Hacker News Reader

Comment by Xcelerate | Hacker News Reader

Unlike previous hype cycles, the potential value of this one is extraordinary if it’s actually unlocked (I mean, what was the theoretical upper limit on the benefit of cryptocurrency for the world? Probably not that much.) Previous attempts at AI/AGI have been constrained by computational resources. It’s quite possible that we already have sufficient computational power and the necessary data for AGI—all we need are the right algorithms.

Even if for some bizarre reason we’ve already tapped the maximum potential of transformer architectures and all of this money goes nowhere, compared to all the other ways that society wastes money, I would be fine with calling this a big bet for humanity that didn’t pay off. It doesn’t mean that it wasn’t worth the attempt though.

View on HN

> Unlike previous hype cycles, the potential value of this one is extraordinary if it’s actually unlocked

That was also true, in AI, of the expert-system hype cycle. And the actual value unlocked was extraordinary, just not at the scale people saw as the potential.

Actually, it was seen as true of all of the hype cycles during the hype cycle, that's what makes it a hype cycle.

> (I mean, what was the theoretical upper limit on the benefit of cryptocurrency for the world? Probably not that much.)

If you believed the people that were as breathless about it as you are about the current AI hype cycle, basically infinite, unlocking ways human potential and interactions, economic and otherwise, are held back by centralized and/or authoritarian systems.

That's what made it a hype cycle.

> It’s quite possible that we already have sufficient computational power and the necessary data for AGI—all we need are the right algorithms.

Yeah, but that's always been true. If software-only AGI is possible, we've always had the data in the natural world, and with no strong theoretical model for the necessary computational power, its always been possible we had enough. What we clearly lacked were the right algorithms (oh, and any reason to believe software-only AGI was possible.)

I think I agree with basically your whole comment but I'm wondering if you could explain what you mean by "software-only AGI". Obviously all software runs on hardware, and creating specialized hardware to run certain types of software is something the computing industry is already very familiar with.

In the far far future, if we did crack AGI, it's not impossible to believe that specialized hardware modules would be built to enable AGI to interface with a "normal" home computer, much like we already add modules to our computers for specialized applications. Would this still count as software-only AI to you?

I've held for a long time that sensory input and real-world agency might be necessary to grow intelligence, so maybe you mean something like that, but even then that's something not incredibly outside the realm of what regular computers could do with some expansion.

There's some discussion of embodiment as an important factor in intelligence such that it would defy pure software implementation. I’m personally of the opinion that even to the extent this is true, it probably just means the compute capacity required for software is higher than we might otherwise think, to simulate the other parts, alternatively, with the right interfaces and hardware, we don't need that cheat. But “everything involved can be simulated in software at the required level”, while I believe it, somewhat speculative.

https://en.m.wikipedia.org/wiki/Portia_(spider)

This spider could be evidence of "software based intelligence" in biological brains - it exhibits much more complex behaviors than other animals it's size, more comparable to cats and dogs.

What I mean is that some believe that their brain is "emulating" all parts of the larger "brain", but one at a time, and passing the "data" that comes out of one into the next.

Just a cool thing.

> Unlike previous hype cycles, the potential value of this one is extraordinary if it’s actually unlocked

That sounds exactly like most hype cycles, it's almost a tautology that the perceived potential value is immense (at least to enough people).

Consider e.g. the hype around "the internet" in early mid nineties, which led to the dot.com collapse. Today the internet has undeniably had a massive impact globally, so the naysayers have been comprehensively proven wrong. On the other hand, the most optimistic views have not begun to come to pass yet (ever?) either. Lots of ideas that were floated in the 90s didn't really work until 10, 15, 20 years later. Some things that are now ubiquitous weren't really conceived of then, etc. etc. As usual, it turned out the technology wasn't the really hard part.

So far the current AI cycle seems to be following the usual playbook.

Any correlation between current models and AGI is, IMO, hype.

GenAI is a remarkably useful tool, but its not one step away from an AGI.

depends on how you define "step". Engineer a 10x/100x version of what we have in terms of LLM (either by being more efficient and/or more/specialized hardware) and let this thing build novel attempts for AGI algorithms 24/7 in a evolutionary setting.

I guess the challenge is more to agree on a fitness function to measure the "AGI"-progress" against, but thats a different topic. But in general scaling up the current GenAI tech and parallelize/specialize the models in a multi-generational way _should_ be a safe ticket to AGI, but the time scale is inknown of course (since we can't even agree on the goal definition)

I like this comment because I think it highlights the exact difference between AI optimists and AI cynics.

I think you'll find that AGI cynics do not agree at all that "engineering a 10x/100x version" of what we have and making it attempt "AGI algorithms 24/7 in an evolutionary setting" is a "safe ticket" to AGI.

I wouldn’t say I’m a cynic, I’d just say how can one possibly know what a safe ticket is in this space? The logic you described is basically simple extrapolation, like in the xkcd wedding dress comic. There’s no guarantee that will get you anywhere in finite time.

"depends on how you define "step". Engineer a 10x/100x version of what we have in terms of LLM (either by being more efficient and/or more/specialized hardware) and let this thing build novel attempts for AGI algorithms 24/7 in a evolutionary setting."

The current LLM's get stuck in loops when a problem is too hard for it. They just keep doing the wrong thing over and over. It's not obvious this sort of ai can "build novel attempts" at hard problems.

> (I mean, what was the theoretical upper limit on the benefit of cryptocurrency for the world? Probably not that much.)

According to the crypto-faithful at the time: solving territorial disputes (Gaza Strip? blockchain solves this!), identity management, bank transfers, payments over the internet with no transaction fees, "the supply chain" (whatever that means), etc. Not as interesting to a layperson as AGI, but if all those (or ANY of those) ended up panning out, crypto would have been a multi-trillion dollar industry and fundamentally transformed vast swathes of modern society.

I do think LLMs are far more useful than blockchain, but claiming "the potential value of this one is extraordinary" is exactly what people said in previous hype cycles.

> crypto would have been a multi-trillion dollar industry

what metric would you like to use, specifically? double check that its a metric that matches other industries

the market cap of the digital spot commodities? the marketcap of the businesses that use the digital spot commodities? the revenue of all participants and service providers? the volume of all shares and futures and spot trades when sliced down to a submetric that represents 'real' trades? all of the above?

> and fundamentally transformed vast swathes of modern society

thats ...a... goal post. I'm not sure if that's a goal post I would have, its market microstructure plumbing. At best, it modifies capital formation, letting different ventures get funding, which it already has.

and then, what time frame? its a pretty good S-curve from 2009. there is a pretty clear chronology of what delays what, everything that has resulted in a seasonal bubble in crypto comes from a software proposal being ratified that allows it to touch another industry that it previously didn't. Many overlapping similarities to IETF proposals for WWW, but I understand this level of discussion might not reach your circles, the point stands that there are plenty of people in the tech space that had the exact same observation and you and chose to contribute to the proposals that make crypto now more accessible to the next group.

There are plenty of proposals now in many different crypto communities, even ones to make ratification more egalitarian and collaborative.

some turn out to be hits for adoption.

I think it is interesting for people to then use that reality to say crypto hasnt fulfilled any lofty idea they overheard an enthusiast say, because it took too long.

Prior proposals and their ratification were necessary for the reported market cap to reach $1bn, but I know I know “market cap!? you cant sell it all at once!” Holding crypto assets and industry to a separate higher standard than all other industries on the planet.

> crypto would have been a multi-trillion dollar industry

The same people saying this are often the same ones betting on sovereign currencies crashing.

> I mean, what was the theoretical upper limit on the benefit of cryptocurrency for the world

The value of potential bank scams that are otherwise illegal was enormous to investors though. Lots of people got extremely wealthy thanks to crypto scams. Then when the legal holes were covered crypto was forgotten extremely quickly since the hype was mostly kept alive by scams.

AI doesn't have nearly as lucrative scams, so I doubt you will see the same investor frenzy.

> AI doesn't have nearly as lucrative scams, so I doubt you will see the same investor frenzy.

Maybe you are right about the “frenzy”, but quantitatively speaking the market cap of Big Tech (including and especially nVidia) is probably larger than the crypto scams ever will be.

As a comparison the market cap of crypto is apparently less than the cap of nVidia. Share prices of other tech companies like Microsoft are also inflated due to expectations of AI related returns.

The “frenzy” may not be as insane but the money is definitely there. Especially with the crypto bubbles bursting the money has to go somewhere, unless you honestly believe they ended up in US treasury bonds or sth

The potential downside is extraordinary too

Touché

Reply on news.ycombinator.com