Metraipolis › archive › dc_fbcccc56f4c2
Improving Instruction Hierarchy by Learning from Goal-Goal Conflicts — arXiv:2610.35093
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2610.35093
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-27T07:58:26.654Z
- SHA-256 of the archived text
ab1f53cbc4eeb16076cb4aba1d0b43900307eefcb445bf6c96601398e36250fa- Stable URL of this page
https://metraipolis.com/documento/dc_fbcccc56f4c2
How to cite
Improving Instruction Hierarchy by Learning from Goal-Goal Conflicts — arXiv:2610.35093 — arXiv (en), https://arxiv.org/abs/2610.35093 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_fbcccc56f4c2 (Metraipolis archive dc_fbcccc56f4c2, sha256 ab1f53cbc4eeb16076cb4aba1d0b43900307eefcb445bf6c96601398e36250fa)
BibTeX
@misc{dcfbcccc56f4c2,
title = {Improving Instruction Hierarchy by Learning from Goal-Goal Conflicts — arXiv:2610.35093},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2610.35093},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_fbcccc56f4c2; sha256 ab1f53cbc4eeb16076cb4aba1d0b43900307eefcb445bf6c96601398e36250fa},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.35093. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
The recent incident in which OpenAI agents autonomously compromised Hugging Face's infrastructure illustrates the importance of instruction hierarchy (IH): an LLM should follow lower-priority instructions, such as user requests, only when they do not conflict with higher-priority instructions, such as system policies. IH can be effectively improved by fine-tuning LLMs on simulated examples of instruction conflicts, whose diversity is key to teaching LLMs to properly handle conflicts at test time. However, our decomposition of a task into goals and constraints reveals a conflict type -- goal-goal conflicts -- that is absent from existing IH training sets from OpenAI and NVIDIA. We show the importance of this conflict type: fine-tuning Qwen3.5-9B and Gemma4-12B-it on examples of this conflict type significantly improves IH on a wide range of real-world tasks unseen in training; adding these examples to existing IH training sets for fine-tuning also gives better IH adherence. The IH gains come with trivial mean utility changes on no-conflict tasks across eight benchmarks. Our work advocates including this new conflict type in IH training, even if these examples do not reflect typical LLM use. 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/2610.35093 — 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 2026-10-27T07:55:30.735Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).