If I loathe LLMs for no other reason, the fact that so much energy needs to go into getting people to understand that they're experiencing confirmation bias, and it's just making up stuff, we can't base core features on that, is enough. But mad props to Perplexity(dot)AI for returning different info from their API than from their front page, and giving me both dramatically wrong summarization and lots of bogus URLs, to help me make that case just a leetle bit faster.