Deedy Das: ‘Already Clean’ is ‘one of the most important results in AI, period’

This is one of the most important results in AI, period.

TLDR: neural networks are not tangled knots (polysemanticity is wrong). You can study their own neurons directly, without first “unfolding” them with a second model. 90% of neurons in Qwen-32B have one job.

Nick Cammarata started OpenAI’s interpretability team and is at Goodfire now. He’s been studying how LLMs work for a decade.

The stakes for understanding and aligning model behavior are the highest it’s ever been. Models have achieved superhuman results in math, can break production software systems, and changes the nature of white collar work. This work may finally allow us to intentionally design LLMs and shape how they behave.

Its a long, phenomenal read.

Quoting @nickcammarata: interpretability might be the most important scientific problem in the world right now, as I don’t think we can solve alignment if we don’t know what we’re building. I think I have good news about it. I co-authored early work in mechanistic interpretability at OpenAI and came away pessimistic, because neural networks appeared to fold into arbitrary tangled knots. years later, I started openai’s first large-scale automated interpretability project, finding about 2% of neurons legible with the primitive tools we had

with modern agentic tools and a much closer look, the picture looks totally different. the vast majority in models up to trillions of parameters are legible, and their circuits beautiful. many famous examples of tangled neurons turned out to be misunderstood. The roadblocks I expected mostly aren’t there. I now think ambitious interpretability will be solved by the time we need it most, and possibly quite soon, as automated scientists keep improving. it may be a bitter lesson for interpretability: the simple thing, at absurd scale, works

https://already-clean.goodfire.com/