Opus 4.8 out-reviews human experts on 124 Harvard papers

Ethan Mollick (@emollick) highlighted a study in which Opus 4.8 outperformed human experts as a peer reviewer "by every measure." According to his post, the study covered 124 Harvard social science papers. It drew about 1,000 likes.

Peer review is how academic work gets checked before publication, and it is famously slow and uneven. Using AI to review papers has been controversial, with concerns about accuracy and missing context.

Why it mattersIf results like this hold up across fields, AI could become a standard part of the review process, perhaps as a first pass before human reviewers. How "every measure" was defined matters, so it is worth reading the study itself before drawing broad conclusions.