Metraipolis › archive › dc_ca1f8f455276

Training Language Models to Fail Gracefully — arXiv:2610.20401

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.20401
License (as archived)
CC0-1.0
Archived
2026-10-15T14:20:19.784Z
SHA-256 of the archived text
f0227d5c03940f0b21ef293bda93459c9cc9935184e49589d7aeda2793b475c4
Stable URL of this page
https://metraipolis.com/documento/dc_ca1f8f455276

How to cite

Training Language Models to Fail Gracefully — arXiv:2610.20401 — arXiv (en), https://arxiv.org/abs/2610.20401 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ca1f8f455276 (Metraipolis archive dc_ca1f8f455276, sha256 f0227d5c03940f0b21ef293bda93459c9cc9935184e49589d7aeda2793b475c4)
BibTeX
@misc{dcca1f8f455276,
  title = {Training Language Models to Fail Gracefully — arXiv:2610.20401},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20401},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ca1f8f455276; sha256 f0227d5c03940f0b21ef293bda93459c9cc9935184e49589d7aeda2793b475c4},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20401. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reinforcement learning makes language models better at solving tasks, but the same mechanism also rewards guessing. When success is the only way to earn reward, opting for a safe fallback is always worse than a long-shot attempt. When a wrong action can corrupt a database or ship broken code, that incentive can be costly. Therefore, we propose training models to fail gracefully: to preserve successful behavior whenever possible while learning to execute a safe fallback policy when acting would be unreliable. However, naively using policy gradients with intermediate reward for fallback behavior is brittle due to unfavorable optimization dynamics. We address this challenge with GracefulRL, which reshapes the optimization landscape to promote graceful failure. Importantly, this behavior emerges directly from outcome-level reward signals, without annotations specifying when to fall back or additional inference-time machinery such as a verifier. Across a range of tasks and scales, we train models that largely preserve their task-solving capability while selectively falling back when acting would be unreliable. More broadly, our results show that reinforcement learning can not only incentivize models to succeed, but also to reason about the limits of their capabilities and change course when they determine that success is out of reach.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.LG, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG).
Pagina del lavoro: https://arxiv.org/abs/2610.20401 — autori e testo completo sono lì (qui omessi i nomi: nessun dato personale nel deposito).
Licenza: titolo e abstract sono metadati arXiv distribuiti in CC0 1.0; il testo completo ha la licenza scelta dagli autori, indicata sulla pagina.
Letto il 2026-10-15T13:53:35.743Z dalla pagina https://arxiv.org/list/cs.LG/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).