For a while, gains in AI have translated into gains in modeling the brain. We test if that will continue to be the case with recent advances in scaling. Surprisingly we get mixed results: while increased scale advances model alignment to behavior, neural alignment saturates.
[preprint] [social media]