The authors frame social adaptation as multi-turn reinforcement learning.
Preference-grouped reinforcement learning improves social adaptation in simulated conversations
The authors train an LLM agent on whole-conversation satisfaction, normalizing rewards across simulated users with similar hidden preferences.
Big Tech
Jingquan Wang · Jun Yin · Xu Han · Yongsheng Mei · Jie Hao · Bin Guo
Amazon
Research Digest··3 min read
Wang and colleagues introduce preference-batched group relative policy optimization, or PB-GRPO, for training conversational agents to infer and accommodate users’ unstated preferences over multiple turns.
Why this paper
From Amazon
In one line
PB-GRPO improves LLM social behavior by combining persona-driven simulation with preference-batched GRPO training.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ✓Limitations stated by the authors (3 noted)
- ·No benchmark numbers found
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§