I’ve noticed that LLMs frequently make assumptions about why I am asking a question, and will tack on additional paragraphs at the end giving advice for things I never asked for.

Recently I asked in a Google search when YouTube encodes videos in AV1 vs VP9. And Gemini gave a very short and incomplete answer, and then moved on to assume I was asking because AV1 playback was draining my battery too quickly. And then gave me instructions to toggle some settings in YouTube that don’t actually exist.

Part of the problem I think is just the cheap and fast model they use in search results just isn’t very good.

But that bad assumption could make people explain themselves and provide more information than they previously would just to convince the AI to give them the answer. I don't like where this is all going, like you 'll only be able to get things done if you convince the Big Brother..