
19 - Mechanistic Interpretability with Neel Nanda
AXRP - the AI X-risk Research Podcast
The Second Mesh Thing to Bear in Mind When Using Transformers
Transformers are fundamentally sequence modeling networks their input is a sequence of tokens which you can basically think of as words or subwords and at each step they're doing the same processing in parallel for every element of the sequence. A decent chunk of transformers computation comes down to rooting information between different positions figuring out what information to root. The main thing that so obviously any smooth information between positions because we don't just want it to be a function of only token of only the current token.
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.