3 ms·

I remember a bunch of years ago that there was software that could take a text and then determine if other texts were written by the same person, based on certa

by ksaj 4y ago

I remember a bunch of years ago that there was software that could take a text and then determine if other texts were written by the same person, based on certain statistical points regarding the writing style and language use. I would expect that one heuristic would be whether a text is completely lacking in such a signature. This might be affected by writing "in the style of" but that might be picked up by some of the plagiarism tools as well. Then follow that with "in the style of" the particular person submitting the work as their own. A lack of any statistical signature, and heuristically not matching the author apparents' statistical signature either, should set off flags as to the authenticity of the writing, AI or otherwise.

dragonwriter 4y ago
> I remember a bunch of years ago that there was software that could take a text and then determine if other texts were written by the same person, based on certain statistical points regarding the writing style and language use. There have been lots of stylometry tools released over the years, sure. > I would expect that one heuristic would be whether a text is completely lacking in such a signature. A text cannot be lacking such a signature, that makes no sense. > This might be affected by writing “in the style of” “Might be”? “In the style of…” functionality in LLMs is a direct outgrowth of stylometry. That’s where the measures which allow it to work come from. > Then follow that with “in the style of” the particular person submitting the work as their own. But to do that, you have to a verifiably original content written by the target author to start with.