| title | Negative Self-Distillation: Learning to Reason by Avoiding Flaws |
|---|---|
| authors | Rongcan Pei, Zhepei Wei, Shuyao Xu, Xinyu Zhu, Wei-Lin Chen, Yu Meng |
| date | 2026-09-10 |
| venue | arXiv (preprint, not peer reviewed) |
| id | arXiv:2609.11699 |
| upvotes | 13 (Hugging Face Daily Papers, 2026-11-01) |
| links | abs pdf hf code |
| abstract | On-Policy Self-Distillation (OPSD) has emerged as a popular paradigm for large language model (LLM) self-improvement, allowing models to act as their own teachers by leveraging privileged information such as ground-truth solutions. However, recent findings indicate that OPSD can severely degrade the performance of LLMs on complex reasoning tasks: By forcing the student to imitate an artificially confident reasoning trace conditioned on privileged information, OPSD inadvertently suppresses expressions of uncertainty and penalizes the exploratory, self-corrective behaviors required to solve challenging problems. To address this, we introduce Negative Self-Distillation (NSD), a new framework that optimizes LLMs by diverging from flawed reasoning rather than imitating privileged solutions. Instead of relying on ground-truth answers or external supervision, NSD uses the model itself to generate a question-specific negative condition (eg, acting as a ``careless reasoner'') and pushes the student's distribution away from this self-generated negative teacher. Naively applying unlearning objectives to achieve this divergence is problematic, as flawed reasoning tokens are confounded with basic linguistic tokens; indiscriminately penalizing both risks catastrophically degrading the model's foundational language capabilities. We resolve this by designing a dynamic gating mechanism that automatically identifies and isolates reasoning-critical tokens, ensuring gradient updates target only behavioral flaws while preserving the model's linguistic priors. Empirically, NSD consistently outperforms OPSD and other label-free, self-bootstrapping reinforcement learning (RL) baselines. |
cite
@misc{pei2026negative,
title={Negative Self-Distillation: Learning to Reason by Avoiding Flaws},
author={Rongcan Pei and Zhepei Wei and Shuyao Xu and Xinyu Zhu and Wei-Lin Chen and Yu Meng},
year={2026},
eprint={2609.11699},
archivePrefix={arXiv},
url={https://arxiv.org/abs/2609.11699}
}