Harvard's RapidPeer Finds Claude Opus Beats Paid Reviewers In 87% Of Cases
The Common Question
Can AI write peer reviews better than human experts?
The Lesson in This Case
Harvard's rapidPeer program found that human journal editors preferred AI-generated reviews over paid human ones in 87 of every 100 head-to-head comparisons, with Claude Opus scoring higher on all seven evaluation criteria. The human reviewers were an unusually strong baseline, a pool that major journal editors had previously praised as among the best they had seen. Senior leaders funding AI pilots in knowledge-work settings should treat the finding as directional rather than definitive: the study has not yet been peer reviewed, the researchers were evaluating their own service, and the AI judge used to score outputs may favor AI-generated text.
Originally reported by Inside Higher Ed. Read the full story here.