COLLABLLM

Summary from the TLDR AI newsletter

CollabLLM is an open-source framework that trains language models to ask strategic questions and guide conversations rather than make assumptions, using "Multiturn-aware Reward" that evaluates responses based on their long-term impact on conversation outcomes rather than immediate helpfulness.