3 ms·

A bit of a rambling essay without much depth, but it did make me wonder: if AI tools weren’t wrapped in a chatbot pretending to be a person, would all of this h

A bit of a rambling essay without much depth, but it did make me wonder: if AI tools weren’t wrapped in a chatbot pretending to be a person, would all of this hullabaloo about AI and human nature go away? That seems to be the root of the unease many people have with these tools: they have the veneer of a human chatting but obviously aren’t quite there. I tend to treat Ai tools as basically just a toolset with an annoying chat interface layered on top, which in my experience leads me to not feel any of the feelings described in the essay and elsewhere. It’s just a tool that makes certain outputs easier to generate, an idea calculator, if you will. As a result, I’m pretty excited about AI, purely because they are such powerful creative tools - and I’m not fooled into thinking this is some sort of human replacement.

Avicebron 1y ago
I wonder if investors would have dumped the same amount of money in if it was pitched as something like "Semantic encoding and retrieval of data in latent space" vs "hey ex-machina though"
- keiferski 1y ago
  Definitely, whether they intrinsically believe it or not, the hunt for AGI is driving a lot of funding rounds.
  - goatlover 1y ago
    They think it will eliminate most payroll costs while driving productivity way up, without destroying the economy or causing a revolution. They also don't take worries about misaligned AGI very seriously.
    - FridgeSeal 1y ago
      I have to wonder if they’ve really thought through the consequences of their own intentions. > eliminate payroll costs And what do they think everyone who’s been shunted out of a livelihood is going to do? Roll over and die? I don’t see “providing large scale social support” being present in any VC pitch deck. How do they imagine that “without destroying the economy” will happen in this scenario?
ambicapter 1y ago
> if AI tools weren’t wrapped in a chatbot pretending to be a person I don't think this is the central issue, considering all the generative AI tools that generates art pieces, including various takes on the cherished styles of still-living artists.
- keiferski 1y ago
  Right, but then the conversation would mostly be akin to photography replacing portraiture - a huge technological change, but not one that makes people question their humanity.
  - numpad0 1y ago
    It's remarks like these that strengthen my suspicion, that continuous exposure to AI output causes model collapse in humans too.
gavmor 1y ago
Yes, it would—the dialogic interface is an anchor weighing us down! Yes, yes, it's an accessible demonstration of the technology's mind-blowing flexibility, but all this "I, you, me, I'm" nonsense clutters the context window and warps the ontology in way that introduces a major epistomological "optical illusion" that exploits (inadvertently?) a pretty fundamental aspect of human cognition—namely our inestimably powerful faculty for "theory of mind." Install the industrial wordsmithing assembly line behind a brutalist facade—any non-skeumorphic GUI from the past 20 years aughta do. Check out eg https://n8n.io https://n8n.io for a quick way to plug one-shot inference into an ad-hoc pipeline.
- niemandhier 1y ago
  I fully agree, where do we get the training data to create a base model? Where do we source the terabyte of coherent text that is devoid of ego?
numpad0 1y ago
it's just not there. Whether it's text, chat, images, video - the quality sometimes appear regular to cursory look, but aren't actually up there with that of humans. That's the problem. --- This gets into cheap scifi territory but: I think someone should put people in fMRI and make them watch live archive recordings of something consistent and devoid of meanings, like a random twitch channels on same theme, or foreign language sports, and sort results by date relative to time of experiment. Expected result would be that the data is absolutely random, maybe except for something linked to seasons or weather, and/or microscopic trend shift in human speech, and/or narrator skill developments. But what if it showed something else?

Avicebron 1y ago

I wonder if investors would have dumped the same amount of money in if it was pitched as something like "Semantic encoding and retrieval of data in latent space" vs "hey ex-machina though"

keiferski 1y ago
Definitely, whether they intrinsically believe it or not, the hunt for AGI is driving a lot of funding rounds.
- goatlover 1y ago
  They think it will eliminate most payroll costs while driving productivity way up, without destroying the economy or causing a revolution. They also don't take worries about misaligned AGI very seriously.
  - FridgeSeal 1y ago
    I have to wonder if they’ve really thought through the consequences of their own intentions. > eliminate payroll costs And what do they think everyone who’s been shunted out of a livelihood is going to do? Roll over and die? I don’t see “providing large scale social support” being present in any VC pitch deck. How do they imagine that “without destroying the economy” will happen in this scenario?

ambicapter 1y ago

> if AI tools weren’t wrapped in a chatbot pretending to be a person I don't think this is the central issue, considering all the generative AI tools that generates art pieces, including various takes on the cherished styles of still-living artists.

keiferski 1y ago
Right, but then the conversation would mostly be akin to photography replacing portraiture - a huge technological change, but not one that makes people question their humanity.
- numpad0 1y ago
  It's remarks like these that strengthen my suspicion, that continuous exposure to AI output causes model collapse in humans too.

gavmor 1y ago

Yes, it would—the dialogic interface is an anchor weighing us down! Yes, yes, it's an accessible demonstration of the technology's mind-blowing flexibility, but all this "I, you, me, I'm" nonsense clutters the context window and warps the ontology in way that introduces a major epistomological "optical illusion" that exploits (inadvertently?) a pretty fundamental aspect of human cognition—namely our inestimably powerful faculty for "theory of mind." Install the industrial wordsmithing assembly line behind a brutalist facade—any non-skeumorphic GUI from the past 20 years aughta do. Check out eg https://n8n.io https://n8n.io for a quick way to plug one-shot inference into an ad-hoc pipeline.

niemandhier 1y ago
I fully agree, where do we get the training data to create a base model? Where do we source the terabyte of coherent text that is devoid of ego?

numpad0 1y ago

it's just not there. Whether it's text, chat, images, video - the quality sometimes appear regular to cursory look, but aren't actually up there with that of humans. That's the problem. --- This gets into cheap scifi territory but: I think someone should put people in fMRI and make them watch live archive recordings of something consistent and devoid of meanings, like a random twitch channels on same theme, or foreign language sports, and sort results by date relative to time of experiment. Expected result would be that the data is absolutely random, maybe except for something linked to seasons or weather, and/or microscopic trend shift in human speech, and/or narrator skill developments. But what if it showed something else?