@alice@lgbtqia.space I'd also like to say that some of these models -- many problems though they might have -- are a lot better at not spewing un-cited bullshit when doing web search and summary. Google seems to not really care about accuracy at all, which is why they don't mind using these shitty models that are much more error prone than (at least in my experience) other options like, say, GPT 5.6. Not an endorsement of AI systems, and even the best ones will still spew utterly false bullshit, but the least that a search company could do is at least use something that returns these kinds of dangerous mistakes *less often*, no?