AI TrackerFrontier desk · AEST
Back to news
normalaitech

Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game

arXiv cs.AI · 11 Nov 2026, 04:47 AEST

A new research tool called Re-Search has been developed to analyze how large language models (LLMs) ideate research ideas. The study reveals that most progress occurs early in the questioning process, highlighting the limitations of LLMs in generating original ideas.

Key facts

Read aloud

A new research tool called Re-Search has been developed to analyze how large language models L L M s ideate research ideas. The study reveals that most progress occurs early in the questioning process, highlighting the limitations of L L M s in generating original ideas. Research ideation is iterative: researchers refine their ideas step by step based on feedback. We introduce Re-Search, which adapts the format of Twenty Questions to research ideation: given only the research problem of a published paper, a model must rediscover the paper's core idea by asking yes-or-no questions. Across six recent L L M s, nearly all progress happens within the first thirty of one hundred questions, which largely decide how the game ends; after that, most models plateau. We find that the bottleneck is generation, not judgment: models recognize a good idea or a good question when one is offered, but often cannot produce it on their own. These findings point to concrete design principles for autoresearch systems: since models can judge candidates they cannot produce, enriching the search with externally sourced ideas and questions may be more effective than improving any single model's generation ability.

Sources in this cluster