Soft Token Attacks Cannot Reliably Audit Unlearning in Large Language Models.

Haokun Chen, Sebastian Szyller, Weilin Xu, Nageen Himayat

Browse the full EMNLP paper archive.