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

How to feed all the power-hungry AI data centers

caseyhandmer.wordpress.com

Post reply on HN
Print view View original

51–60 of 63 posts

Re: How to feed all the power-hungry AI data centers

#51

Post by GaggiX »

parpfish wrote: post #35
GaggiX wrote: post #25

Earlier quoted context omitted.

Has AI not already been adopted by the masses?

Absolutely not. I’d wager that a slim majority of Americans have heard of chatGPT, much less have used it, and even fewer have done something useful/productive with it. Here are stats from 6 months ago, I doubt it’s changed significantly since then: https://www.pewresearch.org/short-reads/2023/08/28/most-amer...

>about four-in-ten adults under 30 have used it

That seems pretty massive honestly, that's way beyond what I would consider adopted by the masses.

Re: How to feed all the power-hungry AI data centers

#52

Post by grecy »

ProllyInfamous wrote: post #37
grecy wrote: post #11

Earlier quoted context omitted.

Bill Gates has a $1B short position on Tesla. I wouldn't use what Bill Gates is investing in as any indicator of what we should be doing.

Tesla has been on a steady [and IMHO rightful] decline since Jan 3, 2024.

I meant in terms of doing the right thing for the planet

Re: How to feed all the power-hungry AI data centers

#53

Post by parpfish »

GaggiX wrote: post #51
parpfish wrote: post #35

Earlier quoted context omitted.

Absolutely not. I’d wager that a slim majority of Americans have heard of chatGPT, much less have used it, and even fewer have done something useful/productive with it. Here are stats from 6 months ago, I doubt it’s changed significantly since then: https://www.pewresearch.org/short-reads/2023/08/28/most-amer...

>about four-in-ten adults under 30 have used it That seems pretty massive honestly, that's way beyond what I would consider adopted by the masses.

That’s four intern among those that have heard of it. So cut that in half to get the percentage of the population.

Or another way of thinking of it: among people that have heard of your free-to-use product, a majority haven’t bothered to try it.

Re: How to feed all the power-hungry AI data centers

#54

Post by infecto »

jairuhme wrote: post #27
infecto wrote: post #8

Earlier quoted context omitted.

It’s always interesting to see these ideas on a tech forum. Should we as humanity stop right here? Draw a line in the sand and not do anything new? Maybe we should move backwards, skipping the Industrial Revolution and go back to each individual growing their own food? This is a serious question. What do you propose? Because you are not able to pick and choose what humanity tries out, so shall we stop entirely?

I'd argue that the innovation with these models is making smaller. Just throwing compute resources to make a model with more parameters is easy and doesn't really expand our knowledge. IMO, larger and larger LLM's aren't that impressive. Being able to shrink that model down, retain its accuracy (to a degree) and be able to run it on smaller hardware is impressive and will more likely lead to AI/ML being intertwined w…

Don’t disagree and I think it is natural evolution similar other aspect of tech. We innovate and then make it more efficient. I don’t want to stop the opportunity to innovate because of electricity usage though.

Re: How to feed all the power-hungry AI data centers

#55

Post by infecto »

korijn wrote: post #18
infecto wrote: post #8

Earlier quoted context omitted.

It’s always interesting to see these ideas on a tech forum. Should we as humanity stop right here? Draw a line in the sand and not do anything new? Maybe we should move backwards, skipping the Industrial Revolution and go back to each individual growing their own food? This is a serious question. What do you propose? Because you are not able to pick and choose what humanity tries out, so shall we stop entirely?

We should invent solutions for problems that we have, instead of wildly trying to apply this tech to everything we can think of. We are skipping the benefit/cost considerations.

Many companies and individuals are figuring out the benefits. There are number of companies that are seeing real value. We are not skipping anything.

Re: How to feed all the power-hungry AI data centers

#56

Post by atoav »

PeterisP wrote: post #45
atoav wrote: post #31

Earlier quoted context omitted.

> Should we as humanity stop right here? Draw a line in the sand and not do anything new? Maybe someone with the knowledge of hindsight in a few generations will say that, yes, we should stop and draw a line in the sand. We are still used to the idea that we can treat externalities to the systems we build as infinite and boundless, while it is getting increasingly clear that they are not. I am not saying we should st…

There are unaccounted externalities in many areas, but energy use of LLMs isn't one of them - people are paying full market price for the energy of training or using LLMs, that energy cost is large enough so that people care a LOT about it, and make intentional tradeoffs about what is worth it. And if people vote with their wallets that they are willing to pay for a years worth of energy on training LLMs on cat pictu…

Jup, and the market price for energy doesn't reflect where it is going towards, which was my point. Whether you train the LLM to predict extreme weather events or put funny hats on pictures of cats, you have to pay the same — in fact, the latter will drive the price up for the former.

Sure, there are still people out there who believe the invisible hand of the market will regulate everything to give us a good outcome despite every real world evidence to the opposite. But in a market more demand for bullshit reasons drives the prices up for everybody, also for things that are essential for the survival of the poorest, the long term sustainability of environment, health services, etc.

The question I laid out was whether it is really a good idea to treat some random bullshit the same as essential services and externalities? Especially considering we are walking into a climate catastrophe that is the result of too much energy use.

Granted, we live in a society that is kinda built on ignoring these connections — otherwise we would go mad or cynical, but lying to ourselves has a hidden cost that will come due at some point.

Re: How to feed all the power-hungry AI data centers

#57

Post by PeterisP »

atoav wrote: post #56
PeterisP wrote: post #45

Earlier quoted context omitted.

There are unaccounted externalities in many areas, but energy use of LLMs isn't one of them - people are paying full market price for the energy of training or using LLMs, that energy cost is large enough so that people care a LOT about it, and make intentional tradeoffs about what is worth it. And if people vote with their wallets that they are willing to pay for a years worth of energy on training LLMs on cat pictu…

Jup, and the market price for energy doesn't reflect where it is going towards, which was my point. Whether you train the LLM to predict extreme weather events or put funny hats on pictures of cats, you have to pay the same — in fact, the latter will drive the price up for the former. Sure, there are still people out there who believe the invisible hand of the market will regulate everything to give us a good outcome…

But as we don't have mass long-term energy storage, future energy is not exchangeable with current energy, thus the future energy cost shouldn't matter for the current training of LLMs, and whether tomorrow it will be a good use of energy to put funny hats on pictures of cats, it's something that people will be able to choose based on the availability of energy - it's up to them to decide what's essential or not, it's up to them to decide whether they are willing to sacrifice some entertainment just to save energy, and that's what they should be doing with money as the mechanism of measure - if a person wants to do something silly, it would be inappropriate for someone else to decide "no, that's random bullshit" and prohibit them to spend their resources (energy or otherwise) on it. The fact that this decision will affect the price of e.g. energy for others is NOT an externality, it's the core economic mechanism of allocating scarce resources - if some people want to spend the same resources on cat picture and they are fully paying the costs of that, then that's directly reflected in the price they're paying, not an unpriced externality.

Re: How to feed all the power-hungry AI data centers

#58

Post by atoav »

PeterisP wrote: post #57
atoav wrote: post #56

Earlier quoted context omitted.

Jup, and the market price for energy doesn't reflect where it is going towards, which was my point. Whether you train the LLM to predict extreme weather events or put funny hats on pictures of cats, you have to pay the same — in fact, the latter will drive the price up for the former. Sure, there are still people out there who believe the invisible hand of the market will regulate everything to give us a good outcome…

But as we don't have mass long-term energy storage, future energy is not exchangeable with current energy, thus the future energy cost shouldn't matter for the current training of LLMs, and whether tomorrow it will be a good use of energy to put funny hats on pictures of cats, it's something that people will be able to choose based on the availability of energy - it's up to them to decide what's essential or not, it'…

Are you sure you aren't deep in cognitive-dissonance land now?

More energy use overall — momentary or over time (=work), doesn't make a difference here — especially with LLMs which aren't exactly producing bursty loads and could be in theory scheduled to any time of the day. And energy corps will just have to produce more energy, which has it's own negative externalities on the world.

That means my stupid project still has an impact on the rest of the world. As of now everybody luckily still has the luxury to figure this out themselves. But we have to realize that depending on how bad things get this won't be feasible forever. And how fast things get bad depends on our collective energy use. If you're in a desert with limited water you won't let your kid pour it on the water because it makes a fun sound wouldn't you?

You would only let your kid do that if you had no mental model of water being a limited resouece and no imagination of the consequences a lack of water would mean for you and your kid.

Energy is a limited resource and producing more of it has it's own secondary costs (e.g. for climate change).

Operating with limited resources doesn't mean we are not allowed to have fun. It just means you don't pour your limited resource on the ground for it.

Re: How to feed all the power-hungry AI data centers

#59

Post by PeterisP »

atoav wrote: post #58
PeterisP wrote: post #57

Earlier quoted context omitted.

But as we don't have mass long-term energy storage, future energy is not exchangeable with current energy, thus the future energy cost shouldn't matter for the current training of LLMs, and whether tomorrow it will be a good use of energy to put funny hats on pictures of cats, it's something that people will be able to choose based on the availability of energy - it's up to them to decide what's essential or not, it'…

Are you sure you aren't deep in cognitive-dissonance land now? More energy use overall — momentary or over time (=work), doesn't make a difference here — especially with LLMs which aren't exactly producing bursty loads and could be in theory scheduled to any time of the day. And energy corps will just have to produce more energy, which has it's own negative externalities on the world. That means my stupid project sti…

One man's "pour your limited resource on the ground" is another man's "valuable use of that resource" - if they're choosing to do something with the energy they purchase, apparently they consider that this is a worthwhile use of that resource. If I'm in the desert and want to pour some water on the ground for growing orchids, that's my choice on how to use it - if it's really important to conserve water, then it will be costly enough so that I won't waste much of it on luxuries, but if it's cheap enough to do so, then people are voting with their wallets that it's NOT really (yet?) so important to conserve it so much as to refuse spending that on luxuries.

Re: How to feed all the power-hungry AI data centers

#60

Post by ProllyInfamous »

grecy wrote: post #52
ProllyInfamous wrote: post #37

Earlier quoted context omitted.

Tesla has been on a steady [and IMHO rightful] decline since Jan 3, 2024.

I meant in terms of doing the right thing for the planet

They're not doing that, either... the only reason you would drive a Tesla is performance. The environmental impact of making a full-EV is much more damaging than using a gas engine.
Post reply on HN