I’m excited to share our new paper: “Metacognition of ChatGPT in confidence judgements”

Yoshizawa S, Onzo A, Nozawa S, Takano T, Ishikawa T and Mogi K (2026) Metacognition of ChatGPT in confidence judgements. Front. Artif. Intell. 9:1694192.
doi.org/10.3389/frai.2…

The Even Bitter Lesson?

数学界完成了一项真正不可思议的工程：Axiom 把 7,500 页的现代代数几何奠基巨著《The Stacks Project》完整用 Lean 形式化了，整整 1,600 万行代码。

现代代数几何最折磨人的地方在于，一个定理的证明往往引用了二三十年前另一篇论文里的引理，而那篇论文的某些前置条件在后来的范畴论体系下定义是有微小漂移的。人类数学家靠直觉和同行评审往往会忽略这些隐蔽断层。

1,600 万行 Lean 代码意味着，从方案论（Scheme）、层论（Sheaf）到概形堆栈（Stack），每一个引理的类型签名和构造过程都被编译器逐行机检通过。

项目地址：axiommath.ai/research/the-s…

Since August, a team within METR has worked with internal Anthropic information to assess whether AI itself is accelerating AI development.

We are now publishing a report summarizing the results of this risk assessment.

[プレスリリース] 自己回帰生成モデルの熱力学を構築――大規模言語モデルに内在する「世界モデル」の定量的理解に向けて―――工学系研究科・工学部
u-tokyo.ac.jp/focus/ja/press…

1. Benchmarks provide one piece of evidence for acceleration. Since Oct 2025, AECI scores for Anthropic’s models have grown at 1.9x [95%: 1.5–2.8] the earlier rate. Analysis choices affect this estimate, and it remains unclear how benchmark scores map to real-world capabilities.

We cut through the mystery around interpretability in training.

A single, mathematically rigorous framework transforms model internals into training feedback and works across domains, problems, and modalities.

We call this *mechanistic shaping*, and we’ve used it to mitigate shortcuts in chest x-rays, and reduce sycophancy and harmfulness. We believe this is a big step forward for technical alignment.

But it carries a known risk: training against a model’s internals can make it harder to monitor. We explore both the technique and this concern below. 🧵

Most think vibecoding at scale is slop. Here’s a counterexample.

Today we release a 16M-line formalization of the entire Stacks Project, critical infra underlying modern math, with optimized code quality and its highly interconnected structure preserved.

A mathematical city 🌆

Our book on Generative AI and Stochastic Thermodynamics is out this month. @wellingmax, @sirui_lu97 and I have been posting about some interesting topics it covers.

Today we have my favourite animations yet, and some groundwork for the coming days: five related Markov kernels.

looped reasoning is completely overrated

Energy-Based Reasoning (EBR) runs circles around looped reasoning

paper dropping this week!

THERES NO WAY LMAO.

Preprint time:
“Beliefs in feedback: Factored representations of interacting agents in transformers”
arxiv.org/pdf/2611.02047

Transformers predicting interactive agents build separate internal representations for tracking each of them ✨

Humans are ~1,000,000x more sample efficient than LLMs, and there are no good explanations.

We’ve found that a big part of the answer is that LLMs are *massively* underparameterized by many orders of magnitude. The brain has 100T synapses learning from ~200M tokens of language. 1T-parameter model on 20T tokens. That’s 7 OOMs apart in parameters per token.

After a lot of painstaking work, we present new data-constrained scaling laws and learning algorithms to scale to (extrapolated) quadrillion-parameter transformers.

w/ @akshayvegesna, @EthanBaron78553, @asatheesh6561
Full paper led by @EthanBaron78553, with @ShmuelBerman, @andrewgwils

Large-scale RL has a problem: weight updates from trainers to samplers are slow and expensive

We’re sharing weight teleportation: faster weight updates than existing intra-cluster techniques with internet speeds, enabling samplers anywhere (maybe even space!)

Weight icient (<0.23s sampler pause time, ~3s non-blocking compression + prepare), the bottleneck ends up being the GPU hardware’s decompression ASIC

We unlock these gains by using GPUs for massively parallel compression of model weight changes, engineering a heterogeneous decompression and patching pipeline, along with other optimizations and overlaps. These techniques are co-designed with hardware-native features in NVIDIA’s Blackwell generation of chips

We’ve open-sourced weight teleportation, so everyone can use it today

"Assessing AI Consciousness" is now available as a book: hutter1.net/ai/bayesconsc.… It develops a Bayesian framework separating theoretical credence from empirical evidence, providing a precise, scalable method for calibrated assessments (with online tool and code): ai-cognition.org/cacophony-tool/ Happy Halloween.

A really nice review bridging control theory, inference and thermodynamics in relation to learning written by excellent scientists from Princeton University . Reading it for a week now and really enjoying it arxiv.org/abs/2609.15897

My mother was a philosophy professor in Oxford ... and at age 8 I was fond of declaring "If there's one thing I'll never be when I'm grown up it's a philosopher!"
Well, decades later, my science led me to figure out some rather interesting things in philosophy ... and now ilosophical essays just published.
amazon.com/Ultimately-The…

「写真を撮られると魂を抜かれる」という迷信が、AI時代に少し現実になるのかもしれない。他人の自我をコピーし、手元に閉じ込める。そんな現代版の藁人形について、noteを書きました。

ペルソナ・キャッチャー：AI時代の藁人形｜Ryota Kanai @kanair_jp note.com/kanair/n/n8bcc…

This is Genie 4, our most realistic and interactive world model yet. 🌐

Explore dynamic environments in real time to see how the scene responds, sounds, and how the people in it move, react, and speak.

maybe the most important graph in the world

What if Anthropic just...stopped? Unilaterally? Right now?

I talked to @DKokotajlo @_NathanCalvin @geoffreyirving about why it's time to stop framing AI as a prisoner's dilemma and just do the right thing.

I hope a lot of people inside and outside Anthropic read this

🧵 1/n