Remix.run Logo
timewizard 2 days ago

That we don't have a single unified explanation doesn't mean that we don't have very good hints, or that we don't have very good understandings of specific components.

Aside from that the measure really, to me, has to be power efficiency. If you're boiling oceans to make all this work then you've not achieved anything worth having.

From my calculations the human brain runs on about 400 calories a day. That's an absurdly small amount of energy. This hints at the direction these technologies must move in to be truly competitive with humans.

threatofrain 2 days ago | parent | next [-]

We'll be experiencing extreme social disruption well before we have to worry about the cost-efficiency of strong AI. We don't even need full "AGI" to experience socially momentous change. We might even be on the verge of self driving cars spreading to more cities.

We don't need very powerful AI to do very powerful things.

yibg 2 days ago | parent | next [-]

It's not just a energy cost issue with AGI though. With autonomous vehicles we might not have the technology, but we can build a good mental model of what the thing can look like and how various pieces can function long before we get there. We have different classifications of incremental steps to get there as well. e.g. level 1, 2 and so on where we can make incremental progress.

With AGI, as far as I know, no one has a good conceptual model of what a functional AGI even looks like. LLM is all the rage now, but we don't even know if it's a stepping stone to get to AGI.

timewizard 2 days ago | parent | prev [-]

> experiencing extreme social disruption

I think this just displays an exceptionally low estimation of human beings. People tend to resist extremities. Violently.

> experience socially momentous change

The technology is owned and costs money to use. It has extremely limited availability to most of the world. It will be as "socially momentous" as every other first world exclusive invention has been over the past several decades. 3D movies were, for a time, "socially momentous."

> on the verge of self driving cars spreading to more cities.

Lidar can't read street lights and vision systems have all sorts of problems. You might be able to code an agent that can drive a car but you've got some other problems that stand in the way of this. AGI is like 1/8th the battle. I referenced just the brain above. Your eyes and ears are actually insanely powerful instruments in their own right. "Real world agency" is more complicated than people like to admit.

> We don't need very powerful AI to do very powerful things.

You've lost sight of the forest for the trees.

achierius 2 days ago | parent [-]

Re.: self driving cars -- vision systems have all sorts of problems sure, but on the other hand that _is_ what we use. The most successful platforms use Lidar + vision -- vision can handle the streetlights, lidar detects objects, etc.

And more practically -- these cars are running in half a dozen cities already. Yes, there's room to go, but pretending there are 'fundamental gaps' to them achieving wider deployment is burying your head in the sand.

adgjlsfhk1 2 days ago | parent | prev | next [-]

note that those are kilocalories, and that is ignoring the calories needed for the circulatory and immune systems which are somewhat necessary for proper function. Using 2000 cal per day/10 hours of thinking gives a consumption of ~200W

mschuster91 2 days ago | parent [-]

> Using 2000 cal per day/10 hours of thinking gives a consumption of ~200W

So, about a tenth or less of a single server packed to the top with GPUs.

stevenwoo 2 days ago | parent | next [-]

That is true but there’s 3.7 billion years of evolutionary “design” to make self replicating, self fueling animals to use that brain. There’s no AI within foreseeable future capable of that. One might look at brains as a side effect of evolution of the self replicating, self fueling bits.

paulryanrogers 2 days ago | parent | prev [-]

Are the wattage ratings of GPUs how many they need continuously or over hours?

adgjlsfhk1 2 days ago | parent [-]

Watts are a unit of power which is energy per time. specifically, 1 Watt is 1 joule per second

HDThoreaun a day ago | parent | prev | next [-]

We are very good at generating energy. Even if AI is an order of magnitude less energy efficient an AI person equivalent would use ~ 4 kilowatt hours/day. At current rates thats like $1. Hardly the limiting factor here I think

imtringued 2 days ago | parent | prev [-]

Energy efficiency is not really a good target since you can brute force it by distilling classical ANNs to spiking neural networks.