This is a good problem. I do not think it should be used as a reference in scientific papers (at least the way it works now), but there should still be a way to at least let people know where you got that information from.

This is more useful with things like Bing Chat that uses external sources to craft the answers, so you have “real” citations. But yeah, ChatGPT is currently impossible to verify.

I wonder if a seed could be implemented in the future to re-generate the same answers.

At the very least when it comes to analyzing the LLMs, e.g. to discuss biases, there should be some common understanding of how to reference the things that are discussed.

But this is a poor attempt. They just point to the general ChatGPT URL and the date, ignoring that at any time multiple models are available.