5 papers
More Convincing, Not More Correct: Self-Play Reward Hacking of Reference-Free LLM Judges
Chenyu Zhou
Training a language model against its own reference-free judgments (the premise of self-rewarding, self-play, and LLM-as-a-judge pipelines) assumes a model's verdict on a shown ans…
Certified Speculative Execution for Untrusted AI Agents
Chenyu Zhou, Qiliang Jiang, Shuning Wu +1
Hard-constrained sequential decision systems have no certified way to spend the test-time compute of modern AI: executing the multi-step drafts of a learned policy or a frozen LLM…
The Verifier is the Curriculum: Execution-Gated Self-Distillation for Cross-Family Game Generation
Chenyu Zhou, Qiliang Jiang, Shuning Wu +1
Post-training a code generator against a learned judge can optimize proxy features that raise the score without improving the artifact. We study the opposite signal: a deterministi…
Mechanism-Guided Selective Unlearning for RLVR-Induced Reasoning
Chenyu Zhou, Qiliang Jiang, Shuning Wu +1
We propose MAST (Mechanism-Aligned Selective Targeting), a mechanism-guided method for unlearning RLVR-induced reasoning with substantially lower collateral damage than standard fu…
The Vision Encoder as a Privacy Boundary: Visual-Token Side Channels in Encoder-Free Vision-Language Models
Chenyu Zhou, Qiliang Jiang, Shuning Wu +1
A vision encoder compresses image pixels into semantic embeddings, implicitly acting as a privacy boundary by preserving semantic content while attenuating pixel-local detail requi…