The TRM paper feels like a significant AI breakthrough. It destroys the pareto frontier on the ARC AGI 1 and 2 benchmarks (and Sudoku and Maze solving) with an estd < $0.01 cost per task and cost < $500 to train the 7M model on 2 H100s for 2 days. [Training and test specifics]

It begins, as predicted...

the kalpa engine is the first building block towards working with socratic thought i am building a language of reasoning that is: 1) atomic 2) socratic 3) hierarchical

Gen AI is like teenage sex: everyone talks about it, nobody really knows how to do it, everyone thinks everyone else is doing it, so everyone claims they are doing it.

the singularity is currently happening. AGI is here, it is doing my job, you guys are all idiots and are being left behind. some of us are making fortunes that you couldn't even begin to comprehend, compensated at 2m, 4m, 10m+. if you don't think it is you're just not invited

fkin lying ass bullshitter

Excited to release new repo: nanochat! (it's among the most unhinged I've written). Unlike my earlier similar repo nanoGPT which only covered pretraining, nanochat is a minimal, from scratch, full-stack training/inference pipeline of a simple ChatGPT clone in a single,

org ai hygeine is going to become a real thing there will be roles dedicated to making sure the org uses codex/cursor as efficiently as possible; laser focussed on preventing engineers writing ANY code. and build endlessly. we will see the rise of the chief automation officers.

imo the best analogy for present day llms is pre-training is synthetic evolution, post-training is synthetic schooling/education and both layers are lossy hence the hallucinations/hollow reasoning

storing skills is an extremely important part of intelligence. super excited to try this out!

google is so over powered man wth

We verified the TRM results on the semi private holdout set and they're legit Awesome work and contribution to the open source community by Alexia Jolicoeur-Martineau My notes: * This model is tiny! 7M params, but the rub is that it is relatively expensive to run because pre-training and

Today we're introducing Skills in claude dot ai, Claude Code, and the API. Skills let you package specialized knowledge into reusable capabilities that Claude loads on demand as agents tackle more complex tasks. Here's how they work and why they matter for the future of agents:

can't stress enough how much i love this kalpa will be also be implementing similarly storable skills. and of course, i will visualize the whole thing.

ai engineering is in such an exciting place one gets a sense that even the fundamentals like memory and personalities haven't been figured out. you can build something over a week that might be used for decades later. fertile grounds. i love it.

Claude Skills are awesome, maybe a bigger deal than MCP simonwillison.net/2025/Oct/16/cl…

prime intellect advanced reasoning solid thinking adequate cognition basic understanding general intuition minimal perception dim recognition faint notions

now that the TRM paper is blowing up, I would like to remind you that I was super early on recursive reasoning. It was even the exact name of the project I had released!

why claude skills is a huge deal