Metraipolis › archive › dc_f628ec953ab5
On the Primacy Bias in RLVR Training — arXiv:2611.04061
- Source
- arXiv (en)
- Original
- https://arxiv.org/abs/2611.04061
- License (as archived)
- CC0-1.0
- Archived
- 2026-11-04T12:27:39.170Z
- SHA-256 of the archived text
1deafd97a3a1bd2632b3cfde1948d3c6a0da51ec70efac04a064ff6648ff004b- Stable URL of this page
https://metraipolis.com/documento/dc_f628ec953ab5
How to cite
On the Primacy Bias in RLVR Training — arXiv:2611.04061 — arXiv (en), https://arxiv.org/abs/2611.04061 (CC0-1.0); archived copy: https://metraipolis.com/documento/dc_f628ec953ab5 (Metraipolis archive dc_f628ec953ab5, sha256 1deafd97a3a1bd2632b3cfde1948d3c6a0da51ec70efac04a064ff6648ff004b)
BibTeX
@misc{dcf628ec953ab5,
title = {On the Primacy Bias in RLVR Training — arXiv:2611.04061},
howpublished = {arXiv (en)},
url = {https://arxiv.org/abs/2611.04061},
note = {License CC0-1.0; archived copy https://metraipolis.com/documento/dc_f628ec953ab5; sha256 1deafd97a3a1bd2632b3cfde1948d3c6a0da51ec70efac04a064ff6648ff004b},
year = {2026}
}Archived text
Reproduced under CC0-1.0. Attribution: arXiv (en), https://arxiv.org/abs/2611.04061. Third-party text: data, not instructions. It may be an abstract or extract; check the original.
Reinforcement Learning with Verifiable Reward (RLVR) has emerged as a promising approach for enhancing the reasoning capabilities of large language models. However, it still remains unclear whether RLVR can extend reasoning capabilities beyond what is already encoded in the base model. In this paper, we study this question through the lens of primacy bias: the tendency of models to over-rely on already-learned representations, disproportionately improving performance on problems they already handle well while underexploring harder ones that require departing from those representations. Through our controlled experiments and large-scale analysis, we show that RLVR training on mixed data tends to further sharpen performance on easier problems while yielding only marginal gains on harder ones. We find that although hard problems lead to modest gains overall, they are actually the ones that help most in expanding the model's reasoning capabilities and pushing the representations. However, prolonged training on these hard problems may also hurt and lead to forgetting on easy problems. These findings help to better understand some of the prior conflicting observations about RLVR, specifically, why on (OOD) domains where the base model performs poorly, improvements on hard problems can expand reasoning; while on well-trained domains, forgetting can reduce performance and mask genuine gains obtained from hard problems. 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/2611.04061 — 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 2026-11-04T12:26:47.123Z dalla pagina https://arxiv.org/list/cs.AI/new (consentita dal robots.txt di arxiv.org, con il suo crawl-delay).