Should we ask AI first?

AI is usually trained to take on an agreeable tone, after all, they are commercial products and it is usually bad business to annoy or confront the customer unnecessarily. Mathematicians on the other hand tend to speak (type) bluntly, without much regard for the other person's feelings, especially on the internet.

I would say that paid versions of the popular AI models are now competent enough to answer basic questions about basically everything in mathematics, but the boundary is blurred: AI can still make basic mistakes. My most recent encounter is ChatGPT telling me that $\left(\begin{smallmatrix} 1 & 0 & 0 \\ 0 & - 1 & 0 \\ 0 & 0 & - 1 \end{smallmatrix} \right) = - I_{3 \times 3}$ (this is not a typo), so you should always take what it says with a pinch of salt.

I don't think that "asking AI first" should be recommended as best practice or seen as due diligence.

A few reasons:

• LLMs are a recent tech, and they change very quickly in the way they operate and the things they allow for free, so it's a bit strange to organize a "best practice" around them at the moment;
• Some people may prefer not to use them because they raise largely spread concerns (I don't want to discuss here how reasonable) about environmental impact, addiction risk, social risks like concentration of power, privacy reasons, etc... Search engines (at least before inescapable "AI summary") proved on a longer time window that they are comparatively "safe", LLMs didn't yet;
• Some people may have gotten the impression that the interaction with LLMs on their questions is mostly a waste of time, so they don't want anymore to experience it. Say a chatbot says "that's solved by the well-known Ivanov's Theorem that goes bla bla bla" only to find later that the theorem was not by Ivanov, that it did not say exactly what the bot claimed, that the closest true theorem was not enough to settle the question... all of this through many rounds of interaction-search-check-interaction-search-check... in the end almost completely useless.
• Search engines already existed long since when MO was born; integrating a new a recent class of tools into MO's best practices is to let AI companies clearly dictate the agenda of the website, which is overall bad I think.

To me, this question doesn't seem materially different from, "Should I use a search engine to try to find the answer to my question before I post it to MathOverflow?" I think the standard reply is that one should exercise a modicum of due diligence before posting an MO question. Spending a few minutes using a conventional search engine surely falls under "due diligence," so by the same token, spending a few minutes using a chatbot would also seem to fall under "due diligence." Now if you have some moral objection to using AI, then of course I'm not saying you must use it. I'm also not saying that you should fork over $200/month and have a cutting-edge AI churn on your problem for hours and hours before you allow yourself to post to MO. But those extremes aside, I don't see any reason, for this purpose, to treat a chatbot differently from a search engine.

Having said that, I can see one possible distinction. It appears that AI has killed StackExchange, or at least has been a significant factor in its demise. Perhaps we are concerned that AI might similarly kill MathOverflow, and we want to take steps to prevent that from happening. I can sympathize with this viewpoint, but I don't see how refraining from consulting an AI before posting to MO will accomplish much. If an AI can quickly answer the question completely, then there's a good chance that you'll get a response of the form, "I asked ChatGPT and it said..." It seems to me that it would be better to pre-empt such responses by spending a few minutes with an AI yourself before posting.