Researchers have developed a new framework to train large language models for multi-turn strategic persuasion. This framework addresses challenges in simulating realistic persuadees and providing effective supervision for dialogue credit assignment. It introduces PersuArena, an environment for generating realistic persuasion trajectories, and Probe-RPO, an online reinforcement learning method that uses introspective probes from the persuadee's internal state to create dense process rewards. Experiments show this approach significantly increases acceptance rates compared to existing methods across various domains. AI
IMPACT This research offers a novel approach to training LLMs for complex strategic dialogue, potentially improving their capabilities in areas like negotiation and advice-giving.
RANK_REASON The cluster contains a research paper detailing a new framework and methodology for training LLMs. [lever_c_demoted from research: ic=1 ai=1.0]
- alphaXiv
- arXiv
- CatalyzeX
- DagsHub
- Gotit.pub
- GRPO
- Hugging Face
- IArxiv
- PersuArena
- Probe-RPO
- ScienceCast
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →
