AI unlearning study finds update patterns matter more than size
A preprint study finds forget–retain update structure tracked language-model relearning better than weight distance, with utility and scope limits.
Developing Light · https://developinglight.com/editorial/developing-light