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

AGI fantasy is a blocker to actual engineering

tomwphillips.co.uk

Post reply on HN
Print view View original

581–590 of 683 posts

Re: AGI fantasy is a blocker to actual engineering

#581

Post by kryogen1c »

rekrsiv wrote: post #272
wat10000 wrote: post #224

Earlier quoted context omitted.

Evolution has won so far because it had a four billion year head start. In two hundred years, technology has gone from "this multi-ton machine can do arithmetic operations on large numbers several times faster than a person" to "this box produces a convincing facsimile of human conversation, but it only emulates a trillion neurons and they're not nearly as sophisticated as real ones." I do think we probably need a ne…

But we don't even need a human brain. We already have those, they take months to grow, take forever to train, and are forever distracted. Our logic-based processes will keep getting smaller and less power hungry as we figure out how to implement them at even lower scales, and eventually we'll be able to solve problems with the same building blocks as evolution but in intelligent ways, of which LLMs will likely only p…

>But we don't even need a human brain. We already have those, they take months to grow, take forever to train

This is a weird argument considering LLMs are composed of the output of countless hours of human brains. That makes LLMs, by definition, logarithmically worse at learning.

Re: AGI fantasy is a blocker to actual engineering

#582

Post by kadushka »

advisedwang wrote: post #344
IgorPartola wrote: post #21

It is ultimately a hardware problem. To simplify it greatly, an LLM neuron is a single input single output function. A human brain neuron takes in thousands of inputs and produces thousands of outputs, to the point that some inputs start being processed before they even get inside the cell by structures on the outside of it. An LLM neuron is an approximation of this. We cannot manufacture a human level neuron to be s…

It's even worse than number of input/outputs, number of neurons, efficiency or directional feedback. The brain also has plasticity! The connections between neurons change dynamically - an extra level of meta.

Connections between LLM neurons also change during training.

Re: AGI fantasy is a blocker to actual engineering

#583

Post by buu700 »

kragen wrote: post #576
buu700 wrote: post #568

Earlier quoted context omitted.

I'm not an expert on the subject, but is there an obvious bottleneck that would prevent Helion from manufacturing an arbitrarily large number of generators that scales with demand? 2000 generators doesn't seem like an overwhelmingly large number if they get paid enough to build out capacity for it, and they've already started on an assembly line manufacturing plant. I'd expect that they probably wouldn't focus on cop…

> I'm not an expert on the subject, but is there an obvious bottleneck that would prevent Helion from manufacturing an arbitrarily large number of generators that scales with demand? No, nothing is obvious to anyone outside the company. > I'd expect that they probably wouldn't focus on copying Orion exactly, Hopefully their next model will come out before another billion dollars of investment goes down the drain. > O…

Yeah, there are lots of unknowns and valid reasons to be pessimistic or optimistic. I lean toward cautious optimism based on the information that is available, but of course no one knows the future until it happens.

Just to clarify, Orion is one site, not a particular generator model. Orion working would be the green light to build and ship more generators, not a sign that they'd need to go back to the drawing board on an "Orion v2".

re: opex, we do have some idea of what it will be. We know that deuterium is cheap, we know that fusion produces low waste, we know that extensive safety protocols to protect against meltdown are not required, we know that preventing theft of fissile materials is not a security concern, and we know that these machines are substantially automated. We may not have precise balance sheets, but it's also not entirely opaque or mysterious.

I'm not sure the examples of historical China and Japan apply. They didn't invent coal power and then decide not to use it (although Japan was nevertheless fairly quick to industrialize). Fusion as of now seems likely to reach commercialization first in the US, and there's currently a lot of demand for power here with insufficient supply to meet projected growth. It's a very different market from when we were actively shipping our high-power-consumption industries overseas.

I don't agree that breakthroughs happen "rarely". I mean, we wouldn't be here in this thread without the massive breakthrough of useful LLMs, which were considered sci-fi before ChatGPT's release less than three years ago; fusion had the breakthrough of ignition at NIF just a few days after that; and we're only in the position of considering solar and batteries as a competitive energy source due to many small breakthroughs that have occurred in recent history. Breakthroughs happen all the time, but we quickly forget their significance and let them become boring.

Re: AGI fantasy is a blocker to actual engineering

#584

Post by rzerowan »

cruffle_duffle wrote: post #399
IgorPartola wrote: post #21

It is ultimately a hardware problem. To simplify it greatly, an LLM neuron is a single input single output function. A human brain neuron takes in thousands of inputs and produces thousands of outputs, to the point that some inputs start being processed before they even get inside the cell by structures on the outside of it. An LLM neuron is an approximation of this. We cannot manufacture a human level neuron to be s…

“And the final kicker: the human brain runs on like two dozen Watts. An LLM takes a year of running on a few MW to train and several KW to run.” I’ve always thought about nature didn’t evolve to use electricity as its primary means of energy. Instead it uses chemistry. It’s quite curious, really. Like a tiny insect is chemistry powered. It doesn’t need to recharge batteries, it needs to eat and breathe oxygen. What i…

"Wetware" as it were , i remeber some research article some time back where they grew some kind of 'brainlets' and had them functional as far as memory was concerned[1]. Would be a interesting to see how that tech would progress while incororatd in the current silicon/photonic devices for input/output. Downside would be the durability of the organic matter and replacement.

[1] https://sciencesensei.com/scientists-created-thinking-brain-...

Re: AGI fantasy is a blocker to actual engineering

#585

Post by thunky »

fallingfrog wrote: post #75
fallingfrog wrote: post #9

Earlier quoted context omitted.

I have some idea of what the way forward is going to look like but I don't want to accelerate the development of such a dangerous technology so I haven't told anyone about it. The people working on AI are very smart and they will solve the associated challenges soon enough. The problem of how to slow down the development of these technologies- a political problem- is much more pressing right now.

[flagged]

If you're so confident that your discovery will be found by someone else soon why not just share it right here and get the credit you seem to want? Otherwise why even bring it up in the first place?

Re: AGI fantasy is a blocker to actual engineering

#586

Post by DrierCycle »

rekrsiv wrote: post #272
wat10000 wrote: post #224

Earlier quoted context omitted.

Evolution has won so far because it had a four billion year head start. In two hundred years, technology has gone from "this multi-ton machine can do arithmetic operations on large numbers several times faster than a person" to "this box produces a convincing facsimile of human conversation, but it only emulates a trillion neurons and they're not nearly as sophisticated as real ones." I do think we probably need a ne…

But we don't even need a human brain. We already have those, they take months to grow, take forever to train, and are forever distracted. Our logic-based processes will keep getting smaller and less power hungry as we figure out how to implement them at even lower scales, and eventually we'll be able to solve problems with the same building blocks as evolution but in intelligent ways, of which LLMs will likely only p…

The brain is logical by design, it implements logic?

Re: AGI fantasy is a blocker to actual engineering

#587

Post by AndrewKemendo »

prmph wrote: post #536
AndrewKemendo wrote: post #236

Earlier quoted context omitted.

None of which have to do with AI or AGI. Nestle is and has been 10000x worse for global water security than all other companies and countries combined because nobody in the value chain cares about someone else’s aquifer. It’s a social-economic problem of externalities being ignored , which transcends any narrow technological use case. What you describe has been true for all exported manufacturing forever.

I think the point is: where does this end? Do we continue to build orders-of-magnitude bigger models guzzling orders-of-magnitude more water and other resources, in pursuit of the elusive AGI? At some we need to end this AGI" rat race and focus on realizing practical benefits from the models we currently have.

Well then vote with your wallet and convince everyone you know to stop the train.

As long as there’s a market, machines are going to continue toto displace labor. That’s not going to stop

Re: AGI fantasy is a blocker to actual engineering

#588

Post by ThrowawayR2 »

bob1029 wrote: post #279
IgorPartola wrote: post #21

It is ultimately a hardware problem. To simplify it greatly, an LLM neuron is a single input single output function. A human brain neuron takes in thousands of inputs and produces thousands of outputs, to the point that some inputs start being processed before they even get inside the cell by structures on the outside of it. An LLM neuron is an approximation of this. We cannot manufacture a human level neuron to be s…

To be fair to the raw capabilities of the semiconductor industry, a 100mm^2 die at 3nm can contain on the order of 1~10 trillion features. I don't know that we are actually that far off in terms of scale. How to arrange these features seems to be the difficult part. The EDA [0] problem is immune to the bitter lesson. There are certainly specific arrangements of matter that can solve this problem better than a GPU/TPU…

Unlike neurons, connections those features can't be rewired on the fly so the comparison is not even meaningful.

Re: AGI fantasy is a blocker to actual engineering

#589

Post by NamlchakKhandro »

danielbln wrote: post #202
paperplaneflyr wrote: post #19

After reading Empire of AI by Karen Hao, actually changed my perspective towards these AI companies, not that they are building world-changing products but the human nature around all this hype. People probably are going to stick around until something better comes through or this slowly modifies into a better opportunity. Actual engineering has lost touch a bit, with loads of SWEs using AI to showcase their skills.…

Can you elaborate on that last sentence a bit? How has engineering lost touch?

The act of putting characters in files isn't the value.

Re: AGI fantasy is a blocker to actual engineering

#590

Post by nandomrumber »

us-merul wrote: post #34
IgorPartola wrote: post #21

It is ultimately a hardware problem. To simplify it greatly, an LLM neuron is a single input single output function. A human brain neuron takes in thousands of inputs and produces thousands of outputs, to the point that some inputs start being processed before they even get inside the cell by structures on the outside of it. An LLM neuron is an approximation of this. We cannot manufacture a human level neuron to be s…

This is a great summary! I've joked with a coworker that while our capabilities can sometimes pale in comparison (such as dealing with massively high-dimensional data), at least we can run on just a few sandwiches per day.

One sandwich is about the energy equivalent of running about two modern desktop CPU’s (flat out) for about an hour.
Post reply on HN