Yeah. That little bullshit disclaimer at the bottom of every AI thing we have to intentionally ignore is doing a lot of heavy lifting, but is probably a really flimsy shield: “AI can make mistakes, so double-check responses,” says Google.

I have actually used gemini intentionally to answer a question, but the question has always been in the form of ‘where do I find info about XYZ’ instead of my real ask, because I can quickly validate that the response to question 1 is a reliable source, where validating an AI hallucination might take multiple times longer to refute than it did to produce.