GPT-5.2 matches expert human peer reviewers in study

Arxiv CS.CL (NLP/language models)

45 scientists spent 469 hours evaluating AI and human reviews on 82 Nature-family papers. AI reviewers matched top-rated human reviewers, but best results came from AI+human collaboration.

GPT

See 1 more source

Seems GPT-5.2 reaches expert level in peer review: 45 scientists took 469 hours evaluating human & A...26 days agoEthan Mollick

Arxiv CS.CL (NLP/language models)·Seungone KimDongkeun YoonKiril GashteovskiJuyoung SukJinheon BaekPranjal AggarwalIan WuViktor ZaverkinSpase PetkoskiDaniel R. SchriderIlija DukovskiFrancesco SantiniBiljana MitreskaYong JeongKyeongha KwonYoung Min SimDragana

27 days ago