Hacker Newsnew | past | comments | ask | show | jobs | submitlogin
LLMs can hide arbitrary undetectable information in their responses (twitter.com/zamirphd)
3 points by cryptohell on Jan 22, 2024 | hide | past | favorite | 3 comments



ChatGPT could encode usernames, timestamps and other session context into its responses in a way that would only be retrievable by OpenAI and provably invisible to everyone else


Disingenuous articles are the anathema of the common man. multiple valuable seconds of precious days are lost to such thoughts. Understanding how information can be stored multidimensionally is just a convoluted way of expressing what is already known. information can be encoded and decoded, transformed, transmogrified, obfuscated, masked… Humans, for example, have employed numerous techniques for smuggling or concealing messages in what might be considered otherwise completely innocuous objects.




Guidelines | FAQ | Lists | API | Security | Legal | Apply to YC | Contact

Search: