Learning to Ideate for Scientific Impact
arXiv:2609.29802v1 Announce Type: new Abstract: Scientific ideation is increasingly mediated by large language models, but current ideation systems are usually trained and evaluated on immediately judgeable proxies such as novelty, clarity, and feasibility. This leaves open whether delayed signals…