While I’m not the person you replied to and don’t know what their argument would be, I’ll take a shot at giving my own answer. In many cases when people post examples of AI giving unhelpful or bad information, there’s often someone who runs off to their favorite LLM to see if it gives a better result, and it usually does, so it gets treated like user error for using the wrong LLM or not wording the prompt properly. When in other examples that person’s favorite LLM which gave the correct answer this time, is the bad example hallucinating or mixing unrelated concepts, and other people are in the comments promoting other LLMs that gave them a good reply this time. None of the LLMs are actually trustworthy consistently enough to be trusted alone, and you won’t really know what answer is trustworthy unless you ask several LLMs and then go research the topic on your own anyway to figure out which answer is the most correct. It’s a valid point that ChatGPT got the answer more right than Gemini this time, but it’s somewhat useless to know that because other times ChatGPT is the one hallucinating wildly, and Gemini has the right answer, but since they’ve all been wrong before who do you trust.

LLMs are like asking an arrogant person who thinks they know everything, who rather than admitting what they don’t know, will pull an answer out of their butt, and while it might be a logical answer, it isn’t based in reality, and may still be wildly wrong. If you already mostly know the answer, maybe asking the arrogant person works, because you already know enough to know if they are speaking from their actual knowledge or making up an answer, but if you don’t already have knowledge on a topic, you won’t know whether the arrogant person is giving useful information or not.