For the record, this (asking the model if the output is theirs) is not a reliable way to determine AI authorship of comments. The false positive rate is quite high, and will return true simply for a comment being well written and lacking idiosyncrasies.

I don’t think ChatGPT can tell if something was written by ChatGPT. At least not reliably.

OpenAI disagrees :p https://openai.com/blog/new-ai-classifier-for-indicating-ai-...

ChatGPT is probably not trained for this. The classifier seems like a different model/product. Can access it in the playground here:

https://platform.openai.com/ai-text-classifier