Gary Marcus · Independent · NYU (Emeritus) · 2024-11-09
Marcus has spent a decade as the field's most-quoted irritant, and this is the post where his core bet appeared to come due. He argues the pure-scaling thesis hit a wall in late 2024, points to the venture-side admissions, and reiterates the structural claim he has been making since 2022: more compute alone will not buy you reasoning or grounded facts.
Whether you ultimately agree is beside the point. The piece is on the list because the shape of his argument — what counts as evidence, what would falsify it, what alternative architectures he wants funded — is the cleanest version of the skeptic case currently in print. Bring it to any "scaling is all you need" conversation.
> We're increasing GPUs at the same rate, we're not getting the intelligent improvements at all.