Pujun Zheng

ARE-bench: Benchmarking Agents' Research Ability in Real-World Researcher Sessions and Workspaces

Jiahang Lin, Wanying Ren, Wanghan Xu, et al.

MoRI: Learning Motivation-Grounded Reasoning for Scientific Ideation in Large Language Models

Chenyang Gu, Jiahao Cheng, Meicong Zhang, et al.

MoRI is a framework that teaches AI language models to generate better scientific ideas by learning to reason from research motivations to actual methods, rather than just mixing existing concepts together. It uses a two-part training approach: first teaching the model to identify research motivations, then using reinforcement learning to ensure the generated ideas are technically rigorous and scientifically sound. The results show this method produces more novel, rigorous, and feasible scientific proposals compared to existing AI approaches.