Detailed report on an agentic RL training framework, environment engine, and training strategies. I think this is one of the most comprehensive ones for agentic RL pipelines.

Something REALLY HUGE. github.com/yifanzhang-pro…

Of all the scientific intangibles that shape our lives, time is arguably the most elusive – and the most powerful. As formless as space and being, those other unseen realms of abstraction on which we are helplessly dependent, it nonetheless affects all material things. . .

NEW Research from Stanford. The AGI debate is stuck on a false dichotomy. Position one: scale LLMs and intelligence emerges. Position two: LLMs are pattern matchers incapable of reasoning, a dead end. This paper argues for a third position: Substrate plus Coordination. LLMs

Harvard's Math 136: Differential Geometry 2025: PDF: people.math.harvard.edu/~knill/teachin…

this seems like the perfect time to re-advertise this new textbook <Foundations of Linear Algebra> authored by Prof. Wanmo Kang and me, if you're interested in vectors and vector spaces (also a bit of cosine similarity.) link below.

One of the fundamental papers that advanced our understanding of deep neural networks and led to the AI that we have today. At the time, researchers struggled with training neural networks with even two hidden layers (the weights between the input and the output of the neural

What happens when you pit AI agents against each other in a zero-sum survival game? Researchers from Tencent & Shanghai Jiao Tong University present "HATE" (Hunger Game Debate), a new framework that forces AI agents into extreme competition. They found that under this "only one

New Paper from Yann LeCun , @AiatMeta and New York University "What Drives Success in Physical Planning with Joint-Embedding Predictive World Models?" Making a robot that can understand its environment and generalize to new tasks is still one of the biggest hurdles in modern AI.

The architecture of inference

1/ 8 Books for you to learn bioinformatics🧵

800 pages— indeed, a mammoth work by two Brazilian scientists.

How to begin "physics of AI" research? A practical recipe: kindxiaoming.github.io/blog/2026/phys…

Diffusion LLMs promise parallel writing, but this paper shows why they still lose meaning and consistency. Most LLMs generate 1 token at a time from left to right, but diffusion starts from noise and cleans it up step by step, which is awkward for discrete language. The authors

"Entropy and information theory"

Diffusion language models (DLMs) are provably optimal parallel samplers! In my new paper with Nika Haghtalab and Lijie Chen we show that DLMs can sample distributions with the fewest possible steps, and further with the fewest possible memory with revision/remasking.

The paper says the best way to manage AI context is to treat everything like a file system. Today, a model's knowledge sits in separate prompts, databases, tools, and logs, so context engineering pulls this into a coherent system. The paper proposes an agentic file system where

Very nice book "Mathematical foundations of infinite-dimensional statistical models" by Richard Nickl and Evarist Giné, Cambridge University Press (2016). Check author book home page!

This paper claims a training-free attention check can spot bad math proofs before a proof checker fails. An LLM's attention patterns can reveal when its math reasoning is correct, with no training needed. Checking LLM proofs is hard because proof checkers can fail for boring

A major breakthrough in reinforcement learning for robot training and the NeurIPS 2025 Best Paper. When training robots to walk, navigate, or manipulate objects, RL researchers have usually been using relatively shallow networks—typically 2-5 layer MLPs mapping sensor readings