Metraipolis › archive › dc_5b552bcef54a
ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction — arXiv:2608.13622
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2608.13622
- License (as archived)
- CC0-1.0
- Archived
- 2026-10-19T02:20:20.275Z
- SHA-256 of the archived text
b1803570a9ac3a33ef4244172c3472105300619dd55a73aa4a8719f11e92673e- Stable URL of this page
https://metraipolis.com/documento/dc_5b552bcef54a
How to cite
ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction — arXiv:2608.13622 — arXiv (en), https://arxiv.org/abs/2608.13622 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_5b552bcef54a (Metraipolis archive dc_5b552bcef54a, sha256 b1803570a9ac3a33ef4244172c3472105300619dd55a73aa4a8719f11e92673e)
BibTeX
@misc{dc5b552bcef54a,
title = {ARC: Fair Relative Advantage Comparison in Open-Ended Real-World Interaction — arXiv:2608.13622},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2608.13622},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_5b552bcef54a; sha256 b1803570a9ac3a33ef4244172c3472105300619dd55a73aa4a8719f11e92673e},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.13622. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Open-ended real-world interaction admits multiple valid behaviors: an agent may answer directly, ask for clarification, provide progress updates, or confirm before acting. This flexibility breaks a core assumption behind group-based RL: rollouts compared within a group are no longer guaranteed to be behaviorally comparable. As a result, reward-model preferences over interaction style can distort relative advantages and steer optimization toward reward-preferred behaviors rather than context-appropriate ones. We formalize this as a \textit{reward fairness problem} and propose \textbf{ARC} (Advantage Regularization via Conditioning), a training recipe that restores fairer relative comparison through strategy-conditioned rollout grouping, together with hybrid rewards and entropy regularization. We study ARC in our proposed \inter, a novel paradigm for responsive, steerable, and execution-aware user-agent interaction that decouples user-visible communication from latent reasoning and tool use. \inter\ also provides the annotation and distillation pipeline for constructing \inter-86K, our strategy-annotated training corpus for supervised and RL training. Empirically, ARC substantially strengthens the core $\tau/\tau^2$ tool-use benchmarks, while \inter\ reduces time-to-first-token from 4.91s to 1.27s relative to a think-style baseline. Together, these results suggest that a central bottleneck in open-ended interactive learning is not only how agents are rewarded, but whether their behaviors are compared fairly in the first place. The ARC implementation and \inter-86K training data will be released.
Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-08-17 in cs.AI).
Categorie arXiv: Artificial Intelligence (cs.AI) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.13622 — 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 2026-10-19T02:02:02.727Z dalla pagina https://arxiv.org/catchup/cs.AI/2026-08-17?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).