Skip to main content

Brains Predict Hierarchies—LLMs Should Too!

·38 words·1 min

This is very interesting study showing that brain predicts hierarchical representations over multiple timescales. Authors argue that LLMs should try to predict similar representation instead of just next word.

Big hint towards rethinking naive autoregressive objective… https://x.com/AIatMeta/status/1638912735143419904

Discussion