D2D: Using Distillation to Amplify and Detect Hidden Model Biases
aminkarbasi · x · 2026-07-06
The paper introduces "Distill to Detect" (D2D), repurposing distillation as a detection mechanism. It distills the distributional differences between a suspect model and its unmodified base model into a small prefix adapter. This bottleneck amplifies and exposes "stealth biases" invisible in normal outputs, making originally hidden preference signals observable in generated text.
More from Research
- HarnessMM cuts agent tokens per task by up to 88% while raising success rates — omarsar0 · 2026-10-26
- Ben Recht: The world needs a truly open language model, and it's easier to build than you think — beenwrekt · 2026-10-26
- New Blog Breaks Black-Box Distillation Into Five Practical Steps — chrisrohlf · 2026-10-26
- DetectorScope turns black-box AI text detectors into auditable ones with sparse autoencoders — _akhaliq · 2026-10-26
- OPST paper shows LLMs self-improve with zero human data, guided only by topic descriptions — _akhaliq · 2026-10-26
- Tencent's HiSTEVE 1.0: hierarchical VLA agent plays Minecraft with native keyboard and mouse, hits 85% SOTA — _akhaliq · 2026-10-26