Exactly. I'd rather have a quick "I don't know" than wait for them to generate a reply that is likely to be wrong because I already asked the same models myself before consulting a human. If they generate a reply, I get to tell them exactly how the response is wrong, since the core content is usually the same. Only then do I get the "I don't know."

There are a few people I can tell are better at prompting than I am. Them, I'll tolerate this from, though most of them do more than function as an AI proxy anyhow and usually will say "I don't know, I'll check X" first.