RE: Explorations in failure - trying to get a LLM to detect LLM generated content on HIVE
You are viewing a single comment's thread:
I read that LLMs are ridiculously prone to using the em dash (—) in their work. I wonder how much content of hive has that 😀
I find it easy to recognise and always call it out, it's ridiculous how many use the accusation of you or hive in general of being backward as a defense. Cocks be that they are
0
0
0.000
I used to use the em dash all the time - a favourite of mine! I've had to curb it's use lest people suspect I am not real.
Holy kak. I don't mind admitting that I genuinely didn't know what it was till I read about AI loving it! 🤣🤣
I didn't know that's what they were called either!!!!
They certainly are. They also like to say things like this - then say not only that, instead. Or something like that - using words like "Testament" and a whole bunch of other tells. AI also likes to work in triplets, but that's something that good speeches have always had, and guess - (oh fuck another em dash) it was probably trained on that sort of content.
It probably was. It's quite funny how it churns out such similar formatting which I suppose just exposes it's training at least for the moment!
I don't think it will be long before being able to tell will be harder and harder