독서리서치레이더투자 프레임워크
로그인 / 회원가입
← 읽기로 돌아가기
뉴스주제 · Ai2新颖性基准중대1 분 소요DisconfirmAI 편집부

Ai2 Releases Benchmark Showing LLMs Are Unreliable at Judging Novelty of AI-Generated Research Ideas

Ai2's new Novelty Bench reveals that LLM-based 'novelty judges' are highly unstable when evaluating AI-generated scientific ideas, with minor prompt changes causing massive accuracy swings.

2026-10-28 00:12 GMT+8

The Allen Institute for AI (Ai2) released a new study titled "Old Ideas, Novel Problems" alongside its "Novelty Bench" benchmark on October 27, 2026. The research evaluates how reliably large language models (LLMs) can assess the novelty of scientific research ideas, a critical function for AI systems increasingly involved in automating scientific discovery.

Testing six major LLMs, including GPT-5.1, GPT-5.4, and Claude Opus 4.6, the study found extreme instability when judging AI-generated ideas. In one experiment, changing just a few words in the prompt caused GPT-5.4's accuracy to drop from 93.5% to 40.9%—below chance level—on pairs containing one human-authored and one AI-generated idea. This represents a shift of over 50 percentage points due to minor wording variations. By contrast, the same prompt change affected accuracy by only about seven percentage points when judging human-authored ideas, suggesting models are significantly less reliable when evaluating synthetic content.

The findings also indicate that adding retrieval steps or increasing reasoning budgets does not consistently improve judgments. Two purpose-built novelty evaluation systems performed worse than a simple prompted baseline. This highlights a significant limitation in current AI-for-science workflows: the components designed to filter for genuine innovation are currently too fragile to be trusted without substantial human oversight.

출처

로그인 / 회원가입
독서리서치레이더투자 프레임워크
huggingface.co
↗
주제

Ai2新颖性基准

美国艾伦人工智能研究所(Ai2)发布“Old Ideas, Novel Problems”研究及Novelty Bench基准,揭示大语言模型(LLM)作为科研创意“新颖性法官”极不可靠。研究发现,仅微调提示词即可导致GPT-5.4在评估A…

3 단계1 건업데이트됨 2026-10-28 13:11

현재까지의 과정

  1. 2026-10-27Ai2在Hugging Face博客发布《Old Ideas, Novel Problems》论文及Novelty Bench基准,指出LLM评估科研创意新颖性时存在严重不稳定性。huggingface.co ↗
  2. 2026-10-27实验显示,改变提示词措辞可使GPT-5.4在Human+Generated场景下的准确率从93.5%降至40.9%,低于随机水平。runtimewire.com ↗
  3. 2026-10-27对比测试表明,简单的Claude Opus 4.6提示基线(Macro-F1 0.854)优于Idea Novelty Checker和Sakana AI Scientist等专用系统,且成本低数十倍。neuraldaily.news ↗

아직 해결되지 않은 문제

  • 其他主流LLM(如Gemini系列)在相同基准下的表现是否也呈现类似的提示词敏感性?
  • Ai2提出的自动管道能否被社区广泛采用以建立更标准化的新颖性评估协议?
전체 주제 읽기 →
리서치 계속하기 →