Large language models struggle to generate good research ideas alone
Re-Search: Dissecting LLM Ideation Through a Two-Player Questioning Game
Artificial Intelligence
Summary
Figuring out new research ideas usually happens step by step, with feedback helping improve them. The authors created a game called Re-Search where language models try to guess a paper’s main idea by asking yes-or-no questions. They found that these models get most of their progress early on but then get stuck. The main problem is that models can recognize good ideas when shown to them but have trouble coming up with these ideas by themselves. This suggests adding ideas from outside sources could help improve research creativity.
What this means in practice
- •For ai system builders: Improve research support tools by combining model judgments with external idea suggestions to overcome generation limits discovered in LLM ideation.
- •For product managers in ai companies: Develop AI assistants that enhance innovation by enriching model-generated research questions with human or external inputs to help guide idea search.$Commercial implications: Enables sale of advanced research ideation AI assistants that improve creative output by integrating external knowledge into model workflows.
Authors
Muhan Gao, Hangxiao Zhu, Xinqi Su, Yu Zhang, Daniel Khashabi, Kuan-Hao Huang
Abstract
Research ideation is iterative: researchers refine their ideas step by step based on feedback. Yet existing evaluations obscure this search process, making it difficult to understand how LLMs explore the space of research ideas and what limits their progress. 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. Built from 200 ICML 2026 papers, Re-Search tracks how a model's idea evolves as it asks questions and receives binary oracle feedback, isolating ideation from execution. Across six recent LLMs, nearly all progress happens within the first 30 of 100 questions, which largely decide how the game ends; after that, most models plateau. Through controlled interventions on the candidate questions and the hypothesis space, 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. We further analyze the ideation traces and find that models fail by either over-exploiting one direction or over-exploring too many. 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.