(Georgia Institute of Technology, 2024-05-05)
Kelklie, Moges
Data privacy has become a key concern of large language models(LLMs), both in the large trove of information they can infer and in the inherent inflexibility of models in forgetting the learned data. LLMs do not have an easy way to delete information; sensitive data could be inferred through prompt engineering, thus raising concerns about data privacy. This article attempts to address the challenge of LLM data privacy and how policy can help mitigate some of the privacy concerns.