Defence against the dark arts of distillation: Unsupervised discovery of subliminal learning
arXiv:2610.13536v1 Announce Type: new Abstract: When a language model is distilled from a teacher that shares its base parameters, the student can inherit behavioural traits of the teacher---including misalignment---even from data that is semantically unrelated to the trait (Cloud et al., 2026). Al…
Read the full story at arXiv stat.ML ↗
Timeline · 1 report
- 2026-10-12 04:00 · arXiv stat.ML
Defence against the dark arts of distillation: Unsupervised discovery of subliminal learning