
19 - Mechanistic Interpretability with Neel Nanda
AXRP - the AI X-risk Research Podcast
Is That the First Principal Component?
The model is always moving to the left along the first principal component it's always moving in one direction over training. skip trigrams which are a thing one that models can do that we didn't get around to you talk about in mathematical framework were things that like improving context learning right things? So i think that's proving me wrong on that guess yes if anything you'd see like slight improvement in the same direction because skip trigrams are things that like improve context learningright things.
00:00
Transcript
Play full episode
Remember Everything You Learn from Podcasts
Save insights instantly, chat with episodes, and build lasting knowledge - all powered by AI.