CollabSciReM: Aligning Language Models for Scientific Conversations via Science-Aware Reward Decomposition.

arXiv:2610.27901v1 Announce Type: cross Abstract: Large Language Models (LLMs) are increasingly used as research collaborators for hypothesis refinement, methodological planning, and technical analysis across multi-turn interactions. In this setting, a useful