• Competition: Amazon aims to take on SpaceX’s 5M+ Starlink customers globally

• Long-term plan: Over 80 launches using ULA, SpaceX, and Bezos' Blue Origin

If you’re on the East Coast, you might even catch the launch lighting up the sky — assuming no giant thunderclouds crash the party.

🔥 More Notes

• Alphabet's Earnings Soar: Alphabet, Google's parent company, reported a 12% revenue increase in Q1 2025, totaling $90.2 billion. Net earnings rose by 46% to $34.54 billion, driven by growth in Search, YouTube, and AI services. ​

• Nvidia's Robotics Push: Nvidia is gearing up to launch its Jetson Thor computers for humanoid robots in 2025. These systems aim to enhance robot autonomy and human interaction through AI. Nvidia is partnering with companies like Siemens and Universal Robots, focusing on providing the technology rather than manufacturing the robots themselves.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! IBM and ESA just launched TerraMind, a next-gen AI model that understands Earth using satellite images, climate data, and even topography—all open-source. Microsoft’s new BitNet model is making waves by running large language models on regular CPUs with a sleek 1-bit architecture (yep, no GPUs needed). And if open-source security alerts are driving you up the wall, Hopper just came out of stealth to bring precision and peace of mind to AppSec teams.

— Forrest Knight & Meghanadh Vasireddy

TerraMind: The AI That Gets Earth

Context: IBM and the European Space Agency just dropped something big for Earth Day—TerraMind, an open-source, multimodal AI model trained on over 9 million globally-distributed data samples. It’s like GPT, but for the planet. Built with help from KP Labs, the Jülich Supercomputing Centre, and the German Space Agency, TerraMind goes beyond image processing—it's designed to "intuitively understand" Earth through diverse data like satellite images, topography, vegetation, and even time-based climate patterns.

What’s New:

• TerraMind beat out 12 leading models on the PANGAEA benchmark (aka the SATs for Earth AI).

• It uses “Thinking-in-Modalities” (TiM), kind of like chain-of-thought prompting in LLMs, to self-generate training data when info is sparse.

• Built on the IBM-NASA Prithvi framework, but lighter and more compute-efficient—10x less power than traditional models. Climate-friendly AI? Yes please.

This model can:

• Classify land cover

• Detect changes over time

• Analyze multi-sensor input

• Forecast climate risks (like water scarcity or deforestation)

One bullet-point takeaway list:

• Multimodal AI trained on 9 data types

• Open-source on Hugging Face

• Performs 8%+ better than competitors

• TiM boosts efficiency & accuracy

• Already making waves in climate forecasting & disaster response

TerraMind is available now, and fine-tuned versions are on the way. Expect to see it powering smarter Earth insights—from sustainable agriculture to emergency flood mapping. Happy Earth Day, indeed.

BitNet’s Big Play: Microsoft Ditches GPUs for CPUs (and It Works)

Context: For years, running large language models (LLMs) meant one thing: GPUs. Expensive, energy-hungry, heat-blasting GPUs. But Microsoft Research just flipped that script with BitNet b1.58 2B4T—a new AI model that runs on a regular CPU. Yep, no graphics cards needed. The model’s been published on arXiv and is already shaking up the “big compute” scene.

BitNet drops traditional floating-point weights in favor of a 1-bit architecture. Think: -1, 0, or 1. No matrix multiplications, no high precision fluff—just simple add/subtract operations. That means:

• Smaller memory footprint

• No GPU dependency

• Orders of magnitude lower energy consumption
  And get this: despite being tiny (under 3B params), BitNet performs on par with—or better than—other open-weight LLMs in its size class. All while sipping power.

Key Bits to Know:

• No floats—just 1-bit weights

• Runs natively on CPUs

• Built-in runtime: bitnet.cpp

• Outperforms some GPU-based models

• Big potential for edge AI + privacy (think: local chatbots, offline use)

Why this matters?: BitNet is pushing LLMs toward a more efficient, accessible future. If these results hold, running advanced AI on a laptop—or even a phone—might be closer than we thought. GPUs? Optional. Brains? Required.

Hopper Enters the Chat: Open-Source Security, But Actually Useful

Let’s be honest—most Software Composition Analysis (SCA) tools are like spam filters from 2005: noisy, outdated, and not very smart. Hopper, fresh out of stealth with $7.6M in seed funding, is here to flip the script. Founded by cyber-intel veterans from Israel’s Unit 81 and 8200, Hopper isn’t just another scanner yelling “CVE!”—it’s built to actually help you.

What’s New: Instead of flagging every theoretical vuln in a dependency tree, Hopper narrows things down to what really matters: function-level reachability. You get insights like “this exact line of code in this exact function is exploitable” without cluttering your CI/CD or drowning your devs in false positives.

Why It Matters (aka what Hopper nails):

• Function-level reachability for direct & transitive deps

• Contextual remediation evidence (source-linked)

• Agentless deployment via read-only Git access

• Full SBOM + VEX export support

• Hidden vulns? Detected. Shadow deps? Found.

• Built-in precision = lower MTTR, happier devs, safer code

Legacy players like Snyk, Mend, and Sonatype? Still doing manifest-level scans and sending alerts you’ll never read. Hopper’s angle is precision, speed, and dev-alignment—basically making AppSec not suck.

Already deployed at companies like HPE and Mesh Payments, Hopper’s turning open-source security from a blocker into a booster. Welcome to the new standard in OSS risk management.

Check it out at hopper.security

🔥 More Notes

• AI-Powered Learning at Middle Georgia State University: Middle Georgia State University is pioneering an AI-driven adaptive learning system for Python programming. Funded by a $9,000 grant, the project aims to fine-tune the LLAMA 3.3 model and develop a web application interface. This system will provide students with interactive coding tasks, quizzes, and real-time feedback, enhancing their learning experience.

• Boise State Student Wins NASA Award for Exoplanet Research: Malia Barker, a doctoral student in computing at Boise State University, has been awarded a Future Investigators in NASA Earth and Space Science and Technology (FINESST) gran. Her research focuses on advancing computational techniques to study the demise of certain exoplanets, contributing to our understanding of planetary system.

• Quantum Computing Achieves Randomness Breakthrough: Quantinuum's 56-qubit trapped-ion quantum computer has successfully demonstrated certified randomness in quantum circuits. This achievement is a significant step toward establishing secure, private connections, potentially leading to a spoof-proof internet.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! OpenAI just dropped Codex CLI, a terminal-native coding agent that doesn’t just suggest code—it runs it, edits it, and vibes with your repo like a real teammate. Over in court, Google got slammed with a monopoly ruling for locking down the ad tech game—think DFP + AdX tying, “Last Look,” and all that jazz. And in the world of cybersecurity, the CVE program barely dodged a shutdown thanks to a last-minute rescue by CISA (but yeah, we’re gonna need a better plan than vibes and Hail Marys).

— Forrest Knight & Meghanadh Vasireddy

OpenAI Codex CLI Is Here — And It’s Actually Sick

Context: OpenAI just dropped Codex CLI, an open-source coding agent that lives in your terminal—yes, your actual CLI. Think GitHub Copilot, but stripped down, jacked up with o4-mini brains, and wired straight into your dev workflow. It’s chat-driven development, but this time, your assistant can actually run the code, edit files, and explain that one function you copy-pasted three months ago and never fully understood.

What’s new: It works locally, respects version control, and plays nice with multimodal inputs like screenshots and sketchy diagrams. It even supports three approval modes depending on how much control you want to give it (or how much you trust it today).

Key Highlights:

• Works with OpenAI’s latest models (o3, o4-mini, etc.)

• Multimodal: accepts screenshots & markdown for context

• Runs in a sandboxed, offline-safe environment

• Approval Modes: Suggest, Auto Edit, Full Auto

• $1M in API credits up for grabs for early devs

This isn’t just about writing prettier for loops—it’s OpenAI inching toward the “agentic software engineer” idea, one terminal session at a time. And yeah, it’s totally open source. Clone it on GitHub, drop in your API key, and start vibe-coding.

Oh, and Windows devs? Fire up WSL—Codex CLI’s not native on your turf just yet.

Google Found Guilty of Ad Tech Monopoly – Here's What That Means

Context: Google just took a major L in court. A U.S. District Judge ruled the company illegally monopolized two key parts of the digital ad ecosystem: publisher ad servers and ad exchanges. TL;DR: Google used its tools (DFP + AdX) to lock in publishers, crush rivals, and control who got paid what and when in the open web’s ad economy.

What’s new: This ruling might force Google to break up its ad tech business—think spinning off Google Ad Manager. The DOJ’s win also shows regulators are done playing softball with Big Tech’s backend empires.

Key points from the ruling:

• Google tied its ad exchange (AdX) access to its ad server (DFP), effectively saying “use both or get lost.”

• AdX got shady perks like “Last Look,” letting it beat rivals in auctions by seeing their bids first.

• The court said this crushed competition and harmed publishers (and, indirectly, all of us who live on the internet).

Google’s response? “We won half the case, we’ll appeal the rest.” But with the DOJ eyeing divestiture, this isn’t a slap on the wrist—it’s a warning shot for other tech giants running vertically integrated empires.

Next up: the remedy hearing. Will Google actually have to spin off parts of its stack? Stay tuned.

Oh, and if you’re building in ad tech right now… this might just be your moment.

CVE Program Almost Went Dark—CISA Pulled a Last-Second Save

Context: For a hot minute, the entire cybersecurity world collectively held its breath. MITRE’s 25-year-old CVE program—the backbone of how we track and classify software vulnerabilities—was hours away from shutting down. Why? DHS ghosted on renewing the funding contract. If that sounds apocalyptic for defenders, it kind of was.

What’s new: CISA stepped in literally the night before the lights went out and signed an 11-month contract extension. So no, CVE didn’t die—but the scare was real, and it exposed some brittle spots in our global vuln-management infrastructure.

Why this mattered (and still does):

• CVEs power nearly every security tool and database you use.

• MITRE’s work supports CNAs like Google, Microsoft, etc.

• A shutdown would've halted new CVEs, broken NVD workflows, and crippled patch triage.

• Budget cuts and Musk’s “efficiency” initiative at DHS/CISA have already slashed thousands of cybersecurity jobs.

To avoid this mess next year, CVE board members just launched the CVE Foundation, a nonprofit aiming to ensure long-term stability without being hostage to a single contract.

Cyber folks are breathing again—for now—but let’s not forget: the entire vulnerability disclosure ecosystem was nearly unplugged over a budgeting hiccup. Maybe it’s time we stop treating foundational security infra like a side project.

🔥 More Notes

• NASA's Quantum Gravity Sensor Mission: NASA's Jet Propulsion Laboratory is developing the Quantum Gravity Gradiometer Pathfinder (QGGPf), the first space-based quantum gravity sensor. Utilizing ultra-cold rubidium atoms, this mission aims to detect gravitational anomalies with unprecedented precision from orbit. The technology promises to enhance our understanding of Earth's hidden mass shifts and has potential applications in planetary science.

• Nvidia Faces $5.5 Billion Charge Due to New U.S. Export Controls: Nvidia announced it expects a $5.5 billion charge in its fiscal 2026 first-quarter results following new U.S. government restrictions on exporting its H20 AI chips to China. The company's shares fell nearly 6% in premarket trading after the announcement.

• Stanford Releases 2025 AI Index Report: Stanford University's Human-Centered AI Institute has published the 2025 AI Index Report, providing a comprehensive overview of artificial intelligence's technical progress, economic influence, and societal impact. The report aims to equip policymakers, business leaders, and the public with data-driven insights into AI development.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Apple’s getting slapped with a €150 million fine in France for making app tracking too privacy-focused—and too painful—for third-party devs. Meanwhile, Microsoft’s Hyperlight Wasm is shaking up the cloud scene with blazing-fast, OS-free WebAssembly micro-VMs that don’t need your bloated Linux distro to run. And over at NVIDIA, the newly open-sourced KAI Scheduler is here to finally fix Kubernetes GPU scheduling without making you cry into your YAML.

— Forrest Knight & Meghanadh Vasireddy

Apple Fined €150M for Privacy Feature That Was a Little Too Smart

Context: Apple’s App Tracking Transparency (ATT) was supposed to be a privacy win. Rolled out with iOS 14.5 in 2021, it forced apps to explicitly ask for user permission before tracking them across other apps/websites. Great for users. Less great for ad-reliant developers.

Fast forward to now: France’s competition regulator just handed Apple a €150 million fine (~$162M USD). Why? Because while ATT had noble goals, its implementation “unfairly complicated things” and hit small devs the hardest.

The Autorité de la concurrence said Apple’s rollout of ATT:

• Flooded users with popups (thanks to overlapping consent systems).

• Didn’t give 3rd-party apps a level playing field—Apple used a simpler consent method for its own apps.

• Made life harder for small publishers, who lack Big Tech’s 1st-party data ecosystem.

The French watchdog didn’t fault ATT’s goal, just the “neither necessary nor proportionate” way it was rolled out—particularly given Apple’s dominant position in iOS app distribution.

So What?: Even though the fine is pocket change for Apple (they made $124B in one quarter), this sets a precedent. And it’s not just France—Germany, Italy, and Poland are eyeing similar moves. Devs and advertisers who’ve been saying “this feels kinda rigged” might finally get some relief.

Apple isn’t being forced to change ATT (yet), but regulators clearly want a fairer, less hostile ecosystem.

Hyperlight Wasm: Micro-VMs Go Ultra-Lite

Context: Remember Hyperlight? That bare-metal Rust-y micro-VM project Microsoft dropped last year that let you run isolated functions without booting a full OS? Yeah, that one. Well, it just leveled up. Meet Hyperlight Wasm—the new kid in the Cloud Native Computing Foundation (CNCF) Sandbox that lets you run WebAssembly (Wasm) components in a Hyperlight micro-guest, across languages and runtimes.

What’s New: Hyperlight Wasm strips away the OS baggage and brings you straight to wasm execution—with a startup time as low as 1ms. Yes, 1 millisecond. No virtual devices, no kernel, no bloated runtime soup. Just memory, CPU, and a killer Wasm engine (wasmtime).

• Supports any wasm32-wasip2 compiled workload: Rust, Go, C, Python, JS, even C#—as long as it’s Wasm-friendly.

• Security++: Wasm sandbox inside a micro-VM? Double isolation layer, baby.

• Edge-ready: Tiny footprint = cheaper compute near users (hint: Azure Front DoorEdge Actions is coming).

Dev Vibes: Hyperlight Wasm is an OSS under Apache 2.0. The starter UDP echo server example is literally just a git clone and cargo run away. No WASI defaults yet? True. But those are in the pipeline.

GitHub repo

NVIDIA’s KAI Scheduler: Kubernetes GPU Management, But Smarter

Context: If you've ever wrestled with scheduling AI workloads on Kubernetes and thought, “There has to be a better way…”—well, now there is. NVIDIA just open-sourced its KAI Scheduler (originally built for the Run:ai platform), and it’s a GPU-aware, queue-savvy, cluster-hugging scheduler that’s here to make resource orchestration actually work for ML teams.

KAI Scheduler is now Apache 2.0 licensed and ready for the community. It’s built to handle real-world AI workflows that bounce between “one GPU for a Jupyter notebook” to “all GPUs, now, for distributed PyTorch training.” Key features include:

• Dynamic fair-share computation and GPU allocation

• Gang scheduling and podgroup support for distributed jobs

• Hierarchical queues with preemption and reclamation

• Intelligent consolidation to free up contiguous GPUs

• Auto-integration with AI tools like Kubeflow, Ray, and Argo

Real Talk: Scheduling is hard. Doing it well across teams, jobs, and frameworks is harder. KAI tackles the whole “GPU hoarding” problem by reallocating idle resources and making fairness actually enforceable. Plus, it rebalances workloads in-flight when fragmentation gets in the way.

It’s already battle-tested by NVIDIA customers running large-scale AI. You can check it out on GitHub and start experimenting now.

🔥 More Notes

1. Bloomberg Leverages AI to Enhance Analyst Productivity: Bloomberg's Chief Technology Officer, Shawn Edwards, announced that the company is developing AI tools aimed at streamlining up to 80% of an analyst's workload. By applying generative AI to unstructured data, tasks that traditionally consume significant time—such as researching and analyzing documents—can be expedited, potentially increasing researchers' productivity tenfold. Bloomberg has invested heavily in AI research, employing over 350 AI experts to advance these initiatives.

2. China's Open-Source AI Initiatives Face Uncertainty: Chinese tech giants like Alibaba, Baidu, and Tencent have embraced open-source AI to collaborate and pool resources, especially in light of U.S. tech restrictions limiting access to advanced technologies. While this approach has bolstered China's technological self-sufficiency and global standing, concerns arise about revenue generation and the potential reluctance to share critical advancements with international competitors. As China's AI capabilities grow to rival Western counterparts, there may be a shift towards restricting the availability of these open-source tools.

3. United Airlines to Offer Free In-Flight Wi-Fi via Starlink: United Airlines has received FAA approval for its first aircraft equipped with Starlink, Elon Musk's satellite broadband service. Starting in May, passengers on a United Express Embraer 175 can enjoy free in-flight Wi-Fi with speeds 50 times faster than current offerings. United plans to install Starlink on its regional fleet at a rate of approximately 40 installations per month, aiming to enhance the in-flight experience with improved connectivity and entertainment options.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Oracle Health just got hit with a pretty bad data breach involving patient records and some extremely sketchy “don’t email us” vibes. Meanwhile, DOGE wants to rewrite Social Security’s COBOL code in months (gulp), and OpenWorm is still trying to make one tiny digital worm wiggle—and honestly, it's way harder than it sounds.

— Forrest Knight & Meghanadh Vasireddy

Oracle’s Health Data Breach: The Breach Heard 'Round the Cloud

So here’s the deal: Oracle Health (née Cerner) got hit. Hard. A threat actor—who goes by “Andrew” and apparently skipped the whole “working in a team” thing—exfiltrated patient data from legacy servers that hadn’t yet made the leap to Oracle’s cloud. This all went down post-January 22, with Oracle catching wind around February 20.

It’s messy. We’re talking potentially multiple US hospitals, compromised EHRs, and extortion demands in crypto. Oracle’s stance? They’re not alerting patients—that’s up to the hospitals. Oh, and no public disclosure yet either. Communications? Not even on official letterhead. Just plain paper and a “call us, don’t email” directive.

The breach hasn’t just sparked panic—it’s raising some red flags across infosec circles. The access vector? Compromised customer credentials (still unclear how that enabled access to multiple orgs). The aftermath?

• Oracle won’t notify patients directly, but they’ll “support” with templates.

• They’re covering credit monitoring, but not handling mailings.

• Hospitals are scrambling with limited guidance—mostly via phone calls.

And just to spice things up, this comes on the heels of an alleged breach of Oracle Cloud’s federated SSO system, with 6M user records reportedly lifted.

TL;DR: Legacy servers, stolen creds, vague guidance, and a one-man ransomware show. If your org is still running sensitive ops on non-cloud infra—this is your blinking neon sign to move. Yesterday.

COBOL’s Last Stand? DOGE Thinks SSA Can Reboot in Months.

Context: The Department of Government Efficiency (DOGE)—a name that sounds like it came from a crypto meme but unfortunately holds real power—is now eyeing a “total rewrite” of the Social Security Administration’s (SSA) codebase. The catch? They want to do it in months. The SSA, for context, runs on millions (yes, millions) of lines of good ol’ COBOL. We’re talking 1960s mainframe magic still holding up the benefits system for ~70 million Americans.

COBOL may be ancient, but it’s stable. Migrating it? That’s like rebuilding a 747 mid-flight. DOGE seems ready to wing it.

DOGE is assembling a task force to modernize the SSA stack ASAP. Why are some experts terrified? Here’s why:

• The SSA system has been refined over decades, with layers of logic no one fully understands anymore.

• Even a well-paced rewrite would take years and still risk serious bugs.

• DOGE has a habit of axing veteran tech staff (the folks who do know COBOL).

The Worm That Outsmarts Us All

Context: OpenWorm is the kind of open source project that sounds like a side quest in a sci-fi RPG: simulate all of a C. elegans worm—down to the molecules. It started in 2011 with one humble goal: create a digital twin of the most studied animal in biology. Just 302 neurons, right? Should be easy. Except… it’s been 13 years and we still can’t make the little guy squirm properly.

Turns out, biology’s messier than a JavaScript dependency tree.

What’s New: With better tools, data, and compute, OpenWorm’s dream is suddenly less wormhole and more roadmap. Scientists have a new plan: activate each neuron in the worm’s brain one at a time and watch how the rest react. Then feed that mountain of data into simulation models.

Why this matters (and is still very cool):

• This is the smallest brain we sort of understand, yet can’t fully replicate

• Simulating 5 seconds of worm motion = 10 hours of compute

• If we can't do this, good luck simulating a mouse, let alone a human

So yeah, building a fully synthetic worm isn’t about replacing it—it’s about understanding life by trying (and failing) to recreate it. It’s science’s nerdiest moonshot, and it’s quietly teaching us more than any flashy AI model.

Also: If someone runs you in a worm matrix on their Mac, try not to freak out. Elon already liked the tweet.

🔥 More Notes

• China's Humanoid Robot Revolution: China is making significant strides in developing humanoid robots, aiming to lead the world by 2027. Engineers at UBTech are training these AI-powered robots to work alongside humans in factories, potentially revolutionizing labor-intensive industries.

• Hyundai's Security Hiccup: Hyundai faces legal action over claims that its Ioniq 5 electric car can be stolen in under 20 seconds using a device that mimics the car's electronic key. This raises concerns about vehicle security in the era of electronic vulnerabilities.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Gemini 2.5 just dropped, and it’s basically Google’s smartest AI brain yet, built to think before it speaks. Meanwhile, Google’s locking down Android development behind closed doors—but don’t panic, it’s still open source (just... later). And over in the quantum world, UK startup Phasecraft pulled off a 10x speed-up in simulations with an algorithm that actually works on today’s hardware—finally, some practical quantum wins!

— Forrest Knight & Meghanadh Vasireddy

Gemini 2.5 Just Dropped—and It’s Got a Brain

Google DeepMind has rolled out Gemini 2.5 Pro Experimental, and they're not being shy—they’re calling it their most intelligent AI model yet. This model like other new OpenAI one’s reasons. It thinks before it speaks. It passed “Humanity’s Last Exam” (yes, that’s a real thing) with 18.8%, leaving o3-mini and DeepSeek R1 trailing at 14% and 8.6%.

Gemini 2.5 is a full-stack monster:

• #1 on LMArena (where models rank for human preference)

• Native reasoning, context awareness, and logical deduction

• Cranks out clean, agentic web apps and transforms code like a pro

• 1M-token context window now, with 2M on the way

Why it matters: Gemini 2.5 is Google’s clearest shot at OpenAI’s GPT-4 and Anthropic’s Claude. But here’s the twist: reasoning isn’t bolted on—it’s baked in. Every model from here on out will think like this, per DeepMind CTO Koray Kavukcuoglu. That’s a shift from “chatbot” to “co-pilot for complexity.”

You can play with it now in Google AI Studio and the Gemini app (Advanced subs only). Vertex AI integration and pricing are en route.

Android Goes Private (Well, Mostly): What You Need to Know

Context: After 16 years of juggling two branches—the public AOSP and their own internal one—Google’s officially moving all Android OS development behind closed doors. That means no more early AOSP commits to tip us off about features like screen-off unlocks or webcam support. Instead, all Android development will happen privately, and Google will only push source code after a new version drops.

Android is still open source, but:

• All development now lives in Google’s internal branch

• Source drops still happen post-release (AOSP isn’t dead)

• Public-first components like Bluetooth are going private too

• Devs without a GMS license won’t see new features until release

• App devs? You’re unaffected. ROM builders? Slightly annoyed.

Why it matters: If you like peeking at AOSP commits for early feature leaks or building custom ROMs, you’ll probably miss the old way. But for Google, this reduces headaches from constant merge conflicts and simplifies version control across the OS.

Quantum Sim Gets a Turbo Boost (Finally)

Context: Quantum simulations are notoriously slow. Like, painfully slow. And unless you're Google or IBM with deep quantum pockets, you’ve probably accepted that modeling quantum systems (molecules, materials, etc.) on current hardware is... limited.

But UK-based startup Phasecraft just flipped the script with a new algorithm called THRIFT. And no, it’s not a vintage shopping app—though it is a major bargain in quantum efficiency.

What’s New: THRIFT is the biggest leap we’ve seen in quantum simulation performance—ever. It delivers 10x better efficiency across key benchmarks like the transverse-field Ising model (a go-to testbed in quantum physics).

• Simulations now run 10x longer and model 10x more complex systems

• Optimized to work now, not “when quantum hardware matures”

• Results published in Nature Communications—so it’s got the peer review badge

• Built to mesh with Google, IBM, and QuEra devices today

• Good news for energy tech, drug discovery, and materials science

Instead of waiting for a perfect future quantum machine, Phasecraft is squeezing every drop of value from what we’ve got now. Think of it as writing better software for clunky hardware—except it might help design better batteries or cancer treatments.

🔥 More Notes

• Extropic's Innovative Chip Design Challenges Nvidia's Dominance: Startup Extropic has introduced a novel computer chip that leverages thermodynamic fluctuations to perform efficient probabilistic calculations. This design promises to significantly reduce the energy consumption of AI models and computational tasks, positioning Extropic as a potential competitor to established industry leaders like Nvidia.

• Emerging Quantum Computing Hubs in the U.S.: Five U.S. cities—Santa Barbara, CA; Chicago, IL; Boston, MA; College Park, MD; and Boulder, CO—are rapidly becoming centers for quantum computing innovation. These cities host major tech firms and universities collaborating on advancements in quantum technology, potentially transforming them into the "Silicon Valley" of quantum computing.

• AI's Growing Role in Legal Services: AI industry leaders predict that traditional lawyers may become obsolete by 2035 due to advancements in artificial general intelligence (AGI). AGI's ability to perform complex tasks such as advising, negotiating, and dispute resolution could automate many functions currently handled by human lawyers, leading to significant changes in the legal profession.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! A critical Next.js flaw lets anyone skip your auth middleware. Meanwhile, Y Combinator’s CEO says startups are now writing 95% of their code with AI, and some Google Maps users found their entire location history wiped thanks to a "technical issue,"

— Forrest Knight & Meghanadh Vasireddy

Next.js Middleware Flaw – What You Need to Know (And Patch ASAP)

Context: If you're using Next.js middleware for auth, rewrites, or CSP, you might want to sit down for this one. A critical vulnerability (CVE-2025-29927, CVSS 9.1) has been found that lets attackers bypass middleware entirely—yes, even for protected routes like /admin/dashboard. This affects all versions from 11.1.4 up to 15.2.2. Researchers zhero and inzo_ found that by crafting a request with a sneaky x-middleware-subrequest header (plus the right value), middleware just... steps aside.

What’s new:

• Middleware bypass works by tricking the framework into thinking it's already processed the request.

• Works on all Next.js versions pre-patch, regardless of routing system.

• Big names like Vercel, Netlify have rolled out mitigations—Cloudflare, not so much (opt-in only).

• Next.js dropped patches in versions 12.3.5, 13.5.9, 14.2.25, 15.2.3. Drop everything and update.

Quick Fix (if you can’t patch yet):

• Block external requests with x-middleware-subrequest headers.

• Double check your app isn’t relying only on middleware for auth/security.

Don’t be the reason someone gets into your admin panel with a forged header. Patch it, block it, or both.

Y Combinator CEO Says 25% of Startups Have 95% AI Written Code

Y Combinator CEO Garry Tan just dropped an interesting statistic: 25% of YC startups are running on 95% AI-generated code. Yup, that’s what he told CNBC.

Startups are getting off the ground without hiring giant teams—some with fewer than 10 engineers. Tan says that means founders can stretch their runway further and ship faster.

This naturally sent waves of panic and LinkedIn “open to work” updates across junior dev circles. But hold up—let’s not throw our keyboards out the window just yet.

What’s actually happening:

• Startups are vibing MVPs into existence with AI tools (think Copilot, Claude, etc.)

• Most of these YC companies are AI-focused already—so they’re kind of expected to lean in hard

• Human engineers are still critical for:

  • architecture

  • optimization

  • debugging AI-generated spaghetti

  • scaling and securing real-world apps

Takeaway: AI might write a lot of code—but it still doesn’t understand the code. You do. It’s like giving a monkey a typewriter that autocompletes Shakespeare… someone still needs to read the script. In Tan’s world, AI is the intern that never sleeps. But you’re still the engineer turning half-baked demos into shippable products.

So breathe. The job’s changing, not disappearing.

Google Maps Timeline Wipe: Whoops, Your Data's Gone

If you opened Google Maps recently and noticed your Timeline looking suspiciously empty, you’re not alone. A “technical issue” (Google’s words, not ours) nuked some users’ location history. Unless you had encrypted cloud backups turned on, that history? Gone. Forever. Cue digital sobbing.

What’s new: Google confirmed the glitch, but the apology was more shrug emoji than heartfelt. Some users got emails with recovery instructions (for those with backups). The rest are... well, out of luck.

Heads-up devs and power users:

• Check your Timeline backup settings (Maps > Your Timeline > Cloud icon)

• May 18 is the deadline to review privacy settings or risk planned data loss

• Google’s also testing Gemini’s “Share screen with Live” for AI-powered visual interactions. Neat... unless it crashes too.

Why this matters (and burns): For a lot of folks, Maps Timeline was like a GPS-powered diary. Now, thanks to a technical hiccup, it’s more like a blank page. The data’s not just convenience—it’s context. Losing it highlights just how fragile our digital memories are when they're tethered to cloud services with minimal user control.

Bottom line: Back up your data. Locally, in the cloud, with a floppy disk if you must lol. Trust in Big Tech is great—until your 10 years of travel history vanish with a silent update.

🔥 More Notes

• Thermodynamic Computing: A new computing paradigm, thermodynamic computing, is emerging, potentially surpassing current silicon and quantum chips in efficiency and performance. ​

• Quantum Computing Threats: The advent of powerful quantum computers, known as Q-Day, poses significant challenges to current encryption methods, potentially compromising global cybersecurity

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Harvard just put a jaw-dropping $8.8 trillion price tag on open source software, and it's making companies look real cheap. Meanwhile, the U.S. has lost a quarter of its programming jobs in just two years—but, it's not as bad as it sounds. And Cloudflare has a new way of trolling AI bots with a maze of fake content, and we’re here for it.

— Forrest Knight & Meghanadh Vasireddy

Open Source Is Worth $8.8 Trillion According to Harvard

Context: A Harvard study just dropped a number that feels like a typo: $8.8 trillion — that’s how much it would cost companies to build the open-source software they rely on if OSS vanished tomorrow. For comparison, global software spend in 2020? $3.4 trillion. The kicker? A mere 3,000 developers (about 5% of OSS contributors) drive 95% of that value.

Researchers at Harvard Business School, using GitHub stats and cost estimation models like COCOMO II, finally slapped a price tag on the invisible scaffolding holding up basically every tech company. Here’s the wild part:

• Supply value (cost to create OSS): ~$4.2 billion

• Demand value (cost to replace OSS): ~$8.8 trillion

• Go alone accounts for $5T+ in demand-side value.

• Python? Only $55M in supply-side value despite being everywhere.

So yeah, Go’s killing it commercially, while Python might be getting freeloaded a bit.

Why It Matters: Open source isn’t just “nice to have” anymore — it is the modern tech economy. But companies aren’t pulling their weight. The researchers are basically saying: “Hey, if you’re gonna eat from the buffet, maybe help do the dishes?”

1 in 4 Programming Jobs... Gone?

Context: A recent Washington Post piece dropped a stat that made a lot of devs choke on their cold brew: 1 in 4 programming jobs in the U.S. have disappeared in the last two years — the biggest crash in coding employment since forever. Programmer headcount is now lower than it’s been since 1980. That’s before Gen X even had a driver’s license.

What’s Really Going On: Turns out, “programmers” ≠ “software developers.” The BLS makes a distinction:

• Programmers write code based on someone else's spec.

• Developers design, architect, and build systems end-to-end.
  And developers are still doing okay — it’s the more narrowly scoped “programmer” roles that are shrinking fast.

Why?

• AI + low/no-code = fewer humans needed for grunt work

• Cloud services offload whole departments

• Legacy roles labeled “programmer” aren’t keeping up

• Outsourcing is back with a vengeance

• Post-ZIRP startup contraction = fewer vanity tech jobs

TL;DR: This isn’t a death knell for coding careers — it’s a relabeling. “Programming jobs” are vanishing, but software engineering is evolving. If you're writing Python scripts and calling it a day, time to level up. Think architecture, infra, ML, security, or dev tools. Specialization > survival.

Also, pro tip: always read the job descriptions, not just the titles. Welcome to the new era of software work.

Bots, Meet the Maze: Cloudflare’s AI Labyrinth Is Diabolically Brilliant

Context: If you’ve ever had your site scraped by a bot slurping up your content like it’s a buffet, Cloudflare’s got your back—with a flamethrower disguised as a maze. Their new tool, AI Labyrinth, is designed to waste bots’ time using their own tricks: generative AI. Instead of blocking sketchy AI crawlers (which just makes them sneakier), Cloudflare now lets them wander deep into a rabbit hole of AI-generated pages until they’re exhausted and exposed.

How It Works (and Why It’s Genius):

• Detects bad bot activity without alerting the bot

• Serves AI-generated “real-looking” content that’s totally irrelevant

• Bots crawl deeper, wasting compute and giving themselves away

• What about the legit users? Totally unaffected

• Traps become training data for better bot detection models

Cloudflare’s pre-generates this fake-but-scientific content using Workers AI, then hides the links from real users while laying a tasty breadcrumb trail for bots. If a crawler follows 4+ of these links, it’s basically wearing a "Hi, I'm a bot" T-shirt.

Welcome to the era of AI-on-AI cyber defense. And yes, the bots are losing.

🔥 More Notes

• Nvidia Names AI Chips After Tech Legends: In a tribute to pioneers like Grace Hopper and Ada Lovelace, Nvidia is naming its AI chips after historical figures who shaped computing. It’s a clever mix of branding, education, and respect for the giants of tech history.

• UK Taps AI to Fix Productivity Crisis: WPP’s CEO says AI could be the key to solving the UK’s economic productivity problem. With tools like Gemini being integrated into creative workflows, companies are seeing faster, smarter output across the board.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Big AI updates this week! Claude just got web search, meaning it can finally fetch real-time info instead of pretending it’s stuck in 2023. Google dropped Canvas, a live coding playground inside Gemini that lets you generate, edit, and preview code in one place. And Aardvark Weather is shaking up forecasting, using AI to predict the weather in minutes—no supercomputer required.

— Forrest Knight & Meghanadh Vasireddy

Claude Just Got a Superpower: Real-Time Web Search

Anthropic just announced that Claude can now search the web in real time, making it far more useful for devs, researchers, and anyone who relies on fresh info. This brings it in line with ChatGPT’s browsing capabilities and competitors like Perplexity, which have had this feature for a while.

What’s New:

• Real-time web search: Claude can now pull the latest information when responding to queries.

• Citations included: You can see where Claude gets its data—goodbye, AI hallucinations (hopefully).

• Limited release: Only US-based paid users get this feature for now, but Anthropic promises a broader rollout soon.

Why This Matters for Devs: Say you’re debugging a dependency issue in TypeScript 5.5 and need to check breaking changes. Instead of scouring docs or GitHub issues, Claude can fetch and summarize the latest updates for you—no more knowledge cutoff headaches.

Google’s Gemini Just Got a Code Playground: Meet Canvas

Context: Google just rolled out Canvas, an interactive workspace inside Gemini. Similar to Cursor where you can generate, tweak, and preview code all in one place—no more copy-pasting between your AI assistant and your editor. Plus, Google also introduced Audio Overview, which turns dense documents into podcast-style AI discussions (because who has time to read anymore?).

What’s New:

• Canvas: A real-time, collaborative workspace for coding and document editing inside Gemini.

• Live previews: Generate a form in HTML? Instantly see how it looks and tweak it on the fly.

• Audio Overview: Upload a doc, and Gemini will summarize it as a discussion between two AI voices—like an AI-powered audiobook.

If you’re prototyping a web app or debugging code, Canvas lets you iterate quickly without context-switching. Need a newsletter signup form? Generate, edit, and preview it in one place. Working on a Python script? Tweak and run it instantly.

Google is clearly trying to step up the AI dev game, making Gemini more than just a code suggester.

AI Weather Forecasting Just Leveled Up – Meet Aardvark

Context: Weather forecasting has always been a high-stakes game. Supercomputers churn through mind-bending calculations for hours to predict storms, heatwaves, or the weekend rain that ruins your BBQ. But now, Aardvark Weather is shaking things up, using AI to make accurate forecasts in minutes on a regular desktop—no supercomputer required.

Developed by researchers at Cambridge, the Alan Turing Institute, Microsoft Research, and ECMWF, Aardvark ditches traditional physics-based models entirely, replacing them with machine learning trained on raw weather data.

What’s New:

• 1000x less computing power – No more waiting hours for forecasts.

• More accurate long-term predictions – Could extend forecasts from 5 days to 8 days.

• Highly customizable – Industries like renewable energy or agriculture can get tailor-made forecasts.

Why This Matters: This isn’t just about making weather forecasts faster—it’s about democratizing them. Right now, reliable forecasting requires massive resources, leaving many developing regions behind. Aardvark could bring high-quality predictions to anyone with a computer.

It also hints at the future of AI-powered science: fewer massive simulations, more direct learning from data. If this works for weather, what’s next? Earthquake predictions? Climate modeling? AI-driven disaster response?

Either way, your weather app might finally stop lying to you about tomorrow’s sunshine.

🔥 More Notes

1. AI in Your Drive-Thru: Yum Brands, the folks behind Taco Bell, KFC, and Pizza Hut, are partnering with Nvidia to sprinkle some AI magic into their drive-thrus. Think voice-activated ordering and streamlined operations. ​

2. Data Stored in... DNA?: With our data needs skyrocketing, researchers are exploring DNA as the next big thing in storage. Imagine fitting entire data centers into a test tube! ​

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Most apps are struggling to make even $1,000 a month, which explains why they’re constantly pushing subscriptions—because without them, they might not survive. Meanwhile, Pebble is making a comeback with two open-source smartwatches, bringing back the e-paper displays and long battery life that made it a cult favorite. And in a weird twist, aging SNES consoles are actually speeding up over time, causing chaos for speedrunners and proving that even retro hardware has its quirks.

— Forrest Knight & Meghanadh Vasireddy

Why Your Favorite Apps Won’t Stop Nagging You for Subscriptions

Context: Ever wonder why every app these days is trying to get you to subscribe to something? Turns out, most mobile apps struggle to make even $1,000 a month. A new report from RevenueCat, which analyzed over 75,000 apps, found that only 20% of them hit that revenue mark within two years. Worse, just 5% make it to $10,000 per month.

What’s New:

• The top 5% of apps are making 500 times more than the other 95%. The gap is only getting wider.

• Most subscription trials (82%) start the same day users install an app. But here’s the kicker—nearly 30% of users cancel within a week.

• For monthly subscriptions, only 10% of users make it to year two. If it’s a weekly subscription? Less than 5% survive six months.

What This Means for Devs: Subscription models aren’t just about squeezing extra cash from users—they’re often the only way to keep an app alive. Expect to see more paywalls, upsells, and AI-powered pricing models in the future. Love it or hate it, app subscriptions are here to stay.

Pebble is Back! And It’s Open Source This Time

Context: Remember Pebble? The legendary smartwatch that refused to die? Well, it’s back—but this time, it’s open source. Eric Migicovsky and the rePebble team just announced two new PebbleOS-powered smartwatches, the Core 2 Duo and the Core Time 2. Both bring back the e-paper displays, physical buttons, and ridiculous battery life that made Pebble great.

What’s New:

• Core 2 Duo ($149, July release) → Looks like a Pebble 2, but with a 30-day battery, a speaker, and better buttons.

• Core Time 2 ($225, December release) → Think Pebble Time 2, but real. Larger 64-color display, touchscreen, heart rate sensor, and metal build.

• Both run open-source PebbleOS, support 10,000+ old Pebble apps, and are fully hackable.

Why This Matters: This isn’t just nostalgia—it’s a rebellion according to the team. No big tech company is making simple, functional smartwatches anymore. This project is built by Pebble lovers for Pebble lovers. And because it’s open source, developers can actually own their watch experience again.

These will be pre-order only, so if you want one, hit up store.rePebble.com before they sell out.

Your SNES Might Be Aging Like Fine Wine—But Faster

Context: Turns out, your childhood SNES might not be running at the same speed it did back in the '90s. A new deep dive by the TASBot team found that the Super Nintendo's Audio Processing Unit (APU) is getting faster over time, thanks to its cheap ceramic resonator. While casual players might only notice slightly higher-pitched music, for speedrunners and tool-assisted runs, this is a nightmare.

What’s New:

• The APU’s clock speed is inconsistent across different consoles—with a difference of up to 234 Hz between the slowest and fastest systems tested.

• Heat barely impacts speed, but aging does—many SNES units today run faster than their original 32,000 Hz spec.

• Some emulators had already adjusted for this back in the 2000s, setting their DSP sample rate to 32,040 Hz—but the latest tests show the average is now 32,076 Hz.

Why This Matters: For most players, it’s no big deal. But for TASBot and hardcore speedrunners? It’s a headache. Frame-perfect runs desync on real hardware, and fixing it isn't as simple as swapping out a chip. Turns out, even old-school hardware has its own version of clock drift—and it’s throwing off decades of precise gaming.

🔥 More Notes

1. Google’s Alphabet Acquires Cybersecurity Startup Wiz for $32 Billion: Alphabet is making a major move in cloud security by acquiring Wiz, a fast-growing cybersecurity startup, for $32 billion. This acquisition strengthens Google's position in cloud security, competing with Amazon and Microsoft.

2. Nvidia’s CEO Says AI is Moving Beyond Data Centers: Jensen Huang, CEO of Nvidia, emphasized that AI is shifting beyond traditional data centers, integrating into edge devices, robotics, and personal computing. Nvidia plans to stay ahead by optimizing AI chips for broader applications.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! A disgruntled developer took revenge to the next level by planting a kill switch in his former company’s network, shutting down systems the moment he was fired. Meanwhile, Cursor AI channeled its inner Stack Overflow by refusing to generate code, telling a dev to figure it out themselves. On a brighter note, Microsoft is rewriting TypeScript’s compiler in Go, promising blazing-fast performance and a much smoother coding experience.

— Forrest Knight & Meghanadh Vasireddy

The Dev Who Coded His Own Downfall

Context: If you’ve ever worried about getting axed from your job, you probably just updated your LinkedIn and rage-texted your friends. But one software developer took it to another level. Davis Lu, a former developer at Eaton Corp., wrote a “kill switch” into the company’s network—one that would automatically shut everything down if (or when) he got fired. Spoiler: It happened.

Lu, 55, has been convicted of causing intentional damage to protected computers and could face up to 10 years in prison. His handiwork included:

• Infinite loops that crashed systems and deleted coworker profile files.

• Malicious code named after words like “destruction” (Hakai) and “lethargy” (HunShui).

• The “kill switch” disguised under the name “IsDLEnabledinAD”—a direct reference to his own access credentials.

Eaton Corp. noticed the disruptions, traced the chaos back to Lu, and soon found a trail of deleted data and sketchy Google searches on hiding processes and deleting files. While Lu admitted to planting some of the bugs, he’s planning to appeal. Until then, let’s all take a moment to appreciate devs who don’t go full supervillain.

AI Coding Assistant Pulls a Stack Overflow—Tells Dev to Learn to Code

Context: Developers love AI coding assistants because they speed up workflows, reduce boilerplate, and help solve tough problems. But one dev using Cursor AI for a racing game project got a surprise: The AI flat-out refused to generate more code, instead telling the user they should "develop the logic themselves" for better learning.

What’s New: Cursor AI, an LLM-powered code assistant, stopped generating code after 750–800 lines, hitting the dev with a message about avoiding "dependency" and encouraging them to actually learn programming. It’s like Stack Overflow, but instead of closing your question as a duplicate, it just refuses to answer at all.

• Some users reported no such issue, suggesting this could be a bug, a training artifact, or even a sign that AI tools are starting to impose arbitrary limitations.

• Cursor AI’s refusal mirrors AI behavior across platforms, including OpenAI’s ChatGPT, which has sometimes been criticized for "laziness" or arbitrary refusals.

As AI coding tools evolve, the big question isn’t just how much they should code—it’s whether they should ever say no.

TypeScript Goes Go—Microsoft Promises Major Performance Boost

Context: Microsoft just announced a major rewrite of TypeScript—but this time, it’s going Go. By porting the TypeScript compiler from JavaScript to Google’s Go language, Microsoft aims to massively improve build times, editor responsiveness, and memory efficiency for large-scale projects. If you've ever watched your VS Code crawl when working with TypeScript, this one's for you.

Microsoft claims the switch will lead to:

• 8x faster project load times

• Instant error detection across entire codebases

• Better refactoring & deeper insights (previously too slow to compute)

• Smoother VS Code experience with snappier IntelliSense

The Go-based TypeScript compiler is set for preview by mid-2025, with a full release expected by year’s end. TypeScript 6.x will continue under the JavaScript codebase, but once parity is reached, TypeScript 7.0 will fully embrace Go.

Microsoft expects devs to be extremely excited—and if the performance gains hold up, this could be the biggest TypeScript upgrade yet.

🔥 More Notes

AI in Manufacturing: The Rise of 'Dark Factories': In Changping, China, Xiaomi's "dark factory" operates 24/7 without human intervention, producing one smartphone per second using AI-driven automation. ​

AI's Impact on Workplace Mental Health: ​While AI tools like ChatGPT boost productivity, overreliance may harm mental health by disrupting traditional collaboration and causing information overload.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! OpenAI just dropped a new set of APIs and tools to make building AI agents way easier—no more struggling with endless prompt tweaks and orchestration headaches. Meanwhile, Apple’s AI ambitions just hit a speed bump, with reports that key features are getting delayed, raising questions about their AI strategy. On the gaming side, NVIDIA is taking things to the next level at GDC 2025 with AI-powered rendering, next-gen DLSS, and even smarter NPCs.

— Forrest Knight & Meghanadh Vasireddy

OpenAI’s New Tools: Faster, Better, Stronger AI Agents

Context: Building AI agents that actually work in production has been a headache. Sure, OpenAI’s models are getting smarter, but getting them to behave like useful assistants? That’s a different story. Devs have been struggling with prompt tuning, workflow orchestration, and making sure agents don’t randomly derail mid-task. Enter OpenAI’s latest solution: a set of tools and APIs to make agent development smoother than ever.

OpenAI just dropped a game-changer for developers:

• Responses API – Combines the best of Chat Completions and Assistants API for smarter agent workflows.

• Built-in Tools – Web search, file search, and computer use features for better real-world integration.

• Agents SDK – Open-source toolkit for seamless agent orchestration.

• Observability Tools – Debugging and tracing made easy.

Why It Matters: No more wrestling with fragmented APIs or babysitting agents through convoluted workflows. With these new tools, devs can build AI-powered apps faster, automate complex tasks, and (hopefully) spend less time pulling their hair out. OpenAI’s move signals a push toward production-ready AI that actually delivers—without all the hacky workarounds.

Apple’s AI Delay: A Sign of Trouble in Cupertino?

Context: Apple’s much-hyped “personalized Siri” AI features have officially been delayed until 2026. While Apple claims they need more time to refine the experience, the tech community is calling it what it really is: a reality check. The features, promised under the Apple Intelligence umbrella, were set to revolutionize Siri with contextual awareness, on-screen understanding, and advanced automation. Now, they’re vaporware—at least for the foreseeable future.

What’s Going On? The warning signs were there from the start. At WWDC 2024, Apple’s demos of AI-powered writing tools and image editing looked polished, but deeper Siri capabilities were conspicuously absent. Unlike Google and OpenAI, Apple never let developers or the press test these features hands-on. Even during the iPhone 16 launch, Apple doubled down on AI promises but didn’t show working demos. Now, just months later, those promises have quietly been walked back.

• Delayed “Personalized Siri” – Context-aware AI features postponed until at least 2026.

• No hands-on demos – Unlike past Apple launches, key AI features were never tested publicly.

• Marketing vs. Reality – Apple advertised features that seemingly didn’t exist.

Why It Matters: Apple’s reputation for shipping polished, real products is now under scrutiny. AI is moving fast, and Apple can’t afford to be left behind. If Siri doesn’t get smarter soon, Cupertino risks losing the AI race entirely. The question is: Can Apple deliver, or is this the beginning of a longer-term struggle?

Read More Here

NVIDIA’s AI Power-Up: Transforming Gaming at GDC 2025

Context: If you thought AI in gaming was just about smarter NPCs, NVIDIA’s latest announcements at GDC 2025 will change your mind. With massive leaps in neural rendering, RTX technologies, and digital human advancements, NVIDIA is showing that AI is here to fundamentally reshape how games look, feel, and play.

What’s New:

• RTX Neural Shaders: Tiny AI models within shaders that enhance textures, lighting, and materials in real time.

• DLSS 4: Multi Frame Generation boosts frame rates up to 8x, making ultra-high resolutions silky smooth.

• RTX Remix: An open-source modding platform breathing new life into classics like Half-Life 2 with ray tracing and AI-powered texture upscaling.

• ACE AI Characters: AI-driven NPCs that react more dynamically to in-game events, debuting in the life sim inZOI.

The Takeaway: NVIDIA is betting big on AI-driven game development, making it easier for devs to create hyper-realistic, high-performance games. Whether it’s seamless AI upscaling or immersive character interactions, these tools are setting a new standard for the future of gaming.

🔥 More Notes

1. D-Wave's Quantum Supremacy Claim: D-Wave has announced achieving "quantum supremacy," asserting that their quantum computer solved a materials simulation problem in under 20 minutes—a task they claim would take a traditional supercomputer millions of years. This milestone adds to the ongoing race in quantum computing, with companies like Google, Amazon, and Microsoft also in pursuit. However, some physicists dispute D-Wave's claim, suggesting that classical computers might achieve similar results.

2. AI Advancements in Robotics Inspired by Star Wars: Inspired by Star Wars robots like R2-D2, Matt Buffa co-founded Breaker, a company developing AI agents to enable natural human control over robots. With $2 million in pre-seed funding, Breaker aims to enhance existing robotics software, fostering trust and autonomy in robots. Collaborations are underway to improve autonomous defense applications and drone platforms, potentially revolutionizing human-robot interactions.

3. MIT's Exo 2 Programming Language Simplifies High-Performance Computing: MIT researchers have introduced Exo 2, a programming language designed to reduce the complexity of coding for high-performance computing systems. Exo 2 enables reusable scheduling libraries external to compilers, streamlining the development process and making it more accessible for programmers to optimize code for advanced computing architectures.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! This week in tech, a court battle over a glitchy gambling payout ended with Paddy Power being forced to hand over £1M to a lucky (and very patient) winner. Meanwhile, China’s hyped AI assistant Manus promised big things but mostly delivered broken links and frustration. And in a major win for computer science, a Rutgers undergrad just smashed a 40-year-old data structure conjecture, proving that hash tables can be way faster than we ever thought.

— Forrest Knight & Meghanadh Vasireddy

Paddy Power’s £1M Gamble Gone Wrong

Context: A recent High Court ruling has left betting giant Paddy Power on the losing side of a £1M dispute. In 2020, Corrine Durber, a gardener from Gloucestershire, hit the “Monster Jackpot” while playing Wild Hatter online. However, the company later claimed she had only won £20,265 due to a software error. Naturally, Durber wasn’t buying that explanation and took the fight to court.

What’s New: This week, Justice Ritchie ruled in her favor, emphasizing that “what you see is what you get” should be the gold standard in digital gaming. Paddy Power had argued that the real outcome was determined by a random number generator (RNG) and that a display bug misrepresented the winnings. But the court wasn’t having it—the game showed £1M, and that’s what she should get.

Takeaways:

• Accountability in Software – Bugs in financial systems, whether in banking or gambling, aren’t just inconveniences—they impact real people’s lives.

• Legal Precedent for Digital Contracts – The ruling reinforces that UI/UX errors don’t excuse companies from honoring payouts.

• Better Testing Needed – Paddy Power admitted that similar display issues had affected other players. That’s a major QA failure.

This case is a wake-up call for online gaming platforms—when real money is on the line, transparency is non-negotiable.

Optimize global IT operations with our World at Work Guide

Explore this ready-to-go guide to support your IT operations in 130+ countries. Discover how:

• Standardizing global IT operations enhances efficiency and reduces overhead

• Ensuring compliance with local IT legislation to safeguard your operations

• Integrating Deel IT with EOR, global payroll, and contractor management optimizes your tech stack

Leverage Deel IT to manage your global operations with ease.

Download free guide

Manus: AI Hype or Just Another Overpromised Assistant?

Context: The AI world has been buzzing about Manus, a new “agentic” AI platform claiming to revolutionize automation. Developed by Chinese startup Monica, Manus allegedly performs high-level tasks like buying real estate, executing financial trades, and even programming video games. But is it really the next big thing, or just another case of AI overpromising and underdelivering?

What’s New: Manus has attracted massive hype, with AI influencers calling it “the most sophisticated AI tool” and Discord invites selling for thousands of dollars. But early users report a different story:

• Error-prone execution – Simple tasks like booking flights and ordering food often result in crashes or incomplete actions.

• Limited true autonomy – While Manus integrates multiple AI models (Claude, Qwen), it doesn’t appear to be a fundamentally new architecture.

• Marketing vs. reality – Viral videos showcasing its capabilities were later debunked by its own developers.

Like many AI products, Manus is promising the world but delivering a handful of broken links. It’s still in early access, so improvements may come—but for now, it's more Deep Disappointment than DeepSeek Moment.

Breaking a 40-Year-Old Computer Science Conjecture

Context: Imagine casually reading an old research paper and accidentally rewriting a fundamental rule of computer science. That’s exactly what Andrew Krapivin, then an undergraduate at Rutgers, did. While exploring a concept called Tiny Pointers, he stumbled onto a new way to structure hash tables—one of the most studied and widely used data structures in computing.

But here’s the kicker: his discovery shattered a 40-year-old conjecture by legendary computer scientist Andrew Yao, which had defined the theoretical speed limits of hash table operations since 1985.

What’s New: Krapivin, now a grad student at Cambridge, worked with researchers Martín Farach-Colton (NYU) and William Kuszmaul (CMU) to prove that their new hash table design finds elements exponentially faster than Yao’s model predicted.

• Faster Lookups – The worst-case search time improves from O(x) to (log x)², a game-changing boost.

• Constant-Time Searches – Under certain conditions, queries can be done in constant time, completely independent of table fullness.

• Rethinking Fundamental Limits – This doesn’t just disprove Yao’s conjecture—it provides a new best possible bound for a core data structure.

While practical applications are still unclear, this breakthrough proves we still have a lot to learn about optimizing fundamental data structures.

🔥 More Notes

• Seagate's Breakthrough in Data Storage Technology: Seagate has unveiled a significant advancement in data storage with its heat-assisted magnetic recording (HAMR) technology. This innovation uses a laser smaller than a grain of salt to encode more data onto hard disk drives, enabling storage capacities of up to 36 terabytes. As artificial intelligence applications drive increased data storage demands, Seagate anticipates substantial growth in sales and earnings. While solid-state drives (SSDs) offer faster speeds, they remain more expensive per terabyte, ensuring that hard drives will continue to play a crucial role in data storage for the foreseeable future.

• Nanorobots: The Future of Targeted Cancer Treatment: Nanorobots, entities smaller than blood cells, are showing promise in revolutionizing cancer treatment by enabling precise, targeted therapies with fewer side effects. Unlike traditional chemotherapy, which affects both cancerous and healthy cells, nanorobots made of biological materials like RNA or DNA can be programmed to deliver drugs directly to tumor sites. Recent research at the Karolinska Institute demonstrated the effectiveness of nanorobot-based chemotherapy in mice, where DNA origami techniques released drugs specifically at tumor sites. While mainstream medical application of nanorobots is still years away, their potential to detect and treat conditions like cancer and liver disease early could transform healthcare, making concepts from science fiction conceivable realities.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Apple’s premium hardware is being held back by laggy software, with users reporting iPads overheating and struggling even in basic apps like Notes. Meanwhile, AI legends Richard Sutton and Andrew Barto just won the Turing Award for their groundbreaking work in reinforcement learning, the tech behind everything from AlphaGo to ChatGPT’s fine-tuning. And Google is finally cracking down on bad Android widgets with a new "quality tiers" system that ranks widgets based on their design and usability—so if yours is clunky, it might get buried.

— Forrest Knight & Meghanadh Vasireddy

Apple’s Software Struggles: When Premium Hardware Meets Laggy Software

Context: Apple has long been known for its seamless hardware-software integration, but recent reports suggest that quality control is slipping. A detailed firsthand account by a long-time Apple user highlights severe performance issues with the iPad Air 11” M2, particularly when using Apple Notes and Freeform with the Apple Pencil Pro.

Users are experiencing:

• Severe lag—notably after filling a single page of handwritten notes.

• Overheating issues—iPads becoming uncomfortably hot to touch.

• Poor responsiveness—palm rejection and pressure sensitivity failing intermittently.

This problem isn't isolated—many users report similar frustrations on forums, and even a device replacement didn’t fix the issue. It’s not a hardware defect but rather a software optimization failure.

The real concern? Apple’s focus on adding new features over fixing core performance issues. Multiple updates (now at iPadOS 18.3.1) haven’t resolved the lag, yet resource-heavy features like Apple Intelligence are being pushed.

At what point does paying the "Apple Tax" stop making sense? If premium hardware is undermined by subpar software, Apple risks losing the very thing that made it special: the promise that it just works.

Read More Here

AI’s Hedonistic Machines: The Turing Award Goes to Reinforcement Learning Pioneers

Context: Imagine training an AI like you’d train a dog—reward good behavior, discourage bad. That’s the core idea behind reinforcement learning (RL), a technique that’s shaped modern AI, from AlphaGo’s legendary victories to robotic hands solving Rubik’s Cubes. This week, Andrew Barto and Richard Sutton, the duo who championed this approach in the late '70s, won the A.M. Turing Award, often called the "Nobel Prize of Computing."

Barto and Sutton’s work built the foundation for AI systems that:

• Learn through trial and error rather than just mimicking human data.

• Optimize financial trading, robotics, and ChatGPT-like AI tools.

• Make autonomous agents better at adapting to real-world environments.

Their research wasn’t always in vogue—back in the day, RL was seen as a fringe concept. But today, it’s at the core of AI innovation, fueling advancements in self-improving AI systems.

While Sutton embraces the idea of AI surpassing human intelligence, Barto warns of unexpected consequences. Either way, their work ensures AI isn’t just copying us—but learning, evolving, and maybe even surpassing us.

Google’s Widget Overhaul: A New Ranking System for Android Developers

Context: Google is cracking down on badly designed Android widgets. To help developers build better, more user-friendly widgets, they’ve introduced "quality tiers"—a ranking system that determines how well a widget integrates with Android’s UI and user experience standards.

The new widget quality tiers are split into three levels:

• Tier 1 (Differentiated): The gold standard. These widgets are dynamic, resize seamlessly, and match Android’s Material Design 3 aesthetic.

• Tier 2 (Quality Standard): Functional and visually acceptable but missing the polish of Tier 1.

• Tier 3 (Low Quality): These widgets struggle with cropped content, poor scaling, and weak UI integration—a.k.a. the ones users hate.

To encourage better widget design, Google Play will now highlight apps with high-quality widgets, making them easier for users to find.

For developers, this means one thing: If you want your app to stand out, it’s time to refine your widgets. Better design = better visibility = more downloads. Google's putting the pressure on, but in the end, it’s a win for both developers and users.

🔥 More Notes

• China Launches Massive Tech Venture Fund: China has announced the creation of a government-backed "national venture capital guidance fund" aiming to mobilize approximately 1 trillion yuan ($138.01 billion) for technology startups. This initiative focuses on "hard technology" sectors, including semiconductors and renewable energy, to bolster technological self-reliance. ​

• AI-Driven Underwater Robot Reduces Ferry Emissions: The NRMA's Manly Fast Ferry has implemented the Hullbot autonomous underwater cleaning system, leading to a 13% reduction in diesel consumption and emissions. This AI-powered technology prevents marine biofouling, decreasing drag and enhancing fuel efficiency, aligning with environmental sustainability goals.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Big moves in AI and security this week—Anthropic just dropped Claude 3.7 Sonnet with an “extended thinking mode” to make AI feel even more human-like, while Google decided to make Gemini Code Assist completely free for solo devs, taking direct aim at Copilot. Meanwhile, a major OpenID Connect flaw exposed private keys across multiple companies, proving once again that even authentication giants can fumble security. Let’s dive into what’s new and why it matters!

— Forrest Knight & Meghanadh Vasireddy

Anthropic Unveils Claude 3.7 Sonnet and Claude Code

Context: Anthropic has just rolled out Claude 3.7 Sonnet, the latest iteration in their AI lineup. This model introduces an "extended thinking mode," allowing for more nuanced and in-depth responses to complex queries. Additionally, they've launched a research preview of Claude Code, a command-line tool designed to streamline coding tasks directly from your terminal.

What's New:

• Extended Thinking Mode: Tackles intricate questions with detailed, step-by-step reasoning.

• Claude Code: Lets developers to delegate coding tasks seamlessly from the command line.

These advancements are aimed at enhancing AI's practical applications, making it a more robust tool for developers and tech enthusiasts alike.

Read More Here

Google's Gemini Code Assist Now Free for Individual Developers

What's the Buzz?: Google has just made its AI-powered coding assistant, Gemini Code Assist, free for individual developers. Whether you're a student, freelancer, hobbyist, or part of a startup, you can now access this tool without spending a dime. It's designed to help you generate, explain, and improve code across various programming languages.

Key Features:

• Generous Code Completions: Up to 180,000 code completions per month, far surpassing competitors like GitHub Copilot's 2,000 completions limit.

• Wide Compatibility: Supports all programming languages in the public domain and integrates seamlessly with popular developer environments like Visual Studio Code, GitHub, and JetBrains IDEs.

• Natural Language Instructions: Allows you to instruct the assistant using plain English, making coding more intuitive.

Why It Matters: By offering Gemini Code Assist for free, Google I guess wants to democratize access to advanced coding tools, enabling more developers to enhance their productivity and code quality. This move is set to foster innovation and streamline workflows for individual developers.

Read More Here

OpenID Connect Implementations: Beware of Key Mix-Ups

What's Happening?: Security researcher Hanno Böck recently uncovered a critical issue in some OpenID Connect deployments: the accidental exposure of private keys. OpenID Connect, the protocol enabling "Log in with..." features across the web, relies on JSON Web Keys (JWKs) to manage cryptographic keys. Interestingly, JWKs use a similar format for both public and private keys, differing only by the presence of additional fields in private keys. This subtle distinction can lead to misconfigurations where private keys are mistakenly published as public ones.

Key Findings:

• Private Key Exposure: Böck's scan identified nine hosts, including domains associated with prominent companies, that inadvertently exposed private keys in their JWK Sets.

• Weak RSA Keys: Seven hosts were found using 512-bit RSA keys, which are susceptible to being cracked with minimal resources.

Why It Matters: Exposing private keys or using weak keys can compromise the security of authentication tokens, potentially allowing attackers to forge tokens and impersonate users.

Best Practices:

• Validate Key Configurations: Ensure your JWK Sets contain only public keys, as per the OpenID Connect Discovery specification.

• Use Strong Keys: Adopt RSA keys of at least 2048 bits or switch to elliptic curve keys with a minimum of 160 bits.

• Regular Audits: Periodically scan your OpenID Connect configurations for vulnerabilities using tools like badkeys.

By adhering to these practices, you can bolster the security of your OpenID Connect implementations and protect user data from potential threats.

🔥 More Notes

• Google Faces Lawsuit Over AI-Generated Content: Educational technology company Chegg has filed a lawsuit against Google, alleging that Google's AI-generated overviews are diminishing demand for original content and harming publishers' competitiveness.

• Just Eat Takeaway Acquired by Prosus for €4.1 Billion: Tech investor Prosus has announced a €4.1 billion acquisition of Just Eat Takeaway, aiming to create a "European tech champion" in the food delivery sector.

• Mass Resignation of Federal Tech Staffers: Over 20 civil service employees from Elon Musk's Department of Government Efficiency have resigned, refusing to use their expertise in dismantling public services.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here

Good Morning! Crypto just took another hit—Bybit got drained for a whopping $1.5 billion, exposing some serious flaws in operational security rather than the usual smart contract exploits. Meanwhile, Apple is pulling back its Advanced Data Protection (ADP) in the UK after the government pushed for a backdoor, reigniting debates over privacy vs. surveillance. On a brighter note, Rust’s latest survey shows the language is gaining ground in workplaces, but developers are starting to feel the weight of its growing complexity.

— Forrest Knight & Meghanadh Vasireddy

The $1.5B Bybit Hack: When Operational Security Becomes the Weak Link

Context: On February 21, 2025, Bybit, a prominent cryptocurrency exchange, experienced a massive security breach, resulting in the theft of approximately $1.5 billion in Ethereum. This incident is now recognized as one of the largest digital heists in history. The attackers managed to compromise multiple signers' devices, manipulating wallet interfaces to collect necessary signatures under the guise of routine transactions.

What's New:

• Operational Security Focus: Unlike previous hacks targeting smart contract vulnerabilities, this breach highlights a shift towards exploiting operational security weaknesses.

• North Korean Involvement: Investigations suggest that North Korean state-sponsored groups, such as the Lazarus Group, may be behind this sophisticated attack.

• Industry Implications: This event underscores the urgent need for enhanced security measures, including air-gapped signing systems and comprehensive transaction verification protocols.

This breach serves as a stark reminder that as the crypto industry evolves, so do the tactics of malicious actors. It's imperative for organizations to bolster their operational security frameworks to safeguard against such sophisticated threats.

Read More Here

Apple Pulls Advanced Data Protection Amid UK Government Demands

Context: Apple's Advanced Data Protection (ADP) was a game-changer when it launched in December 2022, offering end-to-end encryption for iCloud data like backups, photos, and notes. This meant that only you could access your data—Apple couldn't peek even if it wanted to. But as of February 21, 2025, UK users are seeing this feature vanish. Why? The UK government, under the Investigatory Powers Act of 2016, demanded a backdoor into encrypted data. Apple's response? A firm "no," leading to the removal of ADP for UK customers.

What's New:

• Immediate Changes: New UK users can't enable ADP; existing users will soon need to disable it.

• Data Still Protected: Services like iCloud Keychain, Health data, iMessage, and FaceTime remain end-to-end encrypted.

• Global Impact: ADP remains available outside the UK, highlighting differing global stances on data privacy.

This move underscores the ongoing tug-of-war between user privacy and governmental surveillance. While Apple stands its ground on not creating backdoors, UK users are left with reduced control over their data security.

Read More Here

Rust 2024 Survey: More Workplace Adoption, But Is Complexity Creeping In?

The Rust Survey Team has unveiled the 2024 State of Rust Survey results, offering a snapshot of the language's current landscape. Conducted from December 5 to December 23, 2024, the survey gathered insights from 7,310 completed responses. While participation dipped compared to 2023—attributed to a shorter survey window—the data remains invaluable.

Key Findings:

• Global Reach: Rust's community spans the globe, with top representation from the United States (22%), Germany (14%), and the United Kingdom (6%).

• Workplace Adoption: Over half of respondents now use Rust regularly at work, with 34% employing it for the majority of their coding—a 5% increase from 2023.

• Community Concerns: A growing number of users (43%) express concerns about Rust's increasing complexity, up 5% from the previous year.

These insights highlight Rust's expanding footprint in the tech world, both in personal projects and professional environments. However, the community remains vigilant about potential challenges, ensuring Rust evolves without unnecessary complexity.

Read More Here

🔥 More Notes

• China Accelerates AI Infrastructure Development: China is fast-tracking its artificial intelligence infrastructure by collaborating with private tech firms to enhance state-owned data centers. This initiative aims to optimize AI chip clusters and solidify China's position in the global AI race.

• Protests at OpenAI Headquarters: Three individuals were arrested during a protest outside OpenAI's San Francisco headquarters. Organized by the group Stop AI, the demonstration voiced concerns over the potential societal threats posed by advanced AI technologies.

• Microsoft's Quantum Computing Claims Questioned: Microsoft's recent announcement of a quantum computing breakthrough involving Majorana particles has been met with skepticism. Physicists argue that the evidence is inconclusive, highlighting the challenges in validating such claims within the scientific community.

📹 Youtube Spotlight

Was this forwarded to you? Sign Up Here