A recent study published by the research team at the Chinese University of Hong Kong has indicated that encryption methods such as Morse and Caesar ciphers can bypass the protective mechanisms of large language models, allowing inquiries into illegal content to GPT-4 and receiving responses. This research demonstrates the powerful capabilities of GPT-4 and evaluates the defense capabilities of large language models against encrypted passwords through the open-source tool CipherChat.
Danger! Using Morse, Caesar, and other encryption codes to inquire about illegal content on ChatGPT
AIGC 开放社区
52
© Copyright AIbase Base 2024, Click to View Source - https://www.aibase.com/news/667