Hacker News Board Hacker News, browsed like a classic message board
Live data from Hacker News

AGI is an engineering problem, not a model training problem

vincirufus.com

Post reply on HN
Print view View original

51–60 of 442 posts

Re: AGI is an engineering problem, not a model training problem

#51

Post by ysavir »

tshaddox wrote: post #21
glitchc wrote: post #10

We don't know if AGI is even possible outside of a biological construct yet. This is key. Can we land on AGI without some clear indication of possibility (aka Chappie style)? Possibly, but the likelihood is low. Quite low. It's essentially groping in the dark. A good contrast is quantum computing. We know that's possible, even feasible, and now are trying to overcome the engineering hurdles. And people still think th…

> We don't know if AGI is even possible outside of a biological construct yet. This is key. A discovery that AGI is impossible in principle to implement in an electronic computer would require a major fundamental discovery in physics that answers the question “what is the brain doing in order to implement general intelligence?”

Seems the opposite way round to me. We couldn't conclusively say that AGI is possible in principle until some physics (or rather biology) discovery explains how it would be possible. Until then, anything we engineer is an approximation as best.

Re: AGI is an engineering problem, not a model training problem

#52

Post by gwern »

dcre wrote: post #40

The first premise of the argument is that LLMs are plateauing in capability and this is obvious from using them. It is not obvious to me.

It is especially not obvious because this was written using ChatGPT-5. One appreciates the (deliberate?) irony, at least. (Or at least, surely if they had asymptoted, OP should've been able to write this upvoted HN article with an old GPT-4, say...)

Re: AGI is an engineering problem, not a model training problem

#53

Post by tshaddox »

throwaway31131 wrote: post #30
tshaddox wrote: post #21

Earlier quoted context omitted.

> We don't know if AGI is even possible outside of a biological construct yet. This is key. A discovery that AGI is impossible in principle to implement in an electronic computer would require a major fundamental discovery in physics that answers the question “what is the brain doing in order to implement general intelligence?”

We would also need a definition of AGI that is provable or disprovable. We don’t even have a workable definition, never mind a machine.

We don’t need such a definition of general intelligence to conclude that biological humans have it, so I’m not sure why we’d such a definition for AGI.

Re: AGI is an engineering problem, not a model training problem

#54

Post by karmakaze »

sublinear wrote: post #33
karmakaze wrote: post #18

Earlier quoted context omitted.

This is the funniest "I'm a hammer thus AGI is a nail" post I've ever read.

Maybe I'm misunderstanding what you mean by that, but do you have any examples of software engineering that weren't already thoroughly explained by computer science long before?

By this, I meant the original post.. in agreement.

Re: AGI is an engineering problem, not a model training problem

#56

Post by AIPedant »

tshaddox wrote: post #21
glitchc wrote: post #10

We don't know if AGI is even possible outside of a biological construct yet. This is key. Can we land on AGI without some clear indication of possibility (aka Chappie style)? Possibly, but the likelihood is low. Quite low. It's essentially groping in the dark. A good contrast is quantum computing. We know that's possible, even feasible, and now are trying to overcome the engineering hurdles. And people still think th…

> We don't know if AGI is even possible outside of a biological construct yet. This is key. A discovery that AGI is impossible in principle to implement in an electronic computer would require a major fundamental discovery in physics that answers the question “what is the brain doing in order to implement general intelligence?”

It is vacuously true that a Turing machine can implement human intelligence: simply solve the Schrödinger equation for every atom in the human body and local environment. Obviously this is cost-prohibitive and we don’t have even 0.1% of the data required to make the simulation. Maybe we could simulate every single neuron instead, but again it’ll take many decades to gather the data in living human brains, and it would still be extremely expensive computationally since we would need to simulate every protein and mRNA molecule across billions of neurons and glial cells.

So the question is whether human intelligence has higher-level primitives that can be implemented more efficiently - sort of akin to solving differential equations, is there a “symbolic solution” or are we forced to go “numerically” no matter how clever we are?

Re: AGI is an engineering problem, not a model training problem

#58

Post by tshaddox »

aorloff wrote: post #32
tshaddox wrote: post #21

Earlier quoted context omitted.

> We don't know if AGI is even possible outside of a biological construct yet. This is key. A discovery that AGI is impossible in principle to implement in an electronic computer would require a major fundamental discovery in physics that answers the question “what is the brain doing in order to implement general intelligence?”

A question which will be trivial to answer once you properly define what you mean by "brain" Presumably "brains" do not do many of the things that you will measure AGI by, and your brain is having trouble understanding the idea that "brain" is not well understood by brains. Does it make it any easier if we simplify the problem to: what is the human doing that makes (him) intelligent ? If you know your historical cont…

> Does it make it any easier if we simplify the problem to: what is the human doing that makes (him) intelligent ?

Sure, it doesn’t have to be literally just the brain, but my point is you’d need very new physics to answer the question “how does a biological human have general intelligence?”

Re: AGI is an engineering problem, not a model training problem

#59

Post by DrewADesign »

singpolyma3 wrote: post #37
glitchc wrote: post #10

We don't know if AGI is even possible outside of a biological construct yet. This is key. Can we land on AGI without some clear indication of possibility (aka Chappie style)? Possibly, but the likelihood is low. Quite low. It's essentially groping in the dark. A good contrast is quantum computing. We know that's possible, even feasible, and now are trying to overcome the engineering hurdles. And people still think th…

This makes no sense. If you believe in eg a mind or soul then maybe it's possible we cannot make AGI. But if we are purely biological then obviously it's possible to replicate that in principle.

That doesn’t contradict what they said. We may one day design a biological computing system that is capable of it. We don’t entirely understand how neurons work; it’s reasonable to posit that the differences that many AGI boosters assert don’t matter do matter— just not in ways we’ve discovered yet.

Re: AGI is an engineering problem, not a model training problem

#60

Post by bee_rider »

I think the author could have picked a better title. “ is an engineering problem” is a pretty common expression to describe something where the science is done, but the engineering remains. There’s an understanding that that could still mean a ton of work, but there isn’t some fundamental mystery about the basic natural principles of the thing.

Here, AGI is being described as an engineering problem, in contrast to a “model training” problem. That is, I think at least, he’s at least saying that more work needs to be done at an R&D level. I agree with those who are saying it is maybe not even an engineering problem yet, but should be noted that he’s at least pushing away from just running the existing programs harder, which seems to be the plan with trillions of dollars behind it.

Post reply on HN