Hacker News new | ask | show | jobs
by mh- 16 days ago
I believe we're in a fairly limited window of time right now where that's even detectable via tools like Pangram. For small models it'll, maybe not the case. But for anything worth paying for.
1 comments

They only have to slip up once. And they will, because LLM users are lazy, and LLM content naturally regresses towards the mean.