arXivShayan Talaei, Abhinav Chinta, Devvrit Khatri, Amin Karbasi, Azalia Mirhoseini, Amin SaberiWed, Jul 1, 2026, 10:46 AM PDT
score 17.1
New method exposes hidden biases in language models by amplifying them
Original: Distill to Detect: Exposing Stealth Biases in LLMs through Cartridge Distillation
Source: arxiv.org ↗
Writing ELI5 summary…