Metraipolis › archive › dc_ced47271b02f

Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents — arXiv:2610.21045

Source: arXiv (en)
Original: https://arxiv.org/abs/2610.21045
License (as archived): CC0-1.0
Archived: 2026-10-16T08:59:41.650Z
SHA-256 of the archived text: cbf5651e0bc8caaf3d2b884b7df9457f4cd1e46e6e7bd75dcd557c2d88716b9e
Stable URL of this page: https://metraipolis.com/documento/dc_ced47271b02f

How to cite

Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents — arXiv:2610.21045 — arXiv (en), https://arxiv.org/abs/2610.21045 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_ced47271b02f (Metraipolis archive dc_ced47271b02f, sha256 cbf5651e0bc8caaf3d2b884b7df9457f4cd1e46e6e7bd75dcd557c2d88716b9e)

BibTeX

@misc{dcced47271b02f,
  title = {Learning When to Withdraw: Curriculum Coverage and Rule-Specific Failures in Fine-Tuned Language Agents — arXiv:2610.21045},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.21045},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_ced47271b02f; sha256 cbf5651e0bc8caaf3d2b884b7df9457f4cd1e46e6e7bd75dcd557c2d88716b9e},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.21045. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Tool-using language agents must revise prepared actions when evidence changes or authority is revoked. Selective withdrawal means suspending affected actions, preserving supported work, and resuming only after sufficient repair. We study supervised fine-tuning in NAQD-Env, a synthetic environment with explicit dependencies and a deterministic reference policy. Four LoRA checkpoints of Qwen2.5-3B-Instruct compare an original curriculum with an enriched one, the latter under three system-prompt framings with identical views and targets. Three instruction-tuned baselines and all four checkpoints are evaluated under three inference prompts on fresh validation cases, held-out dependency families, and held-out structural combinations. Under the naqd prompt, SFT checkpoints reach decision agreement of 0.77-0.92 and withdrawal recall of 0.46-0.86, versus 0.45-0.63 and at most 0.06 for the baselines. Rule-level diagnostics qualify these gains. The original-curriculum checkpoint wrongly withdraws 26 of 34 actions with surviving alternative support; enriched checkpoints withdraw 0-1. Conversely, all SFT checkpoints narrow prohibited combinations less often than the 7B and 8B baselines, a case absent from both curricula. Delegation results show more withdrawal but poorer preservation of independently authorized actions, and all four checkpoints omit valid event reports. Framing contrasts vary across metrics and splits and do not establish a reproducible advantage or equivalence. One training run per arm, jointly changed curriculum properties, prior evaluation exposure, and baseline-SFT quantization differences limit causal attribution. The contribution is a method for detecting rule-specific regressions that aggregate accuracy can conceal, not evidence that fine-tuning alone ensures agent safety.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.AI, nuovi lavori del giorno).
Categorie arXiv: Artificial Intelligence (cs.AI).
Pagina del lavoro: https://arxiv.org/abs/2610.21045 — 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 2026-10-16T08:44:17.689Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).