Internalizing LLM Reasoning via Discovery and Replay of Latent Actions
Fuente:
arXiv
Saved in:
| Main Authors: | , , , , , |
|---|---|
| Format: | Preprint |
| Published: |
2026
|
| Subjects: | |
| Online Access: | |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1866917248679018496 |
|---|---|
| author | Shi, Zhenning Zhu, Yijia Shi, Junhan Zhang, Xun Wang, Lei Miao, Congcong |
| author_facet | Shi, Zhenning Zhu, Yijia Shi, Junhan Zhang, Xun Wang, Lei Miao, Congcong |
| contents | The internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute. However, existing activation steering methods rely on static control vectors that fail to adapt to the non-stationary evolution of complex reasoning tasks. To address this limitation, we propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem. STIR introduces a synergistic three-stage pipeline: (1) differential intrinsic action induction harvests latent reasoning successes to crystallize steering primitives; (2) sparse control basis construction curates a compact, geometrically diverse tool library; and (3) value-modulated trajectory intervention dynamically injects context-specific impulses via anchor-based gating. Extensive experiments on six arithmetic and logical benchmarks across four representative models demonstrate that STIR improves average accuracy by 1.9% to 7.5% while reducing average token consumption by up to 35% compared to vanilla decoding. These findings demonstrate that the benefits of explicit chain-of-thought can be realized through dynamic latent trajectory control, internalizing the reasoning process to bypass the explicit generation while achieving superior fidelity. Our code is available at https://github.com/sznnzs/LLM-Latent-Action. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2602_04925 |
| institution | arXiv |
| publishDate | 2026 |
| record_format | arxiv |
| spellingShingle | Internalizing LLM Reasoning via Discovery and Replay of Latent Actions Shi, Zhenning Zhu, Yijia Shi, Junhan Zhang, Xun Wang, Lei Miao, Congcong Machine Learning Artificial Intelligence Computation and Language The internalization of chain-of-thought processes into hidden states has emerged as a highly efficient paradigm for scaling test-time compute. However, existing activation steering methods rely on static control vectors that fail to adapt to the non-stationary evolution of complex reasoning tasks. To address this limitation, we propose STIR (Self-Distilled Tools for Internal Reasoning), a framework that reformulates reasoning enhancement as a dynamic latent trajectory control problem. STIR introduces a synergistic three-stage pipeline: (1) differential intrinsic action induction harvests latent reasoning successes to crystallize steering primitives; (2) sparse control basis construction curates a compact, geometrically diverse tool library; and (3) value-modulated trajectory intervention dynamically injects context-specific impulses via anchor-based gating. Extensive experiments on six arithmetic and logical benchmarks across four representative models demonstrate that STIR improves average accuracy by 1.9% to 7.5% while reducing average token consumption by up to 35% compared to vanilla decoding. These findings demonstrate that the benefits of explicit chain-of-thought can be realized through dynamic latent trajectory control, internalizing the reasoning process to bypass the explicit generation while achieving superior fidelity. Our code is available at https://github.com/sznnzs/LLM-Latent-Action. |
| title | Internalizing LLM Reasoning via Discovery and Replay of Latent Actions |
| topic | Machine Learning Artificial Intelligence Computation and Language |
| url | https://arxiv.org/abs/2602.04925 |