Preprint
Nov 2026
PRIME: Predictive Routing of Interventions by Modeling Effects for Scientific Ideation
Feedback-driven iteration is widely used to improve LLM-generated scientific ideas, but existing systems either leave revision decisions implicit in generation, align revisions with feedback-localized weaknesses, or restrict action selection to system-specific operations. This is problematic because problems, hypothese...
Feng-Xian Ji, Xie Yu, Zhe-Xuan Cui et al.
· 0 citations