391
Thanks Google (lemmy.ml)
submitted 5 months ago by to c/funny@sh.itjust.works
50 comments fedilink hide all child comments
 
you are viewing a single comment's thread
view the rest of the comments
[–] 0 points 5 months ago (8 children)

You don't think this is faked, or the prompt heavily massaged to give this response? You really just jumping straight to "yes, this is obviously real and evidence for why AI is bad"?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 8 child comments
    •
  • [–] 6 points 5 months ago (1 child)

    When AI Overview was in beta in May 2023, it was really bad. Infamous example:

    People milked it for memes. But at some point, Google fixed it, so to keep farming engagement, people had to start making fake screenshots and pretended that they were real.

    And it's really easy to fake:

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 1 point 5 months ago (2 children)

    Have you ever interacted with AI?

    LLMs are not capable of generating factual information. That's not how the technology works. It's never been how the technology works. It objectively cannot ever work that way.

    It's just 'hallucinations.' It is glorified autocorrect. Yes, it is easy to believe a trained autocorrect generated this paragraph. It seems like each word/phrase token is the most likely thing after the last one, given the total information shown. It is exactly what I, or anyone that has been made to suffer AI usage, expects from AI.

    Google has become useless, and few search engines still use search term ranking thanks to AI fucking the brains of every single person in the tech industry.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] -1 points 5 months ago (1 child)

    Have you interacted with AI in the past two years?

    Of course LLMs are capable of generating factual information. An LLM wouldn't lie if you asked "is the earth round?". But an LLM won't generate factual information every time, because as much as the technology is getting hyped up, it's still quite limited and probably will be for the foreseeable future.

    But when the answer is obvious and well-known, they are good enough. Just pop OP's question into any chatbot. They all, including Google's AI overview, give "no" here.

    A state-of-the-art glorified autocorrect will not send "Absolutely!" as the first few tokens after "can cockroaches live in your penis". That's GPT-2 level stuff.

    I agree with your last point about Google Search's decline.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •