, modified versions of Werewolf and Resistance-style games).
Reinforcement fine-tuning improves LLM social reading more than game wins
Fine-tuned language model agents learn to infer hidden roles and predict others' decisions, but directly optimizing win–loss outcomes proves unreliable for acquiring general social deduction skills.
Chinese Tech
Lingzhe Zhang · Yunpeng Zhai · Tong Jia · Kening Zheng · Chiming Duan · Minghua He · +4 more
Peking University · Alibaba Group · University of Illinois Chicago
Research Digest··3 min read
Zhang et al.
Why this paper
From Alibaba Group and 2 others
In one line
Reinforcement fine-tuning improves LLMs' social reading and influence in deduction games, but not when only terminal outcomes are optimized.
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
- ·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.
§