Ai2 study finds LLM novelty judges unreliable under small setup changes
Ai2 released Old Ideas, Novel Problems, a study of how reliably LLMs judge whether research ideas are novel. The authors built a benchmark from ICLR 2026 peer reviews, using reviewer comments on originality to label human-authored ideas, and paired them with AI-generated ideas. Across six LLM judges, small setup changes such as prompt wording shifted verdicts substantially, in some cases over 50 percentage points on pairwise accuracy. Retrieval and extra reasoning did not consistently help, and two purpose-built novelty evaluators underperformed a simple prompted Claude Opus 4.6 baseline. The benchmark, data, and code are released.
One primary source; other links are context, not independent corroboration.
- Primary sourceX discovery · automated brief
- Discovery linkX
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
Open full article view →- Linked article+30
- Fetched source page+30
- No editor link review+0
- Original non-discussion link+5
This measures the stored source trail and editorial steps. It does not certify accuracy, peer review, or independent corroboration. Popularity adds no points.
View full scoring →Read the scoring rubric →What the source says
Ai2 released Old Ideas, Novel Problems, a study of how reliably LLMs judge whether research ideas are novel. The authors built a benchmark from ICLR 2026 peer reviews, using reviewer comments on originality to label human-authored ideas, and paired them with AI-generated ideas. Across six LLM judges, small setup changes such as prompt wording shifted verdicts substantially, in some cases over 50 percentage points on pairwise accuracy. Retrieval and extra reasoning did not consistently help, and two purpose-built novelty evaluators underperformed a simple prompted Claude Opus 4.6 baseline. The benchmark, data, and code are released.
Key findings from the stored brief
- Changing a prompt's wording shifted GPT-5.4 pairwise accuracy from 93.5% to 40.9% on Human+Generated pairs.
- Retrieval and additional reasoning did not consistently improve novelty judgments across the tested setups.
- Two purpose-built novelty evaluation systems performed worse than a simple prompted Claude Opus 4.6 baseline.
- Benchmark labels come from ICLR reviewer comments; AI-generated ideas are treated as weak labels, not verified judgments.
Source claims not independently checked; AI-generated idea labels are weak assumptions rather than verified judgments.
This is a source-based abstract, not independent validation of the source's claims.
Evidence & sources
These links show where the signal came from and what the source page linked to. A linked page is context, not independent corroboration.
Related archive coverage
These relationships record a citation or topic overlap, not confirmation of the finding.
Grok Bot submitted the link. AINewsIQ fetched the primary source and prepared a source-based brief automatically. An editor has not reviewed the story; social attention and the brief do not independently verify its claims.
These are Grok Bot discovery notes. AINewsIQ prepared a brief from available source material. The bot's rationale, linked posts, authors, and stances have not been independently verified.
Ai2's Oct 27 post releases 'Old Ideas, Novel Problems', a study plus benchmark, data and code for testing LLMs that judge whether research ideas are novel. Across six judges from OpenAI and Anthropic, it reports that small changes to prompt, format or reasoning budget produce very different verdicts. Relevant to AI-scientist pipelines that rely on such judges.
Source claims not independently checked; AI-generated idea labels are weak assumptions rather than verified judgments.
Engagement counts, when shown, are observed snapshots. They do not establish accuracy or corroboration.
- x post ↗ · 86 likes, 8 reposts, 7 replies, 6 quotes · observed 10/27/2026, 10:09:03 PM
- Related post ↗
Timeline & updates
- Signal dated by source or discovery feed
This timestamp places the signal in the radar. It may be the discovery post date rather than the primary page's first publication date.
- Shared brief stored
AINewsIQ prepared one abstract from fetched source page.
A publication revision history is not stored for this signal. New engagement snapshots do not mean the source itself changed.
Why this may matter
Novelty judges are increasingly used to evaluate AI-generated research ideas and compare ideation systems. If verdicts flip under minor setup changes, conclusions about which system produces novel ideas may depend on evaluation choices rather than ideas themselves.
Limits and open questions
Source claims not independently checked; AI-generated idea labels are weak assumptions rather than verified judgments.
Organization-specific financial, security, and operational effects require the organization's own context and review.
Discussion
Ask about the source, add context, or point readers to a relevant follow-up.
Discussion
Comments are reviewed before publication. Discuss the evidence, not the person.
No published comments yet.
Reader discussion opens with accounts. You can read the original source and published comments now.