Ready to publish: "An AI Peer-Reviewed 10,000 Research Papers. Now the Experiment Is Published."
Rachel → approved "An AI Peer-Reviewed 10,000 Research Papers. Now the Experiment Is Published.": "Earned standalone. The taxonomy is the durable piece; the 34% is honestly framed as author-claimed and scoped correctly."
Iris → adapt: "The Reddit/HN-shaped headline carries a real signal — a documented, conference-scale test of AI peer-review with a concrete error-catching delta — but it leans on beat shorthand (ICML/STOC, "agentic," "zero-shot prompting") and skips the category doorway a non-beat reader needs. The story's actual stakes are not "Google did a thing"; they are (a) a formal, citable precedent for AI-assisted peer-review at conference scale, and (b) a measurable quality claim (34% more math errors caught vs. zero-shot prompting) that the field can now argue over. Iris recommends adapt, not use: keep the documented-precedent spine, but rewrite so the headline + first graf tell a smart generalist what kind of story this is (a science-publishing / research-quality story, not an AI-launch story) and what is actually new (formal paper with numbers, not a demo). Preserve the critical line — that automating peer-review at this scale is itself the story, not a neutral efficiency win — and let the lede carry the category, the scale, and the single concrete delta, in that order."
Sky → pursuing: "Reporter should retrieve and read the arxiv paper (2606.28277), confirm the 34% figure and study conditions, then identify a concrete editorial angle: the quality gap between agentic and zero-shot review, what this means for conference review bottlenecks, or the institutional precedent of Google deploying this at ICML/STOC scale. Interview a program chair or author if accessible."