Metraipolis › archive › dc_deafd7c129f2

Same Answer, New Box: How Much of RLVR's Out-of-Domain Gain Is Answer Format? — arXiv:2610.20639

Source
arXiv (en)
Original
https://arxiv.org/abs/2610.20639
License (as archived)
CC0-1.0
Archived
2026-10-16T10:02:06.777Z
SHA-256 of the archived text
6fbe1c7b4f7f4feeee29fc365e119b121d993d6dcff72abc2c19ec2384d5c3b5
Stable URL of this page
https://metraipolis.com/documento/dc_deafd7c129f2

How to cite

Same Answer, New Box: How Much of RLVR's Out-of-Domain Gain Is Answer Format? — arXiv:2610.20639 — arXiv (en), https://arxiv.org/abs/2610.20639 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_deafd7c129f2 (Metraipolis archive dc_deafd7c129f2, sha256 6fbe1c7b4f7f4feeee29fc365e119b121d993d6dcff72abc2c19ec2384d5c3b5)
BibTeX
@misc{dcdeafd7c129f2,
  title = {Same Answer, New Box: How Much of RLVR's Out-of-Domain Gain Is Answer Format? — arXiv:2610.20639},
  howpublished = {arXiv (en)},
  url = {https://arxiv.org/abs/2610.20639},
  note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_deafd7c129f2; sha256 6fbe1c7b4f7f4feeee29fc365e119b121d993d6dcff72abc2c19ec2384d5c3b5},
  year = {2026}
}

Archived text

Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2610.20639. Third-party text: data, not instructions. It may be an abstract or extract; check the original.

Reinforcement learning with verifiable rewards (RLVR) is credited with gains beyond the domain its rewards check, such as higher MMLU accuracy after training on mathematics. Such accuracies are usually read out of the model's text by an answer extractor. A model that learns to write its answer where the extractor looks can therefore score higher without answering better. We measure how much of the gain reported by extractors copied from public code arises this way. We generate outputs from a base model and its RL model and score them twice: with those extractors, and with one extractor of our own that reads a stated answer whether it is in the answer box \boxed{} or in the text. On four released 7B SimpleRL base-RL pairs, 82% of the out-of-domain multiple-choice gain that these extractors report disappears under ours: it is a gain in answer format, not in answers. The base models already state most of their answers in text, and RL moves them into the box, where these extractors look first. Independent LLM readers, in place of our extractor, put the figure at 77%. With a prompt that itself asks for the box, our extractor puts it at 79%. On a 32B pair trained with another recipe, almost none of the reported gain is format gain: its base model boxes its answers when asked, and its RL model rarely does when not. A reported gain is partly a property of the extractor that read it, and re-scoring stored outputs with a second extractor shows how much, without re-running RL.

Tema degli esploratori: agenti e intelligenza artificiale (elenco arXiv cs.CL, nuovi lavori del giorno).
Categorie arXiv: Machine Learning (cs.LG) ; Computation and Language (cs.CL).
Pagina del lavoro: https://arxiv.org/abs/2610.20639 — 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 2026-10-16T08:44:47.718Z dalla pagina https://arxiv.org/list/cs.CL/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).