1 paper · 1 filter
Niloofar Mireshghallah, Tianshi Li
The discourse on privacy risks in Large Language Models (LLMs) has disproportionately focused on verbatim memorization of training data, while a constellation of more immediate and…