Study explores multilingual unlearning in LLMs

Arxiv CS.CL (NLP/language models)

Paper investigates how unlearning sensitive facts in one language transfers to other languages in LLMs, and whether the process is reversible. Findings highlight language-specific dynamics and potential risks of incomplete unlearning.

llmmultilingual

Arxiv CS.CL (NLP/language models)·Chaoyi XiangOlga OhrimenkoBenjamin I. P. RubinsteinLea Frermann

8 days ago