arXiv:2512.13979v2 Announce Type: replace
Abstract: Large reasoning models achieve strong performance on diverse tasks by producing extended chains of thought. Self-reflection, the ability to review and revise prior reasoning steps, is widely regarded as a key contributor to this performance. However, self-reflection also incurs substantial inference cost, and its governing mechanism remains underexplored. In this work, we study self-reflection through the lens of representation engineering. First, we identify a reflection direction in the model's latent space that separates reflection steps from non-reflection steps, and show that activation along this direction is strongly predictive of answer correctness, suggesting that self-reflection is regulated by the model's internal uncertainty. Next, building on this insight, we propose ReflCtrl, a framework that controls self-reflection via a stepwise steering method: interventions are applied only at the start of each new reasoning step, enabling fine-grained control over reflection frequency without degrading generation quality. Experiments across math and general reasoning benchmarks show that reflection is often redundant, especially in stronger models: ReflCtrl reduces total reasoning tokens by up to 43.2% while preserving accuracy, and substantially outperforms the conventional approach that steers at every token, at matched token budgets.
Abstract: Large reasoning models achieve strong performance on diverse tasks by producing extended chains of thought. Self-reflection, the ability to review and revise prior reasoning steps, is widely regarded as a key contributor to this performance. However, self-reflection also incurs substantial inference cost, and its governing mechanism remains underexplored. In this work, we study self-reflection through the lens of representation engineering. First, we identify a reflection direction in the model's latent space that separates reflection steps from non-reflection steps, and show that activation along this direction is strongly predictive of answer correctness, suggesting that self-reflection is regulated by the model's internal uncertainty. Next, building on this insight, we propose ReflCtrl, a framework that controls self-reflection via a stepwise steering method: interventions are applied only at the start of each new reasoning step, enabling fine-grained control over reflection frequency without degrading generation quality. Experiments across math and general reasoning benchmarks show that reflection is often redundant, especially in stronger models: ReflCtrl reduces total reasoning tokens by up to 43.2% while preserving accuracy, and substantially outperforms the conventional approach that steers at every token, at matched token budgets.
This article has been indexed from cs.AI updates on arXiv.org
Read the original article: