Nick claude tldr

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

oh this is quite well done! some small parts are off but surprisingly good

@nickcammarata This has such clarity. Thank you Michael

@nickcammarata How did you generate this? Just asked Claude to tldr the piece?

I had to download all the files from the site first it was too large for standard Claude to fetch on its own and the pdf of the content was too large. then i asked it to create a short form video that explains what’s going on. It first gave a 60 second video that assumed i knew the topic and then i followed up with asking for a breakdown based on questions someone might have and its ok to be a bit longer. That’s about it. I have eleven labs connected to my account and probably have some skills I’ve collected over time but i think opus 5.5 doesn’t really need that

@nickcammarata Sounds pretty straightforward, thanks. (the video is actually helpful. at least, to a newbie like me ^^)