|
|
|
|
|
|
by lelandfe
1001 days ago
|
|
|
> All adversarial stylometry shares the core idea of faithfully paraphrasing the source text so that the meaning is unchanged but the stylistic signals are obscured Having an LLM rewrite a comment would do this entirely, no? Are you just interested from an academic perspective how one might build something more surgical, that only changes some words in a comment? |
|
|
> Are you just interested from an academic perspective how one might build something more surgical, that only changes some words in a comment?
Yes and no. ChatGPT seems like a blunt instrument, and given it's not purpose built, it could miss certain characteristics that could enable identification.
Also LLMs kind of have their own style, and adopting that particular style is likely self-defeating to getting a message out (e.g. I would tend to ignore something that sounded like it was written by ChatGPT). Manual correction then be needed, but it would be hard with such a system, because I think that would tend to re-introduce the author's style.