from Hacker News

Ask HN: Can ChatGPT Counteract Stylometry?

by desertraven on 3/28/23, 1:28 AM with 2 comments

There's been posts on here demonstrating the ability to identify different users as the same person based on writing style. Three letter agencies likely employ this internet-wide.

Can formulating discussion/comments through chatGPT counteract this practice? One seeming drawback is ChatGPT not communicating in a style that is pleasing to the author. Though ChatGPT is able to take on different writing styles as per instruction.

Any thoughts?

by xena on 3/28/23, 1:42 AM
If your threat model includes government agencies, using a third party like OpenAI that could be easily man in the middled by government agencies through purchasing OpenAI would seem counter to that goal.
It would be an interesting way to defeat stylometry techniques though.
by PaulHoule on 3/28/23, 1:38 AM
Language models ought to be able to do better than that. If they can translate from English to French they ought to be able to “rewrite the following in the style of Ben Thompson.”