登入
洞見 - Large language model security
Increased Vulnerabilities of Large Language Models Due to Fine-Tuning and Quantization
Fine-tuning and quantization of large language models (LLMs) can significantly reduce their jailbreak resistance, leading to increased vulnerabilities.
Enhancing Large Language Model Defense Capabilities through a Multi-Agent Attacker-Disguiser Game
A multi-agent attacker-disguiser game framework is proposed to strengthen the ability of large language models to generate secure responses that disguise defensive intent, avoiding exploitation by malicious attackers.
Robust Semantics-based Watermarking for Large Language Models against Paraphrasing
A semantics-based watermarking framework, SemaMark, is proposed to enhance the robustness of LLM-generated text detection against paraphrasing.
