Presenting this at #ICML with @rjantonello.bsky.social and Aditya Vaidyaโจ
Why do ๐ข๐๐๐๐ก๐ layers in LLMs and speech-audio models best predict brain responses to language?
We show a peak in the dimensionality of ๐ค activations (left) to track high ๐ง predictivity (right)
๐งต(cross-posted from X)
