The Rise and Fall of Bayesian Statistics johndcook.com/blog/2025/08/0…

I am pleased to announce our new paper, which provides an extremely sample-efficient way to create an agent that can perform well in multi-agent, partially-observed, symbolic environments. The key idea is to use LLM-powered code synthesis to learn a code world model (in the form

Replace Pi with Tau/2 in all your code. It's a more intuitive standard when it comes to angles and radians. Tau/2 = 1/2 turn (180 degrees) Tau/3 = 1/3 turn (120 degrees)

Introducing RND1, the most powerful base diffusion language model (DLM) to date. RND1 (Radical Numerics Diffusion) is an experimental DLM with 30B params (3B active) with a sparse MoE architecture. We are making it open source, releasing weights, training details, and code to

“What are some real world use cases for our robot? Maybe it could clean the floor.” Meanwhile:

yeah

It makes me a little sad that LLMs took away the “ask the hivemind a technical question” genre of posts from social media. Social learning was one of the bright spots on the positive side of the social media impact ledger and now learning is that much more atomized.

I’ve had some normie interactions that have me concerned too. The shift is often more subtle than the “psychosis” term implies. The difficult truth though is that human history is a continuous chain of mass psychoses and yet we have to find our way through life nonetheless.

"There are decades where nothing happens; and there are weeks where decades happen." All the best to those who face consequential weeks.

🚀Excited to launch FlashInfer Bench. We believe AI has the potential to help build LLM systems . To accelerate the path, we need an open schema for critical workloads and an AI-driven virtuous circle. First-class integration with FlashInfer, SGLang and vLLM support👉

“AI progress isn’t only about advancing intelligence—it’s about unlocking it through interfaces that understand context, adapt to your intent, and work seamlessly.”

Few people seem to know you can do the Quake 3 invsqrt hack on other functions. But it's pretty doable, e.g. x/(x+1) in raw bits.

These guys put a multi-trillion dollar open source ding in the universe. Beyond its economic impact though, the early days of PyTorch seeded a community and ethos that changed the course of many lives (including mine).