It comes as the AI firm's boss has called for the technology's development to be slowed down, citing "serious" risks.

I’m not an AI techie and I’m not well educated on LLMs/AI in general. The issue I’m having is the dichotomy between how people talk about it. I’ve heard completely opposite ends of the spectrum. From AI is quickly becoming advanced enough to threaten human society, all the way to its a glorified dictionary/search engine that can’t really do much of anything. I know Lemmy has a steep anti-AI slant. But, can someone talk about this in a way I can wrap my head around why the divide seems so big?

  • zarkanian@sh.itjust.worksEnglish
    31·
    2 days ago

    LLMs basically work like auto-correct. They’re much more sophisticated, but it’s the same basic principle. Instead of “Which letter is most likely to follow after this series of letters?” it’s “Which token is most likely to follow after this series of tokens?” A token can be anything, but usually it’s a word.

    Now, given that, consider a statement like

    We really do earnestly believe AI could kill all humans!

    Okay, how? A statistical model generating tokens doesn’t have thought, it doesn’t have emotions, so it can’t want to do something, much less make plans and carry them out.

    Could AI give somebody the ability to kill all humans? How? Under what scenario? Is somebody putting an LLM in charge of their nuclear stockpile? You don’t blame the LLM. You blame the humans who made that decision.

    • humanspiral@lemmy.caEnglish
      2·
      1 day ago

      so sad to keep seeing this criticism. These are tools that work towards goals. Can be war target selections based on “cruely is the point” philosophy, where there are no wrong answers from the point of view of the human agent asking the compute agent. A government spending $Ts to maximize their power/survival omelette is going to break a few human rights principles, and the models they will access will have no guardrails. A bond villain selecting people opposed to skynet/bond villains for “targets to neutralize” can use models that are months ahead of public releases, without any guardrails, on how to most discretely neutralize them. Anthropic has Swatted anti-AI activists without providing evidence of anything they said as wrong.

      Your argument is “their shit is weak and buggy, so nothing can ever go wrong”. Tools to help produce evil shit, buggy or not, are going to be used to produce evil shit.