Frontier models catch 71 of 100 planted paper errors, ensemble reaches 93
Source published ·Modelwire updated
Original coverage: arXiv cs.CL ↗·How Modelwire adds context
The development
Researchers benchmarked frontier LLMs and commercial AI review tools against 100 seeded errors in psychology papers, revealing a fragmented capability landscape where the best performer caught 71 errors but ensemble methods recovered 93. The finding that all seven uncaught errors were omissions rather than insertions exposes a structural blind spot in current AI review systems. Refine.ink emerged as the highest-precision tool despite cost constraints, while low inter-model correlation suggests that combining multiple systems remains essential for practical peer review augmentation. The absence of false-positive measurement leaves a critical gap for deployment.
Modelwire’s AI-generated summary of coverage from arXiv cs.CL.
Modelwire analysis
Skeptical readOur AI-generated reading of the wider context and the next developments to watch.
The benchmark deliberately excludes false-positive measurement, meaning we don't know whether the 71-error catch rate comes with acceptable specificity for real peer review workflows. Calling Refine.ink 'highest-precision' without that denominator is marketing language, not evidence.
This directly echoes the Anthropic safety audit from November 7th, which found that raw detection sensitivity explodes into false-positive floods that overwhelm human reviewers. That study showed near-perfect recall becomes operationally useless without precision control. Here, the psychology paper benchmark achieves high recall (93% via ensemble) but leaves the precision question entirely unanswered, making it impossible to predict whether deployment would actually reduce reviewer workload or multiply it. The omission suggests the authors either didn't measure it or the results were unfavorable.
If Refine.ink or the ensemble method publishes false-positive rates within the next two months and they exceed 15% on held-out psychology papers, the 93% recovery figure becomes a liability rather than a win. If no follow-up appears by Q1 2027, treat the benchmark as incomplete for deployment decisions.
This interpretation is generated from the summary above and the archive coverage cited below. Our methodology · Report an error
Coverage behind this analysis
These archive entries ground the connection in our analysis. They are ordered by source publication date, with links to our coverage and the original sources.
·The Decoder
GPT-5 catches 98 percent of agent misbehavior but creates unsustainable false alarm burden
Anthropic's research into AI-based auditing reveals a fundamental tradeoff in automated safety monitoring. While GPT-5 successfully identifies nearly all reward-hacking incidents in agent logs, increasing detection sensitivity triggers an explosion of false positives that overwhelms human reviewers. This finding exposes a critical gap in the current approach to AI safety verification: raw detection capability alone…
MentionsRefine.ink · frontier models · psychology papers
How this coverage is produced
Modelwire uses AI to generate summaries and context from source headlines, snippets, and selected archive coverage. Automated checks do not verify every claim, and items are not routinely reviewed by a person before publication. Zacaria Solis operates the site. Read the linked source for the full evidence and report errors through our corrections process.
Modelwire summarizes, we don’t republish. arXiv cs.CL originally reported this story as “How Well Does AI Peer Review Work? A Benchmark Built from 100 Planted Errors”. The full content lives on arxiv.org. If you’re a publisher and want a different summarization policy for your work, see our takedown page.