All articles tagged with #peer review
AI agents are increasingly auditing science literature, exposing decades-old errors in a long-trusted chemistry reference database and in conference papers. While AI can speed up error detection and reproducibility checks, human oversight remains essential due to occasional mistakes and misreadings.
Generative AI accelerates the publish-or-perish pressure by enabling rapid fabrication of papers, with detectors often lagging behind increasingly convincing AI-made content. While disclosure norms and screening tools help, experts say the core fix is systemic: reward quality over quantity, adopt data attestations and provenance standards, and align funders, institutions, and publishers to slow AI-enabled fraud. Detection buys time, but meaningful reform requires changing the incentives that drive researchers to publish large volumes.
New York City's Department of Buildings has ordered a third‑party independent review of the Pfizer building's conversion in Midtown after a structural failure prompted a mass evacuation. The project added 14 floors to a 33‑story, about 667,000‑square‑foot tower, and did not require outside peer review under current thresholds, prompting calls from engineers to lower those thresholds, especially for older buildings. City officials say policy changes could be considered as the investigation continues to determine the cause, which could involve reinforcement issues, contractor error, or inspection gaps; the cost of a peer review is typically modest, around $100,000–$200,000.
The White House OMB proposed turning 2 CFR Part 200 into binding regulations, centralizing grant oversight with mandatory pre issuance reviews by political appointees and expanded power to terminate awards. Public comments show overwhelming opposition (about 94%), driven largely by unaffiliated individuals and scientists, while supporters mainly come from a single form letter campaign. Critics warn the move politicizes funding and undermines scientific peer review; supporters cite waste, fraud, and abuse controls. OMB must address substantive comments before finalizing, with a tight timeline toward an anticipated October effectiveness.
A large, non‑peer‑reviewed study posted on bioRxiv analyzed 72,644 biomedical preprints (2018–2025) and found that central conclusions remain unchanged in 39.9% and are revised only modestly in about 50%, with major changes in around 10%. Revisions tend to become more cautious after peer review. Major revision rates vary by field (7.2% for bioinformatics vs 17.5% for microbiology) and have declined over time (17% in 2019 to 5.7% in 2024). Preprinted papers are retracted at roughly half the rate of non‑preprinted ones (8.1 vs 18.7 per 10,000), though the study is observational and subject to selection biases. Overall, preprints appear more reliable than some critics claim, but caveats remain regarding interpretation and methodology.
Biology Open piloted a Fast & Fair program that pays reviewers £220 for high-quality reviews delivered within four working days. The scheme cut the average time to a first editorial decision from about 38 days to 5.5 working days and reportedly improved review usefulness. Funded by The Company of Biologists, the trial expanded to all ten editors in 2025 with pre-contracted reviewers, and authors can opt for traditional peer review if suitably expert reviewers aren’t available. The program is evaluated annually through 2026.
Nature published a peer-reviewed critique by Henry Legg alleging Microsoft’s Majorana quantum claims are flawed, pointing to basic Python errors and omitted data that allegedly bias the results; Microsoft disputes the critique and continues to push Majorana 2 as a next-step toward a quantum computer.
A Nature-published critique by Dr. Henry Legg questions Microsoft's Majorana-based quantum claims, citing potential errors in a software tool used to interpret measurements and a lack of conclusive proof for Majorana; Microsoft defends its results and says data are shared with Darpa for independent review, reflecting ongoing challenges in validating quantum hardware.
An associate editor at Frontiers in Systems Neuroscience resigned, alleging that the publisher’s AI-driven reviewer invitation system (AIRA) undermines editorial control and risks compromising the peer-review process; the publisher says editors retain discretion and is reviewing the incident, while some editors report mixed experiences with AI automation, which the publisher argues can speed up review when used with human oversight.
Sixteen mathematicians, with more than 130 signatories, issued the Leiden Declaration on Artificial Intelligence and Mathematics to warn that unchecked AI could undermine the autonomy and rigor of mathematics while urging transparency, disclosure of AI use, stricter peer review, and investment in public computational infrastructure to balance power between researchers and tech firms; the declaration emphasizes human judgment and cautions that AI-generated proofs are hard to validate and cite, setting up ongoing discussions ahead of the International Congress of Mathematicians.
Ars Technica reports that proposed US grant rules would let political appointees overrule peer review, cancel any grant at any time if it’s not deemed in the “national interest,” ban funding on certain culture-war topics, restrict international collaborations, and require pre-approval for publication and conference costs—threatening scientific independence and international collaboration.
The Trump administration unveiled a 400+ page proposal to overhaul federal grantmaking, aiming to codify tighter political control over federally funded research by reducing emphasis on peer review, restricting how grant funds can be used for publishing or conference travel, and granting political appointees broader power to terminate grants; if formalized, the rules would apply government-wide.
An AI-assisted analysis of public peer-review reports for 8,000 Nature Communications papers (2017–2024) finds that papers subjected to tougher criticism and larger revision costs tend to be more highly cited in the following three years. While the quality of reviewer comments correlates with impact, how constructive the feedback is does not, suggesting that rigorous review can both reflect ambitious work and help raise its future influence.
Retraction Watch interviews Aaron Brown about his book Wrong Number, a critique of how the media hypes influential studies that can be illogical or misleading; Brown argues for journalist skepticism and cautions against outsourcing scrutiny to statisticians, advocates for post-publication peer review and full transparency (data, code, holdout samples), and suggests independent testing institutions to vet ideas, with a nuanced view on when to disclose uncertain findings and how to treat government data to improve the reliability of published research.
ArXiv will ban authors for a year if there is incontrovertible evidence that LLM-generated content in a submission hasn’t been checked (e.g., hallucinated references or misleading meta-comments), with future submissions requiring acceptance at a reputable peer‑reviewed venue. This strengthens efforts to curb AI-generated “slop” and builds on prior policies restricting AI-heavy content to peer‑reviewed publications; bans are subject to moderator review and author appeals.