SearcharxivSearch

arXiv · 2608.27960

When Teacher Guidance Misleads: Reward-Aligned On-Policy Distillation

Abstract

On-policy distillation (OPD) has recently emerged as a popular post-training paradigm for large language models (LLMs), providing an efficient way to transfer the knowledge and capabilities of teacher models into student models. However, teacher guidance on student-generated prefixes is not always reliable. Training should optimize the model to generate responses that are more likely to be correct, or equivalently, to get higher outcome rewards. But during OPD, the teacher model may provide guidance that discourages the student from moving toward correct trajectories or moves the student toward incorrect ones, which is misaligned with outcome reward. Such misaligned guidance is unreliable, as it would mislead the optimization process and ultimately degrade model performance. To mitigate misaligned teacher guidance, we propose Reward-Aligned On-Policy Distillation (RA-OPD). The key insight is to keep only trajectories whose induced updates move the student toward correct trajectories or discourage the student from moving toward incorrect ones. Specifically, for each sampled trajectory, RA-OPD checks whether its trajectory-level distillation return is consistent with its outcome reward and then filters out the misaligned trajectories. RA-OPD selects more reliable trajectories to improve student model performance without requiring additional computational cost. We evaluate RA-OPD on math and code benchmarks using models from the Qwen3 family and the DeepSeek-R1 family. Across seven math benchmarks and three code benchmarks, RA-OPD significantly outperforms standard OPD and other tested OPD variants.

Explore related subjects

Keep this discovery

BibTeXRIS

Siyuan Gan, Yuhan Li, Xiran Wang, Linjian Meng, Boyan Wang, Zhen Zhao, Jing Huo, Yang Gao. 2026-08-28. When Teacher Guidance Misleads: Reward-Aligned On-Policy Distillation. https://arxiv.org/abs/2608.27960

Cite the original work for its findings. Save a collection to share your selection of sources.

Discover connections

Connections use source metadata and explicit phrase matches, not verified experimental comparisons.

KEEP EXPLORING

Related discoveries

One AI Signal, Many Human Judgments: A Bayesian Cascade Analysis of AI-based Credibility Indicators in Online Information Spread

Social media platforms increasingly use AI-based credibility indicators to help users judge misinformation. Unlike individual human-AI decision-making, these indicators are embedded in information spread: users see both an AI prediction and earlier judgments shaped by the same AI, and their own judgments may then enter the public history. Yet how to analytically characterize this process remains under-explored. We therefore introduce a social-learning lens for this setting by extending the classical Bayesian cascade model with the AI indicator as a shared public signal. The resulting Gateway condition compares the evidence from the AI prediction with users' private impressions. Through this view, we show that AI changes what public history means. Crowd agreement may reflect accumulated independent human evidence, or repeated dependence on the same AI prediction. This creates a preservation-correction trade-off: stronger reliance on AI can preserve correct predictions, but can also lock in incorrect ones by blocking corrective private impressions. We calibrate the model using human-subject data on news veracity judgments. Although the AI outperforms human users, the average user weights it below her own impression but above several peer judgments, while individual users vary from discounting the AI to relying on it enough to cascade. Simulations show that over-reliance on a weak AI is especially harmful, and that diversifying AI signals across users can better keep the crowd informative. We conclude with implications for understanding human-AI interaction in information spread and designing misinformation interventions.

cs.HC

"An Endless Stream of AI Slop": How Developers Discuss the Burden of AI-Assisted Software Development

"AI slop", that is, low-quality AI-generated content, is increasingly affecting software development, from generated code and pull requests to documentation and bug reports. However, there is limited empirical research on how developers perceive and respond to this phenomenon. We qualitatively analyzed how developers discuss AI slop in 1,154 Reddit and Hacker News posts, developing a codebook of 15 codes organized into three thematic clusters: Review Friction (how AI slop burdens reviewers, erodes trust, and prompts countermeasures), Quality Degradation (damage to codebases, knowledge resources, and developer competence), and Forces and Consequences (systemic incentives, mandated adoption, craft erosion, and workforce disruption). Our findings frame AI slop as a tragedy of the commons, where individual productivity gains externalize costs onto reviewers, maintainers, and the broader community. We report the concerns developers raise and the mitigation strategies they propose, with implications for tool developers, team leads, and educators.

cs.SE

AI Writers Have a Consistent Stylometric Footprint, but AI Editors Do Not

Text generated by large language models (LLMs) has been shown to be stylometrically distinct from human-written text \citep{andreDetectingAIAuthorship2023, shahDetectingUnmaskingAIGenerated2023, oparaStyloAIDistinguishingAIGenerated2024, soto2024fewshot, liLinguisticDifferencesAI2025, selviogluFeatureExtractionAnalysis2025}. But LLMs are increasingly used not only to generate text but also to edit human writing, and it is unclear whether the two leave the same trace. We show that AI generation leaves a consistent ``stylometric footprint'': a small subset of features, primarily entropy and lexical diversity, consistently separates AI-generated text from human writing across 8 LLMs and 5 domains, while the remaining features depend heavily on the domain and generator. AI editing, however, does not reproduce the same footprint. Relative to their human-written sources, AI-edited texts show only a small increase in lexical diversity and a decrease in entropy, rather than the joint increase that characterizes AI generation. Lexical density, which contributes little to generation, instead becomes the dominant editing-associated signal. Stylometric features therefore separate AI-edited text from AI-generated text but are substantially less effective at separating it from human-written text. Our results suggest that ``AI text'' is not a single phenomenon: generation and editing leave qualitatively different stylometric traces and should be studied separately.

cs.CL