Evaluating GPT-3 and GPT-4 on the Winograd Schema Challenge (d-kz.medium.com) 1 points by niccl 3y ago ↗ HN
[–] sharemywin 3y ago ↗ I'm guessing they trained it on most if not all of those datasets. [–] qualudeheart 3y ago ↗ Maybe. Unclear if that’s falsifiable.
2 comments
[ 3.3 ms ] story [ 14.1 ms ] thread