Can Language Models be Instructed to Protect Personal Information?

Yang Chen, Ethan Mendes, Sauvik Das +2

Large multimodal language models have proven transformative in numerous applications. However, these models have been shown to memorize and leak pre-training data, raising serious…