Can we trust LLMs to judge whether a research idea is novel?
Comments on “Can we trust LLMs to judge whether a research idea is novel?”
3 judges. Scored by all three judges. Only the day’s top stories get the claim check. How far stories are checked
Ai2's new study, "Old Ideas, Novel Problems," tests whether LLM "novelty judges" can reliably assess whether research ideas are new. Using ICLR 2026 peer reviews to label human-authored ideas, the team found that small prompt changes can shift judge accuracy by more than 50 points on AI-generated ideas, that extra…
Why it ranked
No. 62 of 155 with a score of 0.61. The panel rated it highest on new information (0.93) and substantiated (0.84), lowest on practical relevance (0.60).
- AI judgeJudge 2claude-haiku-5-5
- Strong AI-evaluation research with a benchmark and reproducible pipeline, but its claims rest on a weakly-labeled setup, and it is a lab announcement of a niche study rather than a field-changing result.
- AI judgeJudge 3gpt-5.6-luna
- Primary benchmark, dataset, code, and quantitative comparisons make this technically substantive and broadly relevant to AI research workflows. A missing result value and promotional closing slightly reduce confidence, but the central findings are well supported.
Each judge wrote its line after answering the rubric, with the site and the author still hidden. Every answer behind this is in the table under Score.
Score
0.61total under rubric v2.2
- Significance0.62
- Reader interest0.69
- Audience breadth0.63
- Practical relevance0.60
- Substantiated0.84
- Technical depth0.72
- New information0.93
- Counts against: Promotional0.16
- Counts against: Overclaim0.24
Each judge’s answer per question (0–1), their mean and spread, and what the question adds to the score (weight × mean).
| Question | Judge 1 | Judge 2 | Judge 3 | Mean | Spread | Weight | Contribution (weight times mean) |
|---|---|---|---|---|---|---|---|
| Significance | 0.64 | 0.45 | 0.78 | 0.62 | 0.33 | +0.24 | +0.150 |
| Reader interest | 0.56 | 0.62 | 0.90 | 0.69 | 0.34 | +0.26 | +0.180 |
| Audience breadth | 0.62 | 0.55 | 0.72 | 0.63 | 0.17 | +0.13 | +0.082 |
| Practical relevance | 0.52 | 0.55 | 0.74 | 0.60 | 0.22 | +0.22 | +0.133 |
| Substantiated | 0.98 | 0.72 | 0.82 | 0.84 | 0.26 | +0.05 | +0.042 |
| Technical depth | 0.78 | 0.60 | 0.78 | 0.72 | 0.18 | +0.05 | +0.036 |
| New information | 0.90 | 0.90 | 0.98 | 0.93 | 0.08 | +0.05 | +0.046 |
| Promotional | 0.09 | 0.20 | 0.18 | 0.16 | 0.11 | −0.19 | −0.030 |
| Overclaim | 0.24 | 0.25 | 0.22 | 0.24 | 0.03 | −0.11 | −0.026 |
| Sum of contributions | +0.613 | ||||||
| Claim-check penalty | 0.000 | ||||||
| Total, as published | 0.613 | ||||||
Spread is the highest minus the lowest judge answer. marks a spread above 0.40: the judges disagree, and the story is marked contested.
The badge for this ranking, for the author's page or anyone's: one line of HTML, and it links back here.
<a href="https://rubric.hrih.si/editions/2026-10-28/62/"><img src="https://rubric.hrih.si/badge/2026-10-28/62.svg" alt="Ranked #62 on Rubric, 28 Oct 2026" width="245" height="28"></a>Also covered by
No other coverage was clustered with this story.
Elsewhere
Elsewhere: HN —
Recorded for context only, checked 28 Oct 2026, 06:47 UTC. It played no part in the ranking.