Reinforcing Agentic Creativity in Scientific Ideation with Night Science.

arXiv:2609.35706v1 Announce Type: new Abstract: Large language models (LLMs) excel at structured, verifiable tasks, but their low-entropy bias can produce homogeneous and predictable outputs, limiting their utility for open-ended scientific ideation. Effective