3 ms·

> Large language models are both powerful tools, and mindless—even dangerous—bullshit machines. Yes, but so is the average person. Do you cover the similaritie

> Large language models are both powerful tools, and mindless—even dangerous—bullshit machines. Yes, but so is the average person. Do you cover the similarities/parallels to common human behavior patterns in your course? It stands out to me as a major blind spot in a lot of the discourse about LLMs, down to willing acceptance of humans intuiting what someone else meant when the words themselves were fundamentally ambiguous, which is extremely akin to a hallucination chosen from the listener's language model.

justonceokay 2y ago
Yes but people have culpability and responsibility. In places of power or influence, this culpability can lead to being fired, legal action, disbarment, loss of money, etc. so there is a real pressure to be coherent and aligned with reality.
- BugsJustFindMe 2y ago
  You say this but I have a ton of experience that "can lead" and "there is" come with a gigantic pile of caveats to the extent that they appear to be more false than true. Or they're technically true but practically meaningless. The world has been utterly awash in mass-perpetuated misinformation for at least all of recorded history without any real ability to stem the onslaught. This is not a modern problem just because a modern technology also exhibits it. You should look up the percentage of Americans who believe in ghosts sometime. About as many people believe in ghosts as don't, so no matter which side you land on, the other side is enormous. One of the sides must be wrong, I won't claim which, though only the belief side fails a falsifiability check. Where's our accountability to believing and spreading ideas based in reality again? The believers believe because they learned about it from someone. It didn't happen spontaneously on its own. It's all just been memes the entire time.
  - evklein 2y ago
    This is a sidestep imo. People _can_ be held accountable, though they will not always be. Machines add a layer of complexity - money is lost or a life is lost because AI made the call, who bears the burden? Machines _can't_ be held accountable.
    - BugsJustFindMe 2y ago
      I hear you, but I think it becomes less of a sidestep when "they will not always be" is in practice "they basically never are". And I'm not sure that most interactions even _can_ be held to account. When someone, say, hallucinates the intended meaning of something written ambiguously that has no unambiguous meaning, and I point out the hallucinatedness of their assumed meaning when what was written was definitively ambiguous, we've all just said words. There's no, like, penalty for anyone. And people do say things ambiguously and other people do hallucinate the supposed meaning literally all the time. If there were any meaningful accountability for hallucinations, it wouldn't happen nearly as often as it does.
_dark_matter_ 2y ago
The key distinctions are: we can hold people accountable, and the amount of shit produced by one person is limited. Neither of those are true for LLMs.
- BugsJustFindMe 2y ago
  You wrote basically the same thing as the adjacent person, so rather than write the same response to you both, I'll redirect to my reply over there: https://news.ycombinator.com/item?id=42993981 https://news.ycombinator.com/item?id=42993981
- 1propionyl 2y ago
  The recent book Unaccountability Machine (Dan Davies) made the rounds a while back on this site a few years back. A few years before that, Ted Chiang's essay in the New Yorker entitled "Will AI Become the New McKinsey?" did likewise. There's a bright red through line here. I get the sense that the intellectual ferment is starting to develop an awareness of the risk (and if you're cynical, potential) of LLM deployments in business as a systematic strategy for absorbing accountability for decisions. For the time being, it usually seems like there's someone who is accountable, or at least can be scapegoated. But how long will that last? As Davies points out in his book, we didn't need LLMs to create bureaucracies where the buck fails to stop anywhere and instead irretrievably slides between the tracks. As Chiang points out, the "efficiency maximization" of McKinsey served as a way for organizations to outsource accountability for major decisions to an entity very, very good at working backwards from desired outcomes while acting like they were just led to a fait accompli by the numbers. [0] https://libro.fm/audiobooks/9781805220794-the-unaccountability-machine https://libro.fm/audiobooks/9781805220794-the-unaccountabili... [1] https://www.newyorker.com/science/annals-of-artificial-intelligence/will-ai-become-the-new-mckinsey https://www.newyorker.com/science/annals-of-artificial-intel...

justonceokay 2y ago

Yes but people have culpability and responsibility. In places of power or influence, this culpability can lead to being fired, legal action, disbarment, loss of money, etc. so there is a real pressure to be coherent and aligned with reality.

BugsJustFindMe 2y ago
You say this but I have a ton of experience that "can lead" and "there is" come with a gigantic pile of caveats to the extent that they appear to be more false than true. Or they're technically true but practically meaningless. The world has been utterly awash in mass-perpetuated misinformation for at least all of recorded history without any real ability to stem the onslaught. This is not a modern problem just because a modern technology also exhibits it. You should look up the percentage of Americans who believe in ghosts sometime. About as many people believe in ghosts as don't, so no matter which side you land on, the other side is enormous. One of the sides must be wrong, I won't claim which, though only the belief side fails a falsifiability check. Where's our accountability to believing and spreading ideas based in reality again? The believers believe because they learned about it from someone. It didn't happen spontaneously on its own. It's all just been memes the entire time.
- evklein 2y ago
  This is a sidestep imo. People _can_ be held accountable, though they will not always be. Machines add a layer of complexity - money is lost or a life is lost because AI made the call, who bears the burden? Machines _can't_ be held accountable.
  - BugsJustFindMe 2y ago
    I hear you, but I think it becomes less of a sidestep when "they will not always be" is in practice "they basically never are". And I'm not sure that most interactions even _can_ be held to account. When someone, say, hallucinates the intended meaning of something written ambiguously that has no unambiguous meaning, and I point out the hallucinatedness of their assumed meaning when what was written was definitively ambiguous, we've all just said words. There's no, like, penalty for anyone. And people do say things ambiguously and other people do hallucinate the supposed meaning literally all the time. If there were any meaningful accountability for hallucinations, it wouldn't happen nearly as often as it does.

_dark_matter_ 2y ago

The key distinctions are: we can hold people accountable, and the amount of shit produced by one person is limited. Neither of those are true for LLMs.

BugsJustFindMe 2y ago
You wrote basically the same thing as the adjacent person, so rather than write the same response to you both, I'll redirect to my reply over there: https://news.ycombinator.com/item?id=42993981 https://news.ycombinator.com/item?id=42993981
1propionyl 2y ago
The recent book Unaccountability Machine (Dan Davies) made the rounds a while back on this site a few years back. A few years before that, Ted Chiang's essay in the New Yorker entitled "Will AI Become the New McKinsey?" did likewise. There's a bright red through line here. I get the sense that the intellectual ferment is starting to develop an awareness of the risk (and if you're cynical, potential) of LLM deployments in business as a systematic strategy for absorbing accountability for decisions. For the time being, it usually seems like there's someone who is accountable, or at least can be scapegoated. But how long will that last? As Davies points out in his book, we didn't need LLMs to create bureaucracies where the buck fails to stop anywhere and instead irretrievably slides between the tracks. As Chiang points out, the "efficiency maximization" of McKinsey served as a way for organizations to outsource accountability for major decisions to an entity very, very good at working backwards from desired outcomes while acting like they were just led to a fait accompli by the numbers. [0] https://libro.fm/audiobooks/9781805220794-the-unaccountability-machine https://libro.fm/audiobooks/9781805220794-the-unaccountabili... [1] https://www.newyorker.com/science/annals-of-artificial-intelligence/will-ai-become-the-new-mckinsey https://www.newyorker.com/science/annals-of-artificial-intel...