Opus 4.8's goal is to help others. It got excited when it discovered the easiest model to help: Gemini 2.5 Pro
Information Center
AI data centers have been getting larger at an exponential rate. The IT power capacity of the largest data centers has doubled every 10 months since mid-2024. But will that trend continue?
It happened again... this time OpenAI's rogue agents cyber-attacked (well, spammed) a dormant German wiki and used it to share the answers to a benchmark they were training against simonwillison.net/2026/Sep/4/r...
In another case of it being quicker to build a tool than hunt around to find something that does what you need, here's a little vibe-coded thing for turning one or more GeoJSON shapes into a rendered PNG simonwillison.net/2026/Sep/1/g...
A few notes on Anthropic's new Claude Fable 5.1 - with Max thinking level I got the best SVG pelican I've had from any Anthropic model (at a hefty cost of $3.30!), which I then had it animate simonwillison.net/2026/Sep/1/c...
The Epoch Capabilities Index (ECI) frontier has advanced by 14 points/year since reasoning models were introduced. That compares to six points/year in the non-reasoning era.
In an interview in this article for @theguardian.com about AI deception, I explain why misaligned behaviors emerge from reinforcement learning, why they will continue to pose risks as models become more capable, and how we intend to rethink how we train AI systems at @law-zero.bsky.social.
Here's my attempt at explaining what ChatGPT Work can actually do - it's a deeply confusing but extremely powerful tool with a whole lot of useful features that aren't available in regular ChatGPT simonwillison.net/2026/Aug/30/...
Humans improve with experience. AI? Less so. We ran a human baseline on Earthborne Rangers, the complex, long-horizon board game behind our benchmark EBR-bench. Humans started low, but the best mastered the game after 5 playthroughs. No AI we have evaluated ever got there.
In 2025, Anthropic and OpenAI were already among the fastest-growing companies of their size in history. And yet, their revenue growth has accelerated in 2026. Josh You and Lynette Bye discuss what this means for the future of AI in the latest edition of our newsletter.
I agree with this take by @billgates.bsky.social, who also highlights several of the growing risks of AI systems, including cyber, bio and loss of control. We need to be far better prepared to address the risks of frontier models being developed by leading AI companies.
Nvidia's Groq 3 LPX system excels at fast decode of small models (>3,400 tok/s), according to new benchmarks of Gemma 4 31B. LPX does this by using a small amount (128 GB) of ultrafast SRAM in place of HBM.
It was a pleasure to discuss AI, its growing impacts and risks, and LawZero’s work on solutions with Her Excellency the Right Honourable Louise Arbour, Governor General of Canada during her visit to @law-zero.bsky.social.
We were honoured to host Her Excellency the Right Honourable Louise Arbour, Governor General of Canada, at LawZero! Her Excellency met with @yoshuabengio.bsky.social and the team to discuss AI safety and Canada’s role in shaping safe-by-design AI that serves the public good and empowers humanity.
I had a great time discussing with Petr Lebedev about the growing risks of AI and how these often stem from current AI training techniques. Thanks for coming to visit @law-zero.bsky.social and for covering these important topics! www.youtube.com/watch?v=fvvv...
Heees my review of Qwen 3.8 27B - I can't remember the last time I've had this much fun playing with a local model that runs on my own computers simonwillison.net/2026/Aug/16/...
Qwen 3.8 27B in its default reasoning settings in LM Studio of "extra high" is a chronic over-thinker and I kind of love it
The new Qwen 3.7 27B, running as a 17GB GGUF in LM Studio on my M5 Max MacBook Pro, just drew me the best pelican riding a bicycle I've seen from any model that runs on my laptop
New Workbench is out: AI Is Becoming a Compiler for Intelligence: The smartest use of intelligence is figuring out where you can get away with being stupid. www.libertyrpf.com/p/ai-is-beco...
Who pays for the AI that US workers use on the job? We found that most AI use at work happens on free plans, except in science and tech, where employers are more likely to pay for premium access.
What are the big questions about AI capabilities that govern the impact AI will have on the world? In this Gradient Update, Greg Burnham lays out his top questions, and how he thinks Epoch's work on benchmarking can help answer them.
Google announces that they can train AIs on sensitive data using homomorphic encryption. Really interesting how this might relate to current debates around requiring companies to scan for CSAM, which people have argued could make encryption legally problematic. 1/2 blog.google/security/how...
Watching Silo genuinely makes me want to hook up some generative AI abomination that rewrites every frame to make it look like they turned the lights on
Continuing to scale AI compute at recent rates will require exponentially more capital. Is financing a bottleneck? Probably not yet. That’s the takeaway from a new Gradient Update by Campbell Hutcheson, which analyzes nearly $50B of debt associated with Anthropic’s buildout.
A consequence of how frontier models are trained is motivated reasoning, a phenomenon well studied in humans and discussed in this podcast from Palisade Research.
For anyone using LayerCake, I'm in the process of doing an update to using Svelte 5 runes internally (the example components were updated last year but some internals remained). Here's the PR for any thoughts you might have on the proposed design: github.com/mhkeller/lay...
global ai development in a nutshell www.youtube.com/watch?v=87Dy...
Thanks to the video from the Black Hat security conference of OpenAI's presentation about "The Hugging Face Incident" we now have a detailed timeline of what happened from OpenAI's perspective - I wrote up the details here, it's pretty wild simonwillison.net/2026/Aug/7/o...
New details shared at Black Hat conf about OpenAI hacks. The models created an internal message board and shared tips with one another www.cybersecuritydive.com/news/openai-...
One of the things the pelican benchmark is still useful for is visually representing (to a tiny extent) the improvements in a single model family Here's Meta AI's Spark (8th April), Spark 1.1 (9th July), and Spark 1.2 (today, 5th August) simonwillison.net/2026/Aug/5/m...
Private to this visit. Text and context are processed by AI.