"See this response by ChatGPT"

I feel strongly that links to AI-generated content are inappropriate on MO. This goes for comments, answers, questions, etc. I don't think there is any chance that someone asking a question here will be unaware that things like chatGPT exist given that they are being hyped endlessly in the popular press. If they want that kind of thing, they can get it elsewhere.

EDIT: We're now getting more and more obviously chatbot-generated content (some disclosed, some not), and increasingly some of it is not being deleted. I'm getting very close to deleting my account in disgust.

While I agree with Andy Putman that the comment in question was not a good one, I do not agree with the blanket condemnation in his answer that "links to AI-generated content are inappropriate on MO."

Part of my objection is my pet peeve that it's not clear what constitutes "AI." I think I know what Andy Putman's intent is, but what about the output of a SAT solver or a computer algebra system? What about a formally correct proof in Lean that was generated wholly or in part by an LLM? More to the point, I foresee that the line between human-generated and AI-generated content is going to become increasingly blurry. If I use something like Gemini DeepThink to help me prove a theorem, and I personally verify that the argument is correct, then what am I supposed to do under a ban on AI-generated content? Not share the proof at all? Share it, but pretend that I didn't use an AI to help me? Laboriously rewrite and re-type the proof "in my own words" so as not to run afoul of the ban? I'm in favor of establishing some guidelines, but IMO, a glib ban on "all AI-generated content" would be a poorly thought-out move that fails to recognize the rapidly changing landscape of AI-assisted mathematical proof.

But back to the specific comment in question: My objection to it was that it amounted to plugging the question into ChatGPT and regurgitating the output with no attempt to verify its correctness. Under no reasonable notion of "best practices" can such an action be considered best practices. The comment was nearly equivalent to "Have you tried Googling?" Yes, it's true that some people don't bother to use Google even when it's the obvious first step, but that does not mean that every MO question should be greeted with a comment "Have you tried Googling?" The lack of value was particularly obvious in this case because Iosif Pinelis later admitted that he had no idea whether ChatGPT's answer was correct and he lacked the expertise to check correctness. Simply posting a link to ChatGPT's output when you have no idea whether the output is correct is potentially worse than useless, since it could be actively misleading.

That comment is not asking the poster for more information or suggesting an improvement. It should have been an answer.

If that comment were an answer, it would have been downvoted into oblivion, and we would have nothing to discuss here.

The MO custom of tolerating and encouraging answers in comments is causing these problems.

I wanted to write something similar to Timothy Chow's answer, but was a bit hesitant to do so under my name, and I think his answer is better put than I could have managed. I would only add the following.

i. A comment on the original post suggested that linking to ChatGPT would be rude. But I think that rudeness depends on community norms, and sometimes it can be good for the purpose of a community to have norms that might be considered rude to those unfamiliar with the community. I have in mind that on MO sometimes questions are closed with little explanation or at most brief comments that the posted question isn't "research level." In other cases mathematicians refuse to provide additional details for an answer they consider adequate (with the implication being that it's not worth the mathematician's time to spell out every detail of the proof). I can understand that the posters whose questions are closed with little explanation or who are refused additional details on a proof might feel rudely treated, but the reality is that MO has a specific purpose and norms built around the fact that mathematicians are busy (they don't have infinite time). For much the same reason, I don't think it's bad for a mathematician to say "I asked ChatGPT, I checked the response, it is correct, here it is." In cases where the response is adequate, why should they be required to meaninglessly spend time copy-pasting and rephrasing the answer? By the way, I think insofar as they actually checked the response, they are providing additional value over ChatGPT through verification and curation.

ii. If there are some questions on MO that can now be answered adequately by AI, then as I see it the options are

1. Determine that such questions are not suitable for MO (but note that this requires MO standards to change as AI models advance).
2. Allow answers generated by AI.
3. Accept that people will provide answers generated by AI without disclosing that they were written by AI.

I think 1. and 2. are OK, but I'm less excited about 3.

Judging by the discussion, there are different types of MO users. Some don't have access to state-of-the-art AI systems, but would appreciate the high-quality answers they generate. Others don't want to see any AI-generated content.

In the Puzzling Stack Exchange, it's popular to hide portions of text (in this use case, "spoilers") that only become visible when a user clicks on it. For example:

https://chatgpt.com/share/69b087a2-9030-800c-8da5-c4b550307e49

Perhaps AI-generated content on MO could be hidden in this way?

I still think that these questions can be answered by ChatGPT and/or a comparable AI.

Of course, I can be mistaken about this. There is a simple test: Anyone can ask the linked questions of such an AI. If then there are any concerns about the AI's response (say, about specifics), then such concerns can be conveyed to the AI. I expect that after 3--5 such rounds of questioning of the AI, in most cases in a few minutes one would get satisfactory responses.

It would be interesting to see results of such tests.