Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning
Fuente:
arXiv
Enregistré dans:
| Auteurs principaux: | , , , , , , , , |
|---|---|
| Format: | Preprint |
| Publié: |
2025
|
| Sujets: | |
| Accès en ligne: | |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866915563995922432 |
|---|---|
| author | Huan, Maggie Li, Yuetai Zheng, Tuney Xu, Xiaoyu Kim, Seungone Du, Minxin Poovendran, Radha Neubig, Graham Yue, Xiang |
| author_facet | Huan, Maggie Li, Yuetai Zheng, Tuney Xu, Xiaoyu Kim, Seungone Du, Minxin Poovendran, Radha Neubig, Graham Yue, Xiang |
| contents | Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments on Qwen3-14B models using math-only data but different tuning methods. We find that reinforcement learning (RL)-tuned models generalize well across domains, while supervised fine-tuning (SFT)-tuned models often forget general capabilities. Latent-space representation and token-space distribution shift analyses reveal that SFT induces substantial representation and output drift, while RL preserves general-domain structure. Our results suggest a need to rethink standard post-training recipes, particularly the reliance on SFT-distilled data for advancing reasoning models. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2507_00432 |
| institution | arXiv |
| publishDate | 2025 |
| record_format | arxiv |
| spellingShingle | Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning Huan, Maggie Li, Yuetai Zheng, Tuney Xu, Xiaoyu Kim, Seungone Du, Minxin Poovendran, Radha Neubig, Graham Yue, Xiang Artificial Intelligence Computation and Language Math reasoning has become the poster child of progress in large language models (LLMs), with new models rapidly surpassing human-level performance on benchmarks like MATH and AIME. But as math leaderboards improve week by week, it is worth asking: do these gains reflect broader problem-solving ability or just narrow overfitting? To answer this question, we evaluate over 20 open-weight reasoning-tuned models across a broad suite of tasks, including math, scientific QA, agent planning, coding, and standard instruction-following. We surprisingly find that most models that succeed in math fail to transfer their gains to other domains. To rigorously study this phenomenon, we conduct controlled experiments on Qwen3-14B models using math-only data but different tuning methods. We find that reinforcement learning (RL)-tuned models generalize well across domains, while supervised fine-tuning (SFT)-tuned models often forget general capabilities. Latent-space representation and token-space distribution shift analyses reveal that SFT induces substantial representation and output drift, while RL preserves general-domain structure. Our results suggest a need to rethink standard post-training recipes, particularly the reliance on SFT-distilled data for advancing reasoning models. |
| title | Does Math Reasoning Improve General LLM Capabilities? Understanding Transferability of LLM Reasoning |
| topic | Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2507.00432 |