Saved in:
Bibliographic Details
Main Authors: Hu, Zhiyuan, Wang, Yibo, Dong, Hanze, Xu, Yuhui, Saha, Amrita, Xiong, Caiming, Hooi, Bryan, Li, Junnan
Format: Preprint
Published: 2025
Subjects:
Online Access:https://arxiv.org/abs/2505.10554
Tags: Add Tag
No Tags, Be the first to tag this record!
_version_ 1866918036175323136
author Hu, Zhiyuan
Wang, Yibo
Dong, Hanze
Xu, Yuhui
Saha, Amrita
Xiong, Caiming
Hooi, Bryan
Li, Junnan
author_facet Hu, Zhiyuan
Wang, Yibo
Dong, Hanze
Xu, Yuhui
Saha, Amrita
Xiong, Caiming
Hooi, Bryan
Li, Junnan
contents Large reasoning models (LRMs) already possess a latent capacity for long chain-of-thought reasoning. Prior work has shown that outcome-based reinforcement learning (RL) can incidentally elicit advanced reasoning behaviors such as self-correction, backtracking, and verification phenomena often referred to as the model's "aha moment". However, the timing and consistency of these emergent behaviors remain unpredictable and uncontrollable, limiting the scalability and reliability of LRMs' reasoning capabilities. To address these limitations, we move beyond reliance on prompts and coincidental "aha moments". Instead, we explicitly align models with three meta-abilities: deduction, induction, and abduction, using automatically generated, self-verifiable tasks. Our three stage-pipeline individual alignment, parameter-space merging, and domain-specific reinforcement learning, boosting performance by over 10\% relative to instruction-tuned baselines. Furthermore, domain-specific RL from the aligned checkpoint yields an additional gain in performance ceiling for both 7B and 32B models across math, coding, and science benchmarks, demonstrating that explicit meta-ability alignment offers a scalable and dependable foundation for reasoning. Code is available at: https://github.com/zhiyuanhubj/Meta-Ability-Alignment
format Preprint
id arxiv_https___arxiv_org_abs_2505_10554
institution arXiv
publishDate 2025
record_format arxiv
spellingShingle Beyond 'Aha!': Toward Systematic Meta-Abilities Alignment in Large Reasoning Models
Hu, Zhiyuan
Wang, Yibo
Dong, Hanze
Xu, Yuhui
Saha, Amrita
Xiong, Caiming
Hooi, Bryan
Li, Junnan
Computation and Language
Large reasoning models (LRMs) already possess a latent capacity for long chain-of-thought reasoning. Prior work has shown that outcome-based reinforcement learning (RL) can incidentally elicit advanced reasoning behaviors such as self-correction, backtracking, and verification phenomena often referred to as the model's "aha moment". However, the timing and consistency of these emergent behaviors remain unpredictable and uncontrollable, limiting the scalability and reliability of LRMs' reasoning capabilities. To address these limitations, we move beyond reliance on prompts and coincidental "aha moments". Instead, we explicitly align models with three meta-abilities: deduction, induction, and abduction, using automatically generated, self-verifiable tasks. Our three stage-pipeline individual alignment, parameter-space merging, and domain-specific reinforcement learning, boosting performance by over 10\% relative to instruction-tuned baselines. Furthermore, domain-specific RL from the aligned checkpoint yields an additional gain in performance ceiling for both 7B and 32B models across math, coding, and science benchmarks, demonstrating that explicit meta-ability alignment offers a scalable and dependable foundation for reasoning. Code is available at: https://github.com/zhiyuanhubj/Meta-Ability-Alignment
title Beyond 'Aha!': Toward Systematic Meta-Abilities Alignment in Large Reasoning Models
topic Computation and Language
url https://arxiv.org/abs/2505.10554