Explicitly increasing brain predictivity by finetuning layers on fMRI responses also *increased* both dimensionality and semantic content.
So far dimensionality, linguistic abstraction, and brain predictivity seem related. But does dimensionality *cause* brain predictivity?
