3mo ago

We did the math on AI’s energy footprint. Here’s the story you haven’t heard.

www.technologyreview.com We did the math on AI’s energy footprint. Here’s the story you haven’t heard.

The emissions from individual AI text, image, and video queries seem small—until you add up what the industry isn’t tracking and consider where it’s heading next.

Absolutely needed: to get high efficiency for this beast ... as it gets better, we'll become too dependent.

"all of this growth is for a new technology that’s still finding its footing, and in many applications—education, medical advice, legal analysis—might be the wrong tool for the job,,,"

156 comments

as it gets better
Bold assumption.
- Historically AI always got much better. Usually after the field collapsed in an AI winter and several years went by in search for a new technique to then repeat the hype cycle. Tech bros want it to get better without that winter stage though.
  
  AI usually got better when people realized it wasn't going to do all it was hyped up for but was useful for a certain set of tasks.
  Then it turned from world-changing hotness to super boring tech your washing machine uses to fine-tune its washing program.
  
  The spice must flow
  
  Each winter marks the beginning and end of a generation of AI. We are now seeing more progress and as long as there is no technical limit it seems that its progress will not be interrupted.
  
  The issue this time around is infrastructure. The current AI Summer depends on massive datacenters with equally massive electrical needs. If companies can't monetize that enough, they'll pull the plug and none of this will be available to general public anymore.
  This system can go backwards. Yes, the R&D will still be there after the AI Winter cycle hits, but none of the infrastructure.
  
  That's part of why they installed Donald Trump as the dictator of the United States. The other is the network states.
  
  Historically "AI" still doesn't exist.
- Yeah, I think there was some efforts, until we found out that adding billions of parameters to a model would allow both to write the useless part in emails that nobody reads and to strip out the useless part in emails that nobody reads.
  
  I want my emails to be a series of noises that only computers can hear and communicate with

The energy issue almost feels like a red herring for distracting all idiots from actual AI problems and lemmy is just gobbling it up every day. It's so tiring.
- That's because it IS an issue, together with many other issues like disinformation, over reliance, wrong tools for wrong (most) jobs, etc.
  
  Ypu know what I don't hear on lemmy? People complaining that crypto world consumes more energy than AI world and one of those is far more useless in grand scheme of things.
  So how comes it is A issue for AI, but everyone seemingly has forgotten about crypto?
  Last I heard, securing one transaction on chain is equalivent to powering US household for many days (feel free to fact check). In comparison, generating LLM text for entire hour on your PC is pretty much the same as gaming for two hours (very approx., your gpu is unlikely at 100% load), which means gaming world is far more destructive energy wise. Are you getting triggered yet?
- lemmy is just gobbling it up every day. It's so tiring.
  Are you fucking serious? All I ever see on Lemmy is prople saying "AI slop" over and over and over and over again... in like every comment section of every post. It could be a picture that was actually hand-drawn, or a photograph that was definitely not AI, or articles written by someone "sounding like AI". The AI hate on Lemmy is WAY overpowering any support.
  
  I think you misunderstood me here as we're in agreement already
- Partly, yep. Seems like every time I try to pin down an AI on a detail of a question worth asking - a math question, or a date in history, it'll confidently reply with the first answer it finds ... right or wrong.
  
  I don't think accuracy is an issue either. I've been on the web since inception and we always had a terribly inaccurate information landscape. It's really about individual ability to put together found information to an accurate world model and LLMs is a tool just like any other.
  The real issues imo are effects on society be it information manipulation, breaking our education and workforce systems. But all of that is overshadowed by meme issues like energy use or inaccuracy as these are easy to understand for any person while sociology, politics and macro economics are really hard.

How does crypto mining play into all of the electrical need? I know they used to use a butt load.
- I found this article from last year: https://www.eia.gov/todayinenergy/detail.php?id=61364
  Our preliminary estimates suggest that annual electricity use from cryptocurrency mining probably represents from 0.6% to 2.3% of U.S. electricity consumption.
  The wide range should not be too surprising, it's a mess to keep track of, especially with the current administration. Since then, with Trump immediately pledging to support the "industry", I can only imagine it consuming even more now.
  
  That's a huge amount of electricity even at it's lowest. Are they building the AI to crypto mine is also another question. I could see these sneaky bastards combining the two somehow.
- It should be clarified that it's 99.99% Bitcoin mining that's wasting all that energy, any other crypto that still uses mining is basically irrelevant when compared to it

Solar powered server farms in space. Self-powered, self-cooling, 'outside the environment'. Is this a stupid idea?
Edit: So it would seem the answer is yes. Good chat :) Thanks.
- Launch cost is astronomical.
  Maintenance access is horrible.
  Temperature delta is insane, upto 250C.
- I don’t understand the self-cooling. Isn’t it harder to keep things cool in space since there is no conduction or convection cooling? I mean everything is in a vacuum. The only place for heat to go is radiative and that’s terribly inefficient. Seems like a massive engineering problem.
  
  It is, infrared radiators weight a shit ton and are inefficient, big and unwieldy. Still the only viable option for cooling in space. AI would take an hugemongous square footage of it just so the GPUs won't melt.
- You can cool servers way better on Earth than you can in space. Down here you can transfer heat away from the server with conduction and convection, but in space you really only have radiation. Cooling spacecraft is an engineering challenge. One might imagine a server stuck inside a glass thermos that's sitting out in the sun.
- If the end goal is so little Timmy can ask a robot if nazis exist and it spits out misinformation or so Ai bots can flood social media with endless regurgitated bullshit, then no, it's just more garbage in space.
  Ai is interesting,... necessary? A lot of people can be fed and housed for the cost of giant, experimental solar powered Ai computers in space so that they have more excuses not to pay people a living wage.
  
  When I think about the potential of AI, I like to think of Iain M. Banks' Culture more than Skynet. We could probably all live in a post-scarcity society even without AI if we put our minds into it, but let's free ourselves of unnecessary or unwilling labour while we're at it, eh?
- Afaik space isn't self cooling. Overheating of spacecraft is a thing. I think they can only cool through infrared radiation or something.
- https://www.computerworld.com/article/3990472/china-takes-edge-computing-to-orbit-with-first-space-based-processing-network.html
  So someone thinks it's not entirely stupid
- Do you know how much energy you need to launch a kilogram into Earth orbit?
- very stupid. One of the most difficult things in space is cooling stuff. Sending up a bunch of space heaters in a box (almost all of the energy pumped into a computer is turned into heat. The actual computatiion takes next to nothing in comparison) is definitely not a good idea. Definitely not one thought up by a technical person.
- I assume yes, I know very little but I know space is very hard and harsh environment. Also it would be very expensive I assume. And it would need to be big.
- also, you can make computers much more cheap and reliable, more maintainable and much much faster, if you protect them from space radiation by operating them down here, under the protection of earth's atmosphere.

Its worth it for school essays and prawn jesus though.

Does the article answer the question of what is the footprint of a prompt?
- Depends on the prompt, the model, the parameters, which DCs, time of day, location in the world, and other factors. They answer the question but there's so many variables that can affect footprint (and big hyperscalers do not release this data so you have to under a lot)
- Basically nothing worth getting angry about

The matrix is getting more and more real every day

Fire bad, who cook with fire, fire burn, fire pollute, fire baaaaad
- Yes, "AI" is literally contributing to the burning of the planet.
  https://www.cleanairfund.org/news-item/wildfires-climate-change-and-air-pollution-a-vicious-cycle/
  
  So is the computer you're using
- Nothing wrong with examining potential issues for emerging technology before they become actual issues.
  
  For sure, and I agree. But that isn't what happens around here. Instead we turn to panic rather than skepticism. We are cynical more than anything. And I don't trust cynicism about topical subjects.
  This wasn't well reasoned objectivism. It was journalist and artist fearing for their jobs about technology they don't understand. They generate a lot of content to generate panic. The mob saw the panic and adopted it. You're not a true lefty unless you accept that AI is some new danger to the threat of the lowly creative.
- What’s your point here?
  
  They're trying to compare "AI" to fire. If you don't see the point, I can't blame you.

At some point, someone said the same thing about:
electricity
books
cars
computers
medicine
houses
Is this /c/technology or /c/anti_technology because it’s hard to tell most of the time.
- A better analogy for AI is the discovery of asbestos or the invention of single-use plastics. Terrible fucking idea.
  
  I think it’s probably a bit early to tell for certain on that assessment. There is definitely pros and cons to all technology. Electricity production causes environmental damage, building wooden houses require logging. Plastics are a byproduct of a withering industry. Asbestos might have saved more lives than it took, but there were probably much better ways to solve fire resistant buildings.
  Why all these destructive things? Capitalism requires maximizing profits above all else. So, really the question is how will capitalism fuck us over with AI? So, so many ways. That’s why it’s important that we build community understanding of this technology in order to combat it. It’s not going away. It’s here to stay. So we either put our heads in the sand and pretend it’s not here or we can embrace it and learn how it works and how to defeat it and come up with open source tooling to combat it.
  I’m in the latter camp. I love technology breakthroughs and want to learn first hand the capabilities to understand how it will be used against me and how I can use it.
  
  Well, it's a bit better than that, simply because you can train AI with solar power. Probably nobody does that currently, as it's easier, faster-to-market and probably (for whatever corrupt reason) cheaper for business to let it run on burning fossils/nuclear. Currently there's an insane amount of waste, often 1000s of models are trained and only the best performing one is deployed - and then it's just a fancy autocomplete. The better use is for prediction of material failure, new medicine and protein folding, generally improved processes.
  With asbestos you get some convenience, but it'll be for eternity a pain to find a waste management facility that will accept it.
  
  Plastics are great, what are you smoking, plastics?
- I'm genuinely excited about the possibilities of AI, just not in the hands of a bunch of self-serving, amoral cunts.
  
  I completely agree. However the genie is out of the bottle. Not much we can do to prevent it at this point, but there is plenty we can do to learn about it and defend against is abuse against us.
- I don't think books ever had the same amount of discussion of how they impact our global carbon footprint, and where it comes to "houses" - I doubt people in the neolithic said about their new invention what is being discussed with AI. It is a disingenuous comparison. (And sure, someone somewhere may have said something like that about basically anything, but usually not a large part of professionals from within the field, like is the case with AI.)
  This is also not simply Ludditism, the nature of how AI is used currently goes far beyond where it is genuinely useful in a case of investor hype FOMO, and the hidden costs for our efforts against climate change are real, as are the problems for creatives - who sadly need a lot of the "bullshit work" that AI can substitute to survive while honing their craft - as is the quality drop in journalism, as are fundamental questions about how far generative AI models can truly evolve in quality for the massive amount of energy invested, so the usual "just wait until the tech gets better" is not the easy way out to justify draining said energy (and fresh water) on top of what crypto mining has been wasting with data centres in the past years.
  Now, those problems aren't simply problems of the technology, but also of how that technology manifests within market dynamics. But the technology still is not just neutral, and even if we view it as an inevitability, that inevitability does not have to manifest without regulation and within the context of hyped, often unwanted application to basically everything.
  Without mechanisms to address problems and to enforce regulation, in lieu of fundamental changes to what market/investment dynamics demand, this is indeed a very questionable technology at this point. And also: To truly love something abstract, like "technology", means being able to - sometimes harshly - criticise it. Think the meme of a "tech bro" with a fully automated house vs the IT guy who barely has tech stuff beyond their PC and some stuff tinkered on passionately in their own time.
  
  I’m not sure regulation is going to be an open for this in the US anytime soon. Maybe EU can show us the way.
- Cars are literally privileged garbage that's destroying the planet. Great comparison on that one.
  Is this /c/technology or /c/anti_technology because it’s hard to tell most of the time.
  Well only one of those is allowed to exist so you figure it out.
  https://en.wikipedia.org/wiki/Scientism
- What benefit for society does this crop of large language models and generative "AI" offer?
  We already see students use it for homework, meaning they don't learn their stuff.
  We also see people treat the output of LLMs as gospel truth, despite the fact that LLMs often hallucinate complete BS!
  LLMs and generative "AI" rely on stolen artwork. Which is a net negative for society.
  
  Some people think you can use it as a replacement for therapy or to fight loneliness. Turns out, simply reading fiction is better.
  https://neurosciencenews.com/reading-emapthy-loneliness-28972/
  
  Well for me, I enjoy pair programming my own projects with offline models. I also bounce ideas off it to attempt to ground myself in some type of reality (some models are better for this than others… probably has risk of delusions of grandeur. Some models will just verbally suck you off which is annoying).
  I built ansible tooling for deploying k3s kubernetes and Ceph-backed Proxmox clusters and VMs and containers and services. Utilized the help of LLMs to structure my playbooks and figure out how roles work.
  I love learning new things and LLMs have a lot to offer in that regard. You have to watch out for the bullshit and independently look at other sources as well, but it’s a great starting point and I can sometimes have sone deep conversations around some topics.
- You might have heard of these fossil fuels we're busily running out of. And fossil is still 80% of primary energy use so there is no renewable energy transition, and renewable infrastructure is being built almost exclusively using fossil fuels.
  So this means future energy rationing. What's the business case for AI?
  
  What's the business case for AI?
  printing money by deception. oh and also votes, probably.
  
  Restricting our energy use is not a very good end game. We need to learn how to unlock more energy production without destruction of the environment. This will happen through technological development. Temporarily rationing or conservation may be needed, but permanent is not the answer.
- When AI was getting popular, the media released an absolute war against it. A lot of us are swayed by what the media tell us
  
  That’s probably going to be one of the main weapons against us, the media generating fake news using AI to control us.
- Is this /c/technology or /c/anti_technology because it’s hard to tell most of the time.
  People here are generally anti-anything. That's what echo chambers are for.
  
  I am biased, I am having a ton of fun with LLMs and they are helping me achieve some personal goals. Do they use energy? Sure. Will new, more powerful technologies come along later that require even greater amounts of energy? I hope so one day. We need to find cleaner more abundant energy sources.
  
  It's much better to be a critical thinker than mindlessly accepting whatever BS from some grifter just because it's "accepted wisdom" in a completely brainwashed society.
  
  Sounds like most of Lemmy. Honestly sometimes I feel it’s worse than Reddit with the constant bashing on anything except Linux, Firefox, or - for some reason - Steam. Still glad I left Reddit though.

156 comments