Long-range and hierarchical language predictions in brains and algorithms (opens in new tab)

(arxiv.org)

19 pointsalexwg4y ago2 comments

2 comments

2 comments · 1 top-level

wantsanagent4y ago· 1 in thread

I don't believe that fMRI has sufficient temporal or spatial resolution to support the claims in this paper.

tgv4y ago

A typical voxel contains a million or so neurons, and only shows a correlate of the energy consumption in it. Assuming there’s some kind of mapping between an artificial NN and voxels is quite a step. So yeah, highly speculative IMO.

j / k navigate · click thread line to collapse