The authors developed Self-Diagnosis-guided Terminal Credit Redistribution, or FAULT, for training language-model agents on multistep tasks.
Self-diagnosis pinpoints which agent decisions deserve reinforcement learning credit
FAULT converts verified natural-language error diagnoses into step-level training signals while preserving the trajectory’s terminal outcome.
Chinese Tech
Yihua Zhu · Qianying Liu · Weixu Qiao · Xuan Ren · Weiwei Xu · Wenbo Li · +7 more
Alibaba Group · Kyoto University · NII LLMC · Peking University · University of California, Los Angeles
Research Digest··3 min read
Zhu and colleagues address a weakness in agentic reinforcement learning: a final success or failure score does not reveal which action caused the outcome, and provides no comparative signal when every rollout receives the same result.
Why this paper
From University of the Chinese Academy of Sciences and 8 others
In one line
In agentic RL, self-diagnosed errors turned into step-level credit anchored by terminal outcomes recover learning signal from same-outcome groups.
What we could check
- ·No code link found
- ·No weights link found
- ·No dataset link found
- ·No compute details found
- ·No stated limitations found
- ✓Reports numbers on named benchmarks
Observed from the paper text and links we have. Absence here means we did not find it, not that it does not exist.
§