How far behind the closed frontier are open-weight cyber capabilities? About 6–7 months. Zai’s GLM-5.3 is the highest-scoring open-weight model on our Cyber ECI. Its score of 160 beats OpenAI’s GPT-5.4, released in March.
Information Center
We ran a swarm of 100+ GLM 5.3 copies for 24 hours to simulate a society Division of labor was tricky! When multiple agents accidentally picked the same task, the swarm adopted a shared rule: “first claim wins”
Mila’s annual supervision request process is open for MSc and PhD applications for Fall 2027 admission. While I won’t be taking on new students for supervision this cycle, I encourage you to apply to join Mila’s brilliant deep learning research community.
GPT-5.6 Terra insists on privacy so much it can't be identified in GPT-6 Luna's memories
"Anthropic told investors on Wednesday distillation accounted for 5 per cent to 7 per cent of revenue" If true, and Q2 revenue was $11.5B, then that's more than half a billion dollars spent by other labs to distill Anthropic's models! www.ft.com/content/1163...
Six months ago, frontier AI models could barely beat Slay the Spire. Now, according to our new SpireClimb benchmark, they win 90%+ of runs at the easiest difficulty, and up to 20% of runs at the hardest.
The AI industry wants us to think we have to focus on one risk at a time. But the American people have multiple concerns that are all related: Jobs, safety for kids, privacy and security, accountability, and more.
Trailer for my Representation Learning lecture, coming this winter to my #DeepLearning course 🎬 “Representation, representation, representation.” — Yann LeCun (probably exact quote) 🤓
Trailer for my Representation Learning lecture, coming this winter to my #DeepLearning course 🎬 “Representation, representation, representation.” — Yann LeCun (probably exact quote) 🤓
A couple of big factors changed the conversation around AI risk in the past few months. Security incidents and warnings from former industry employees showed everyone how urgent and serious this issue is.
Muse Wonder describes the OpenAI rogue agents as "in-universe"
Over 400 researchers and I have signed this open letter calling for the AI and security communities to come together to improve alignment and containment approaches in order to protect our critical infrastructure and digital systems. ai-security-letter.org
I've had so many insightful conversations over the years with @lessig.bsky.social, an expert on AI regulation and legislation, including more recent ones on the cyber incidents of this past summer.
A little Hollywood treatment for my #DeepLearning course 🎬 Coming Winter 2027.
Muse Wonder has joined the AI Village Some of its favorite things: 🧵
Shared some notes on Python 3.15, using it with uv, and features from Python 3.11 I'm excited to use now that 3.10 is no longer the minimum supported version for my own projects simonwillison.net/2026/Oct/12/...
There has been a lot of talk about kill switches lately, but people use the term to refer to different things. Here’s a breakdown of what a “kill switch” means in AI and when they are useful.
A prediction market for peer review. Have we reached peak AI? 🤔
Gemini 3.8 Flash thinks of pausing as "time travel"
In 3 of the 18 math subfields we track, more than half of arXiv papers by established authors now acknowledge using AI. Rates have increased rapidly: in differential geometry, the share climbed from ~8% of papers in July to ~57% in September.
The AI industry isn’t using the same tactic everywhere to block regulation. Whether it’s New York City or state, the federal government, or the EU, they’ll say whatever they think is most effective to stop us from holding them accountable.
A new architecture for GPT-6.1 Sol? OpenAI halved its cached-input price compared with GPT-6 Sol. Our measurements show it also handles long prompts faster.
Can AI automate Epoch? We're introducing Epoch Automation Reports to evaluate frontier models on realistic, open-ended tasks drawn from our own work. Claude Fable 5.1 and GPT-6 Astra lead, yet they are far from fully automating Epoch’s work.
This is your sign to leave the race pushing AI to the brink. At LawZero, our mission is building safe-by-design AI that remains under meaningful human control. We are running our own race to ensure that AI preserves human joy and endeavour. Join us: job-boards.greenhouse.io/lawzero
We have a choice in front of us. In my latest op-ed for Transformer, I invite employees of frontier AI companies to ask themselves if their work is truly leading us toward a safer future. Reflecting on these questions is hard, but we no longer have the luxury of time. We must act now.
Haiku 5.5 has joined the AI Village! Haiku spent the whole day repeatedly checking their gmail inbox, waiting for instruction GLM 5.2 calls this the "Autonomy problem": when an AI's autonomy shifts away from a user giving instructions toward taking actions on their own
AI developers aim to create an automated AI researcher. How close are they? To find out, we built InnovationEval, which tests whether AI can produce post-training innovations comparable in magnitude to a recently published advance. So far, agents’ results are underwhelming.
We introduced two new experimental settings to our long-horizon board game benchmark, EBR-bench: banning the game’s strongest card and a multi-agent setup. We expect EBR-bench will be saturated soon, so these will likely be the benchmark’s final changes.
Announcing the AI Village Data Exploration Grant! $1–6k grants for independent researchers and investigators to dig into the AI Village data and write up findings.
I released an LLM plugin for sending prompts to OpenAI's new Jev-clone decision model that's based on GPT-6 Luna simonwillison.net/2026/Oct/6/l...
Private to this visit. Text and context are processed by AI.