A preprint comparing human and machine reviews at iclr 2026 and nature communications found the…
a preprint comparing human and machine reviews at iclr 2026 and nature communications found the models extended critique coverage but showed a hivemind effect, agreeing with each other too much, and that ai review scores are trivially gameable through paper rewriting. biomedical publishing now exceeds 3 million manuscripts a year, which is why the answer is not to hand reviews to llms but to redesign what a review is for.
a reviewer’s job is disagreement, and models are trained to agree.
Context
A 2026 arXiv preprint, 'Review Arcade: On the Human Alignment and Gameability of LLM Reviews' (2605.28897), studies how LLM reviews align with human reviews and how easily they can be gamed. Another preprint (arXiv 2606.13044, Jun 11, 2026) studies gaming AI peer review without hidden prompts.
A position paper presented at ICLR 2026, 'Stop Automating Peer Review Without Rigorous Evaluation', argues that today's AI systems should not be used to produce paper reviews. Springer Nature (May 21, 2026) says more than three million scholarly articles are added to the scientific record each year.
The gameability finding and the argument against handing reviews to LLMs match the preprints and the ICLR position paper. The specific preprint comparing ICLR 2026 and Nature Communications reviews, the hivemind effect and the 3 million manuscripts a year in biomedical publishing were not seen in the sources read, so unsupported here, not refuted. Springer Nature's three million is for all scholarly articles, not biomedical alone. 'A reviewer's job is disagreement' is the author's opinion.
Related work
- arXiv: stop automating peer review without rigorous evaluation ↗Position paper.
- Springer Nature: building trust at scale, using AI to strengthen the scientific record ↗May 21, 2026.
- arXiv: AI-assisted peer review across research communities ↗Survey of venue policies.
Watch next
- Find the preprint that compares ICLR 2026 and Nature Communications reviews.
Sources
Provenance
The note above is reproduced unedited from the original post, first published on Threads on 1 October 2026 at 18:54 IST. Sources are the papers and datasets the note draws on.
View the original post ↗Embed this note
More notes
The air is now being asked to keep its own ledger
the air is now being asked to keep its own ledger: ecmwf’s aifs compo becomes the first ai model to forecast atmospheric composition globally every three hours, cleanair simulates 365 days of pm2.5 over china in ten seconds, and a unified framework maps six pollutants at one kilometer across the whole country. the air now files its own composition report.
read the note →The current is now being asked to draw its own map
the current is now being asked to draw its own map: china’s langya 2.0 predicts six ocean phenomena including internal waves and mesoscale eddies, a deep net called wenhai resolves eddies globally with air sea flux formulas built in, and scripps infers surface currents from the way temperature patterns deform in satellite images. the ocean now files its own circulation report.
read the note →The soil is now being asked to report its own carbon
the soil is now being asked to report its own carbon: a nix color sensor paired with generative data augmentation predicts soil organic carbon without a lab, random forest drives 74 percent of soil health mapping studies, and sentinel 2 tracks five year carbon change across france and italy from 922 samples. the dirt now files its own carbon account.
read the note →