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

Covering electricity price increases from our data centers

anthropic.com

Post reply on HN
Print view View original

101–110 of 110 posts

Re: Covering electricity price increases from our data centers

#101

Post by keeda »

gosub100 wrote: post #89
keeda wrote: post #19

Earlier quoted context omitted.

> These things are so hideously inefficient. Quite the opposite, really. I did some napkin math for energy and water consumption, and compared to humans these things are very resource efficient. If LLMs improve productivity by even 5% (studies actually peg productivity gains across various professions at 15 - 30%, and these are from 2024!) the resource savings by accelerating all knowledge workers are significant. Si…

Are you excluding the cost of training the AI from the calculation?

In the initial analysis of a single worker, yes, but when scaling up per-human savings to use by the wider population, the aggregate resource savings compensate for training resource usage within a few days, weeks at most.

Re: Covering electricity price increases from our data centers

#102

Post by keybored »

keeda wrote: post #100
keybored wrote: post #64

Earlier quoted context omitted.

What I gather from your analysis and gleeful exclamation marks is that I should start rioting now rather than wait.

Hah, interesting, I intended my exclamation marks come across as surprise or emphasis rather than glee! If it helps better interpret my posts, these days my frame of mind is "We're living in very interesting times" in the sense of https://en.wikipedia.org/wiki/May_you_live_in_interesting_ti...

It’s all so interesting!

Have fun watching the car crash.

Re: Covering electricity price increases from our data centers

#103

Post by pmontra »

digitalPhonix wrote: post #84
pmontra wrote: post #38

Earlier quoted context omitted.

100 W is only the start. Let's say that I consume 100 W all along the day. I use an LLM for coding assistance in the old way of asking questions and copy pasting code. It's much faster than me at writing that code. I don't think it ever works 1 hour for me per day. It's probably 10 cumulative minutes, probably much less. Round it up to 12 minutes to make it 1/5 of a hour or round it down to 6 minutes for a 1/10. So i…

If you as a human are coding for 24 hours a day as the benchmark for LLM efficiency we have other problems.

I believe that we consume 100 W on average no matter what we do, except intense physical activities, which consume more.

Re: Covering electricity price increases from our data centers

#104

Post by Deanallen »

deaux wrote: post #65
venk12 wrote: post #56

We've moved past asking where the energy comes from or how our planet will survive this critical phase. These days, it's about framing - every country is scrambling to up their game just to stay in power. The companies that are riding this wave are spending millions in marketing, lobbying and billions on consuming energy so that they can make trillions in valuation. I am also an ardent user of AI - but sometimes I do…

> I am also an ardent user of AI - but sometimes I do feel guilty when I use so many tokens - because I know I am burning energy, and feeding part of this mission. If there is a solution, I would like to be a part of it. This is by far the best article I've seen on it [0]. Which leads me to conclude: if you use coding agents, then yes, it's definitely a concern. Yet if you drive daily, even an EV, it's very small com…

It’s not clear from the article, but do these statistics taken into account the amount of electricity and water required to train the model in addition to inference?

For example, the article says their daily average use of Claude code is similar to the dishwasher running. Is that just including inference or also training Opus 4.5?

Re: Covering electricity price increases from our data centers

#106

Post by digitalPhonix »

pmontra wrote: post #103
digitalPhonix wrote: post #84

Earlier quoted context omitted.

If you as a human are coding for 24 hours a day as the benchmark for LLM efficiency we have other problems.

I believe that we consume 100 W on average no matter what we do, except intense physical activities, which consume more.

Right, but you do stuff other than work 24 hours a day, right? You have fun, relax, etc.

Counting the 100w for 24 hours for a human doesn’t match up with counting the power usage from “AI” for only the 10 minutes it’s doing a task.

Also - units issue: 100 watts for a day is 2400 watt-hours. It’s a moot point anyway because the power draw for the frontier models is an order of magnitude off that the division by 24 is basically meaningless.

Re: Covering electricity price increases from our data centers

#107

Post by deaux »

Deanallen wrote: post #104
deaux wrote: post #65

Earlier quoted context omitted.

> I am also an ardent user of AI - but sometimes I do feel guilty when I use so many tokens - because I know I am burning energy, and feeding part of this mission. If there is a solution, I would like to be a part of it. This is by far the best article I've seen on it [0]. Which leads me to conclude: if you use coding agents, then yes, it's definitely a concern. Yet if you drive daily, even an EV, it's very small com…

It’s not clear from the article, but do these statistics taken into account the amount of electricity and water required to train the model in addition to inference? For example, the article says their daily average use of Claude code is similar to the dishwasher running. Is that just including inference or also training Opus 4.5?

This is a great question. To my understanding the industry consensus is that for the big three providers, energy spend on inference had already surpassed training by summer last year, and the former's share only keeps increasing. The problem is that there's no hard data in public.

What we need to do here is write an article that makes a wild claim in either direction ("99% is inference!"), post it on HN, and wait for the comments to roll in that prove it right or wrong.

Re: Covering electricity price increases from our data centers

#108

Post by delaminator »

DoctorOetker wrote: post #95
delaminator wrote: post #67

Earlier quoted context omitted.

Yes I can, on utilitarian grounds. We do it all the time, sending data centers to space is not zero cost externality.

Please continue: 1. humans use soda cans, that people could do without for 100k years 2. ??? 3. Don't put datacenters into space?

"aluminium smelting uses a lot of electricity we should do that"

"data centers use a lot of electricity we shouldn't do that"

"why?"

"computers are different"

Re: Covering electricity price increases from our data centers

#109

Post by DoctorOetker »

rsync wrote: post #98
DoctorOetker wrote: post #49

Earlier quoted context omitted.

I wish HN would have a downvote penalty mechanism, where the number of replies on comments you downvote should be at least x percent, say 50%. Oh no the burden of actually explaining why you want to de-emphasize a comment.

Let me explain to you that I downvoted this comment because you complained about your downvotes. Please don't interrupt the discussion to meta-discuss the scoring system.

look at the timestamps, when I wrote that comment there was no discussion at all, only the downvote and my request for discussion... which worked

but now you say I "interrupted" the back-then-non-existent discussion... whatever rsync, whatever...

Re: Covering electricity price increases from our data centers

#110

Post by DoctorOetker »

delaminator wrote: post #108
DoctorOetker wrote: post #95

Earlier quoted context omitted.

Please continue: 1. humans use soda cans, that people could do without for 100k years 2. ??? 3. Don't put datacenters into space?

"aluminium smelting uses a lot of electricity we should do that" "data centers use a lot of electricity we shouldn't do that" "why?" "computers are different"

for the simple reason that other energy consuming industries have physical products in and physical products out, which is costly to transport from space, while uploading extra corpus tokens and downloading new weights is essentially free from space (compared to the hardware shipping costs).

people have been correctly indoctrinated about global warming and the dominant heating terms coming from excess CO2 concentration, but because of this over-emphasis they neglect the prompt heating that comes with nearly all energy generation mechanisms (from fossil fuels, to solar panels to nuclear energy).

when the whole world starts raising their living conditions, and when a computational race erupts, there is no taming of total human energy consumption.

but what we can do is offload the bulk of computational energy consumption, like training common goods such as LLM weights...

Post reply on HN