I debated whether to share my story on here, but I guess I will. I think there's an idea out there that millions of violent criminals are pouring across the border, carrying machetes and drugs, looking to harm Americans. Certainly, while some people fall into that category, the

Made of meat

🚨Excited to announce that the Sovereign AI Unit is investing £8m in OpenBind: building the world's largest protein-ligand dataset. Backed by a top UK consortium, inc. Isomorphic Labs and Nobel Laureate David Baker, this will help the UK become the home for AI drug design 1/6🧵

AlphaFold was only made possible by a data resource that is a bit of an unsung hero. That data was the Protein Data Bank (PDB). 50 years to curate and billions of £ later, and you had enough solid data on protein structures to do something of Nobel prize winning significance.

The sad thing is that Apple shipped before Mozilla. Things not looking good for Firefox...

Super proud to release our most fun AI experiment to date: AISheets 🗒️ Thousands of AI models meet spreadsheets. Build, analyze, and automate your data using open-source LLMs in one slick, fast and simple app. Surprisingly powerful! 🚀 Try it: hf.co/aisheets

I really love my job in AI these days and it can still be so hard to be happy about it w the pace of progress/effort required. Heart goes out to all my friends putting in crazy hours at a place they kind of dread just to keep up. Worth trying to find the better fit.

175 chars now! vec3 a=r.xxy,d=a-2.*round(FC.zxy),p=cos(t+a);for(p.x=t/.1;o.w++<9e2&&snoise3D(a=ceil(p+=min(a=mod(-p,sign(d))/d+1e-6,min(a.y,a.z)).x*d)/28.)+1.>length(a.yz););o.rgb=fwidth(p);

Your weekly reminder not to build LLM systems that combine access to private data with exposure to untrusted tokens and exfiltration vectors (the "lethal trifecta"). This time it was Microsoft 365 Copilot (now patched, they closed the exfiltration holes) simonwillison.net/2025/Jun/11/ec…

"Design Patterns for Securing LLM Agents against Prompt Injections" is an excellent new paper that provides six design patterns to help protect LLM tool-using systems (call them "agents" if you like) against prompt injection attacks

Congrats to Simon Willison (Simon Willison) on 23 years (!!) of blogging. Really excellent LLM blog, I sub & read everything: simonwillison.net (e.g. I sub via RSS/Atom on NetNewsWire) +If you consistently enjoy the content like I do, sponsor on GitHub: github.com/sponsors/simonw

When I was a lawyer, I went on a secondment to my law firm's Paris office; the trainees I managed there were all extremely bright, but because of French academic culture, would produce 20 page long memos full of legal jargon that clients couldn't care less about. So they had to

Six weeks after ChatGPT I argued that we were already in a Long Singularity For 20,000 centuries of human history, nothing much happened. We spent 19,960 centuries on variations of one tool. Things only accelerated two centuries ago. Surprisingly, we have (mostly) kept adjusting

Many PhDs (my past self included) fall into the trap of thinking that publishing in top-tier conferences is the ultimate goal. But publishing ≠ impact. Muon was just a blog post. It got Keller into OpenAI, he might be training GPT-5 with it now. I'm grateful he listed me as

Geoffrey Hinton needed financial security for his family, so at 65 he sold his neural net company to Google. His former student Ilya Sutskever left Google to co-found OpenAI, where he built ChatGPT. Now Hinton warns about AI risks while OpenAI leads the revolution he sparked.

🚨Breaking: New DeepSeek-r1 (0528) just tied for #1 in WebDev Arena, matching Claude Opus 4! More highlights: 💠 #6 Overall on Text Arena 💠 #2 in Coding, #4 in Hard Prompts, #5 in Math category 💠 MIT-licensed, currently the best open model on the leaderboard! Huge congrats

Wow, in 1887 C.S.Peirce coined the term "Logical Machines."

This is a clever way of showing that LLMs still do a very bad job expressing uncertainty, and that reasoning models are even worse at saying “I don’t know”

learningfromexamples.com/p/what-academi…

(3/n) Introducing Evolutionary Policy Optimization (EPO): a hybrid policy optimization method that integrates genetic algorithms. A master agent learns stably and efficiently from pooled experience across a population of agents.