Current “unlearning” methods only suppress capabilities instead of truly unlearning the capabilities. But if you distill an unlearned model into a randomly initialized model, the resulting network is actually robust to relearning. We show why this works, how well it works, and how to trade off compute for robustness.
Unlearn-and-Distill applies unlearning to a bad behavior and then distills the unlearned model into a new model. Distillation makes it way harder to retrain the new model to do the bad thing. Produced as part of the ML Alignment & Theory Scholars Program in the winter 2024–25 cohort of the shard theory stream.
Read our paper on ArXiv and enjoy an interactive demo.
Robust unlearning probably reduces AI risk Maybe some future AI has long-term goals and humanity is in its way. Maybe future open-weight AIs have tons of bioterror expertise. If a system has dangerous knowledge, that system becomes [...]
---
Outline:(01:01) Robust unlearning probably reduces AI risk
(02:42) Perfect data filtering is the current unlearning gold standard
(03:24) Oracle matching does not guarantee robust unlearning
(05:05) Distillation robustifies unlearning
(07:46) Trading unlearning robustness for compute
(09:49) UNDO is better than other unlearning methods
(11:19) Where this leaves us
(11:22) Limitations
(12:12) Insights and speculation
(15:00) Future directions
(15:35) Conclusion
(16:07) Acknowledgments
(16:50) Citation
The original text contained 2 footnotes which were omitted from this narration. ---
First published: June 13th, 2025
Source: https://www.lesswrong.com/posts/anX4QrNjhJqGFvrBr/distillation-robustifies-unlearning ---
Narrated by
TYPE III AUDIO.
---
Images from the article:
