Metraipolis › archive › dc_70846fdad678

Co-Evolving Actor-Conditioned Critics for Non-Verifiable Generation — arXiv:2608.30397

Source
arXiv (en)
Original
https://arxiv.org/abs/2608.30397
License (as archived)
CC0-1.0
Archived
2026-10-17T08:07:31.765Z
SHA-256 of the archived text
b913c90e1b6c2151bb2944d1e1e9fc055cc269f0726b51730a736be7492647b8
Stable URL of this page
https://metraipolis.com/documento/dc_70846fdad678

How to cite

Co-Evolving Actor-Conditioned Critics for Non-Verifiable Generation — arXiv:2608.30397 — arXiv (en), https://arxiv.org/abs/2608.30397 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_70846fdad678 (Metraipolis archive dc_70846fdad678, sha256 b913c90e1b6c2151bb2944d1e1e9fc055cc269f0726b51730a736be7492647b8)
BibTeX
@misc{dc70846fdad678,
  title = {Co-Evolving Actor-Conditioned Critics for Non-Verifiable Generation — arXiv:2608.30397},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2608.30397},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_70846fdad678; sha256 b913c90e1b6c2151bb2944d1e1e9fc055cc269f0726b51730a736be7492647b8},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2608.30397. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Natural-language critiques provide supervision beyond scalar rewards for non-verifiable generation, which lacks deterministic verifiers. In critique-guided refinement, a critic gives feedback on an initial response and an actor revises it. However, final revision quality does not reveal whether the critique was actually useful: a capable actor may improve without following the feedback, while valid feedback may fail if the actor cannot execute it. We frame critique as actor-conditioned revision guidance, where usefulness depends on whether the feedback helps the target actor address the intended weakness. We introduce TAIScore (Targeted Actionable Improvement Score), a reward that evaluates the instruction, initial response, critique, and revision together, assessing whether the critique targets a real weakness, whether the actor follows it, and whether the intended aspect improves. We use this reward to train an actor-tailored critic with GRPO, and use critique-guided refinements to construct DPO preference pairs for the actor, forming a co-evolving critic-actor loop where the critic adapts to the actor's changing capability. Experiments show that an 8B critic trained with TAIScore outperforms both a zero-shot 120B critic and critics trained with outcome-only or critique-only reward signals. Co-evolving the critic and actor further improves performance, suggesting that effective critique supervision should adapt as the actor changes.

Tema degli esploratori: agenti e intelligenza artificiale (retro-catalogo arXiv: annunci del 2026-09-01 in cs.CL).
Categorie arXiv: Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2608.30397 — 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 2026-10-17T07:47:52.811Z dalla pagina https://arxiv.org/catchup/cs.CL/2026-09-01?abs=True (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).