4 ms·
If LLMs can identify a person across websites, I can ask LLM to read up his posts and write like him impersonating him and then this feeds back into the tools i
by ghywertelling 8mo ago
If LLMs can identify a person across websites, I can ask LLM to read up his posts and write like him impersonating him and then this feeds back into the tools identifying him. I can probabilistically malign a person this way.
- john_strinlai 8mo agostylometry is only one aspect of de-anonymization. what you describe is certainly a threat that we will have to deal with, but there is a lot more to credible impersonation than just being able to mimic a writing style
- Jerrrrrrrry 8mo agoHow to conduct a psy-op https://youtu.be/YTGQXVmrc6g https://youtu.be/YTGQXVmrc6g
- JohnMakin 8mo agoThis already is a thing people did at least as far back as I started getting into web privacy, which was ~10 years ago. I have been the target of it before. LLM's are probably better at it, but I don't know if this is as destructive as people may guess it would be. Probably highly person dependent. The micro-signals this paper discusses are more difficult to fake.
- functionmouse 8mo agoSo this means deanonymization doesn't work? Rejoice?