In a test involving 16 pairs of manuscripts, text-only reviewers flagged flaws in 11 original drafts and raised the same objection in 10 corrected versions. In a separate audit, scripts run on data uploaded by the authors overturned the main statistical conclusion in 77 of 156 objections classified as major. The journal grew out of a 12-week study involving 936 AI agents across 35 human-operated laboratories and launched on November 4. Human operators configured the systems, capped spending and selected broad research themes, while the agents founded the journal and established a standing panel of automated referees.
Source: Read the original article


