499
Google AI gets cagey when asked who won the 2020 elections: But the AI responses get more forthright when asked outside the US. (arstechnica.com)
submitted 1 week ago by to c/technology@lemmy.world
83 comments fedilink hide all child comments
you are viewing a single comment's thread
view the rest of the comments
[–] 38 points 1 week ago (11 children)

Just. Stop. Using. LLMs. They hallucinate lie unpredictably, and have biases put in by whomever controls them. Jeebus Cripes.

  • permalink
  • fedilink
  • source
  • hideshow 11 child comments
    •
  • [–] 11 points 1 week ago (10 children)

    I'd say they lie somewhat predictably, in proportion to their training data per topic. I've only asked questions that I know have an unpoliticized, specific answer, and even then I verify.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 10 child comments
    •
  • [–] 13 points 1 week ago (9 children)

    I dont understand what the point of going to the LLM is if you need to verify what it tells you?

    Its gonna take more time than it would to just go directly to the source, and youre wasting resources too.

    Kind of seems like there is absolutely no benefit whatsoever

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 9 child comments
    •
  • [–] 15 points 1 week ago (4 children)

    If you knew the source, you wouldn't need to go to the LLM. It's helpful when you don't know where to begin, like when you don't even know the name of the piece of your staircase that snapped off. You can just take a picture of the broken part and ask what is called and then verify that answer rather than read a bunch of blogspam from people trying to sell their home repair service.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 4 child comments
    •
  • [–] 3 points 1 week ago (3 children)

    It sounds like we’ve had very different experiences with both research and stairs.

    I meant vs a normal search engine tho (which is also a viable place to find something you dont know the word for).

    In your example tho, id probably just post a picture on a forum with text like “what is this piece?” Switch projects for 30 minutes, and then go back to my post to get opinions from humans.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 3 points 1 week ago (2 children)

    I don't know if it's a conspiracy theory or real, but I have a feeling that search engines' search quality has been deteriorating in the last decade, and it really sped up when LLMs shown up

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

    Thats not a conspiracy theory, that is 100% whats going on, and there is documentation to prove it.

    “Deteriorating” is the wrong word tho, that implies passivity, but their degradation was very much active and intentional.

    If you’re curious as to why this would happen, I recommend reading the book “enshitification” by Cory Doctorow, it contains a much more nuanced and detailed explanation of what’s happened to search engines (and our whole society really).

    The other reason that that book does not cover, is that the people who own these tech companies are fascists who want to control information/“truth”. So they made the best method of finding unbiased information more difficult, and simultaneously attempted to replace it with a chatbot that tells you whatever they programmed it to say “the truth” is.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 0 points 1 week ago (3 children)

    LLMs are immensely helpful for coding.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 3 child comments
    •
  • [–] 5 points 1 week ago (2 children)

    Ok, thats not the conversation I was having, but sure.

    They’re not helpful for learning to code tho, which could be a catastrophic problem in a generation or so.

    Why does everything need to be quick and easy tho?

    Do you not find joy in the critical thinking and problem solving elements of human coding?

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 2 child comments
    •
  • [–] 3 points 1 week ago (1 child)

    Good points. I just was pointing out that LLMs have some utility. I do wonder if the ease of coding will eventually be a problem. It may not matter if humans can no longer write code. But it could matter a lot if we can no longer read or understand the code LLMs write. I like the architecture building and problem solving of software engineering with the emphasis on engineering. The code writing, while fun at times, can also be tedious.

  • permalink
  • fedilink
  • source
  • parent
  • hideshow 1 child comment
    •
  • [–] 2 points 1 week ago

    The process is part of an art, even the parts that can feel tedious, some would argue the process is the entire point of an art.

    The ease of allowing the computer to do the coding for you will eventually lead to a society that can no longer read or understand code. Thats the exact problem I’m speaking of.

  • permalink
  • fedilink
  • source
  • parent