Opus 4.8 reportedly outperforms expert peer reviewers on 124 Harvard social science papers

A post says Opus 4.8 beat human experts as a peer reviewer in a study of 124 Harvard social science papers covering statistical analyses, formal models and qualitative work. Gary King and Steven Worthington say AI-generated reviews outperformed highly rated human reviews by every measure they constructed. They discuss potential ways to speed publication and improve its quality, alongside concerning downsides.