Research

Paper

AI LLM March 16, 2026

HindSight: Evaluating Research Idea Generation via Future Impact

Authors

Bo Jiang

Abstract

Evaluating AI-generated research ideas typically relies on LLM judges or human panels -- both subjective and disconnected from actual research impact. We introduce \hs{}, a time-split evaluation framework that measures idea quality by matching generated ideas against real future publications and scoring them by citation impact and venue acceptance. Using a temporal cutoff~$T$, we restrict an idea generation system to pre-$T$ literature, then evaluate its outputs against papers published in the subsequent 30 months. Experiments across 10 AI/ML research topics reveal a striking disconnect: LLM-as-Judge finds no significant difference between retrieval-augmented and vanilla idea generation ($p{=}0.584$), while \hs{} shows the retrieval-augmented system produces 2.5$\times$ higher-scoring ideas ($p{<}0.001$). Moreover, \hs{} scores are \emph{negatively} correlated with LLM-judged novelty ($ρ{=}{-}0.29$, $p{<}0.01$), suggesting that LLMs systematically overvalue novel-sounding ideas that never materialize in real research.

Metadata

arXiv ID: 2603.15164
Provider: ARXIV
Primary Category: cs.CL
Published: 2026-03-16
Fetched: 2026-03-17 06:02

Related papers

Raw Data (Debug)
{
  "raw_xml": "<entry>\n    <id>http://arxiv.org/abs/2603.15164v1</id>\n    <title>HindSight: Evaluating Research Idea Generation via Future Impact</title>\n    <updated>2026-03-16T11:59:24Z</updated>\n    <link href='https://arxiv.org/abs/2603.15164v1' rel='alternate' type='text/html'/>\n    <link href='https://arxiv.org/pdf/2603.15164v1' rel='related' title='pdf' type='application/pdf'/>\n    <summary>Evaluating AI-generated research ideas typically relies on LLM judges or human panels -- both subjective and disconnected from actual research impact. We introduce \\hs{}, a time-split evaluation framework that measures idea quality by matching generated ideas against real future publications and scoring them by citation impact and venue acceptance. Using a temporal cutoff~$T$, we restrict an idea generation system to pre-$T$ literature, then evaluate its outputs against papers published in the subsequent 30 months. Experiments across 10 AI/ML research topics reveal a striking disconnect: LLM-as-Judge finds no significant difference between retrieval-augmented and vanilla idea generation ($p{=}0.584$), while \\hs{} shows the retrieval-augmented system produces 2.5$\\times$ higher-scoring ideas ($p{&lt;}0.001$). Moreover, \\hs{} scores are \\emph{negatively} correlated with LLM-judged novelty ($ρ{=}{-}0.29$, $p{&lt;}0.01$), suggesting that LLMs systematically overvalue novel-sounding ideas that never materialize in real research.</summary>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.CL'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.AI'/>\n    <category scheme='http://arxiv.org/schemas/atom' term='cs.LG'/>\n    <published>2026-03-16T11:59:24Z</published>\n    <arxiv:primary_category term='cs.CL'/>\n    <author>\n      <name>Bo Jiang</name>\n    </author>\n  </entry>"
}