logoalt Hacker News

handbanana_ • yesterday at 9:00 PM • 1 reply • view on HN

Not the same. The LLM--assuming it's using diffusion, has the exact image of the original in its training data. There is really no human equivalent to that advantage.


Replies

verdverm • yesterday at 10:50 PM

I do not believe they have exact copies, but enough ingrained to produce very similar images. I would be very surprised if they were byte identical, with a margin of error, to the original.

I suspect this applies more to audio/visual than it does text or code.