Of course you can ask an LLM if the content is AI-generated. It's easily at least 50% accurate too.

You know, I wouldn't be surprised if the AI was less than 50% accurate. I'm not claiming that in general, but I'm also certain it would be possible to construct a dataset such that the AI would do far worse than a coin flip.

You know that it's not possible to do worse than a coin flip, right? If you're getting it 100% wrong, I'll just do the opposite of what you say, and have a 100% correct predictor.

The threshold isn't 50% because the distribution of human and AI written cases isn't naturally 50-50. So a coin flip will underperform always guessing the more frequent class. Where it gets interesting is if the base is unknown or variable over time or between application domains. Like, since AI written text is being generated faster than the human kind, soon guessing AI every time will be 99% accurate. That doesn't mean such a detector is useful.

When we say "coin flip" in these situations we mean "chance", ie the prior distribution. Otherwise a predictor of the winning lottery numbers that's "no better than a coin flip" would mean it wins the jackpot half the time.

Yup! My point is that the 'coin flip baseline' model that's as good as chance isn't actually trivial to create, for an unbalanced and time varying underlying distribution.

Only if you have that data available to you, the brain to analyze it and the freedom to chose.

Given the context of what I personally think about the content, the AI is much more accurate than that.

(Is it any wonder why people fall in love with bots that always agree with them?)

You can ask, but you can't really trust the results as the differences between AI generated text and human text are getting smaller every day AND LLMs aren't typically created to specifically identify the difference.

LLMs aren't intelligent. They're just very fancy math and a lot of smoke and mirrors.

If an LLM were capable of reliably detecting AI-generated content, then it would also be capable of producing content that does not appear to be AI generated.

So either A) LLMs are intentionally, universally making their content look AI-generated despite having the capability to do otherwise or B) LLMs cannot reliably detect AI output and their responses on the subject are complete BS.

> If an LLM were capable of reliably detecting AI-generated content, then it would also be capable of producing content that does not appear to be AI generated.

Those are two very different classes of problems. You do not automatically get one if you have the other, in any resource constrained situation. Yes, you can infinitely iterate a RNG into a content producer given a classifier, but that presumes infinite resources.