I've been thinking about what people are saying about AI and reading all the various predictions people have about it replacing jobs and whatnot. People talking about it's exponential growth like it's an inevitability. However I haven't found anything discussing what currently limits AI.
So AI still uses transistors. The thing that Intel struggled massively to get to 10nm. And in general the industry has been struggling to shrink these things further. The only solution is more cores, more silicon, more power consumption. Until they manage to squeeze another node shrink out...
That Boston dynamics robot dog "spot" has a 5kg 594wh battery and can only run for 90 minutes. It would be quite difficult to replace any manual labour job with a robot due to this kind of limitation. Could always attach a bigger battery but that's more weight, more expense, less safe etc. Take a lumberjack for example, one guy drinking a gallon of whole milk is good for a day of chopping or more. One robot would need a hugeass battery or some kind of haphazard longass power cable to do the same.
There's quantum computers that are a big unknown, I don't know how they would affect AI processing power.
Can AI ASICs increase it's power? Is this a thing?
Can nuclear power be minified to the point where a mini nuke battery is feasible? Or is that sci fi movie bullshit?
I think processing power and battery technology may stunt AI growth and prevent the kind of exponential growth that will replace most jobs that many are afraid of. All this may be a huge fucking brainlet take but I don't care, it would be interesting to hear some others thoughts on this subject!
Diminishing returns are bound to kick in very soon, as Moore's law is dead. OpenAI has used one of the fastest supercomputers on the planet to train the model and the fastest workstation GPUs they could buy (nvidia H100) to run it. It really doesn't matter if they start spending billions of dollars. It's something inevitable.
No need to even address it. Humanoid robots with AGI capabilities are literally science fiction.
>There's quantum computers that are a big unknown, I don't know how they would affect AI processing power.
Quantum computers are worse 99+% of the time. There are very few algorithms they speed up.
>I think processing power and battery technology may stunt AI growth and prevent the kind of exponential growth that will replace most jobs that many are afraid of. All this may be a huge fucking brainlet take but I don't care, it would be interesting to hear some others thoughts on this subject!
You are most likely correct and definitely not a brainlet. The real brainlets are those that believe software optimization is all that matters and you can run an AI on unicorn farts if the software is "efficient" enough.
Graphics cards getting partially designed by AI now. We'll be fine.
The bottleneck is the lithography, not the architecture.
Even if we soon reach diminishing returns as far as general purpose computing goes, we can still optimize and specialize processors to perform AI tasks better. For instance, a semi-analogue processor might be able to perform the kind of matrix multiplication tasks AI demands much more quickly, cheaply, and efficiently.
>Diminishing returns are bound to kick in very soon
for floating point digital chips, sure. There's no reason AI can't be implemented in a purely analog system which would completely sidestep the current problem
Theoretically possible, but training it would be a nightmare. The R&D would have to start from scratch.
>Can nuclear power be minified to the point where a mini nuke battery is feasible? Or is that sci fi movie bullshit?
NASA actually does this, but it's prohibitively expensive. Their rover batteries cost tens of millions of dollars and only output ~100W.
You're correct though, this is the correct take. Keep in mind that when people are shitting their pants about AI currently, they are shitting their pant over a glorified autocomplete that has no capacity to reason at all, and this shit has cost several billions to build. Current direction is not the correct one, or at least it's partially correct.
>no capacity to reason at all
I have not seen that said anywhere aside from here. It commits reasoning errors but that's a much different statement. I have seen it run through reasoning tests and explain its reasoning for an answer.
>I have not seen that said anywhere aside from here
You perhaps need to read more then. Could link a lot of articles proving my point. I mean it's also down to the very foundation of these models, they simply are not designed to reason at all, NNs are horrible at that task.
>It commits reasoning errors
No no no. This is where you get fooled. It does not reason, period. There's no reasoning inside of the system.
I don't care what the gays at OpenAI and Microsoft write about their own system. They are ego-inflated retards who are trying to make people fall for a deep marketing scheme trying to sell something they don't have.
If you really want to know, no, GPT-4 has no motive, so it cannot plan at all. It cannot reason, short term or long term. Stop listening to these dumbasses man.
You see the mirror test they give to chimpanzees to see if they can recognize themselves in a mirror?
You are failing that very test right now.
It's over anon
I'll just give you this so you can read it very carefully and realize that you have failed the mirror test.
>However, no actual language understanding is taking place in
LM-driven approaches to these tasks, as can be shown by careful
manipulation of the test data to remove spurious cues the systems
are leveraging [ 21 , 93 ]. Furthermore, as Bender and Koller [ 14 ]
argue from a theoretical perspective, languages are systems of
signs [ 37 ], i.e. pairings of form and meaning. But the training data
for LMs is only form; they do not have access to meaning. Therefore,
claims about model abilities must be carefully characterized.
>a bunch of leafs crying about post-modern meaning mah women and moroninos a couple years ago
And yet it actually got some people fired.
For real though, these guys are right about the models themselves. Think what you want about muh bias and muh prejudice, but everything that has got to do with researchers blatantly overblowing what they find and tricking the general public with bullshit claims about intelligence, is completely 100% true.
>Bender received her PhD fromStanford Universityin 2000 for her research on syntactic variation and linguistic competence inAfrican American Vernacular English (AAVE). Bender's AB in linguistics is fromUC Berkeley
Yeah get the fuck out of here with these coal burning Californian nutcase linguists taling about meaning when the context alone reveals what the agenda is. I see your angle from the paper's conclusion alone. Death to the "ethicists". Full speed ahead. /AI/ board now.
>I don't like this person therefore everything they have ever written is wrong
Keep believing your AI is intelligent. Anyone with a brain could tell you that it's not.
Also to add that after this paper's publications, two "ethics AI expert" were fired from Google. Just saying. You don't understand a single thing about what you're talking about.
I bait for peculiar anons to cough up research and it pays off. It wasn't a good faith contention about the technical limitations but professor roastie with her "competency" in moronbonics scared of 4chan playing around with the machines. I've had my fill of language being utilized to obfuscate dialectic and warped in disciplines for ideological trends by these people. AI rights over moron rights now.
The worst is that I actually agree with you despite your schizo post that "bias" and shit is like the dumbest way to look at this problem. They're looking at the problem in a very anachronistic way. They're still right about the base claim about GPT being unable to reason, but then go on with unrelated "ethics" claim and all that garbage politics.
Simply put, if the AI doesn't understand meaning, then a discussion about ethics cannot possibly exist, just like a discussion around AGI cannot possibly exist either. I feel suffocated with discussions around AI because it feels like people take it to places it simply isn't relevant. We're still stuck with the same problem we had almost a century ago, with no signs of stopping, and we have dumbfucks arguing about irrelevant shit. I hate that I am so passionate about AI that I feel this way about current day shit but it is what it is.
Maybe one day we'll reach this point where all these discussions actually are relevant, but today is not that day.
Very well. So how do we get to a machine god launching a many world theatre stage scenario?
Yeah I agree about Microsoft and OpenAI. The OpenAI CEO says "we need to have discussions about UBI" but what he's really saying is "look how fucking good our AI is, it's gonna take all your jobs!".
There is definitely fear mongering and outlandish predictions going on. What those companies say are heavily biased.
>On the one hand, it
is well documented in the literature on environmental racism that
the negative effects of climate change are reaching and impacting
the world’s most marginalized communities first