Can I read Prompt Inversion Attack Against Collaborative Inference of Large Language Models on EtoBox?

Prompt Inversion Attack Against Collaborative Inference of Large Language Models by 非凡 牛 is a document available to read on EtoBox.

What is Prompt Inversion Attack Against Collaborative Inference of Large Language Models about?

This paper introduces the prompt inversion attack (PIA), a method that exploits privacy vulnerabilities in collaborative inference of large language models (LLMs) by allowing malicious participants to recover input prompts from intermediate activations. The authors demonstrate that their PIA method significantly outperforms existing techniques, achieving high token accuracy in recovering prompts, thus highlighting substantial privacy risks in collaborative LLM systems. The study also discusses potential def

Author

非凡 牛

Language

EN