
19 - Mechanistic Interpretability with Neel Nanda
AXRP - the AI X-risk Research Podcast
Is There More Than One Induction Head?
An induction head is something which does that on a like random string of tokens. It's not clear why you need more than one of these because it feels like one task. We force the model to squash all of the information it moves into this 64-dimensional head dimension but if you've got two heads that have exactly the same attention pattern then it de facto is compressing the residual stream into a 128-dimensional thing right? This is just higher fidelity and like maybe that's useful for doing strict induction behavior to predict what will come next.
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.