Stanford Proposes D2D to Detect Hidden AI Bias

Stanford AI Lab introduced Distill to Detect (D2D), an auditing method that amplifies and detects hidden biases in fine-tuned LLMs. It distills differences into a smaller model to expose subtle biases.

2026-07-09 ~ 2026-07-10 · 2 related posts

• D2D Audits Biases in Fine-Tuned Models — StanfordAILab · 2026-07-09
• D2D: Amplifying Hidden Biases in Models — aminkarbasi · 2026-07-10