user preferences

Weak Rewards for Persistent LLM User Models

This paper explores using weak rewards from retrieval-augmented interaction to model user preferences in chat agents.