i love how much the current wave of #AI fucks with the cultural tropes we had about computer intelligence. instead of cold, rigid, logical planners, they are moody lumps of intuitions, who respond differently when you ask nicely, have trouble staying on topic, and may just make up stuff that "sounds about right".
i think many false expectations are because we can't quite imagine computers doing things like that.
@gimulnautti well there are also fairly crucial differences here, like from what i understand LLMs currently don't really have any autonomy over their "experience". at least not on their own. they can't really experiment. which brought be to the odd conclusion that they do have networks of intuition that i would call understanding but not knowledge. their entire relation to reality is mediated by whoever trains them.
@sofia Indeed. Even if computationally we are based on the exact same process, the scale and efficiency differences are colossal. And will be for a while still too.
Consciousness isn’t needed for any of the tasks we want them to perform, not even for faking consciousness which is one of the tasks(Virtual companions).
But some theories argue that consciousness isn’t much of a trick either. It can be derived from a system predicting what it will predict next. Just multiply enough.
@gimulnautti yeah, i think conciousness is way overhyped. what you say sound pretty much what i said a bit ago: consciousness is the prediction of the present.
@gimulnautti though even already gets weird and blurry at places, like with synthetic data that they have some control over: like letting them program having it evaluated and letting them judge the result and them being trained on that. i think that is an agential experience in a way, though not exactly with reality, i guess.
@gimulnautti i'd be a bit more careful with "not much different from human brains". human learning certainly resembles backprop in some ways, and probably got some equivalent of attention mechanisms. but we also got so much other interesting architecture going on, and i think most working AI systems are only very loosely inspired by those.
@gimulnautti it's think that while there are certainly some emergent similarities we see, LLMs internally are probably much more alien from us than when we'd somehow somehow taught an octopus to speak or something like that.
but human-likeness isn't really the point of AI, i think, so it doesn't "speak against" LLMs…
i think the alienness is part of what makes it so exciting, it will be interesting to see where we see convergence and where we won't. :)
@sofia Yes it’s going to very exciting! As long as we remember to put the shields up against parties with ill intent exploiting the dumb-level AI’s to poison the well of democracy itself.
We get over that speed bump, should be smooth sailing.
@gimulnautti yeah, i think the big problem is that we still have such shitty social incentives that "demagogue", "liar" and "murderer" are basically still job descriptions…
i hope i get around to reading your text. btw: if you know how to get decent text-to-speech on linux i beg you to tell me 😅.
Add comment