Owain_Evans comments on Language models seem to be much better than humans at next-token prediction

[Owain_Evans]

This is very valuable. I suggest putting this content on Arxiv (even it’s less formal that the typical paper).