this post was submitted on 05 Nov 2023
2 points (51.7% liked)
Technology
59217 readers
2864 users here now
This is a most excellent place for technology news and articles.
Our Rules
- Follow the lemmy.world rules.
- Only tech related content.
- Be excellent to each another!
- Mod approved content bots can post up to 10 articles per day.
- Threads asking for personal tech support may be deleted.
- Politics threads may be removed.
- No memes allowed as posts, OK to post as comments.
- Only approved bots from the list below, to ask if your bot can be added please contact us.
- Check for duplicates before posting, duplicates may be removed
Approved Bots
founded 1 year ago
MODERATORS
you are viewing a single comment's thread
view the rest of the comments
view the rest of the comments
How much electricity was used to train Copilot? How much MORE is going to be used in the future.
Feels to me like you don't understand the problem set and you're just impressed by a tool spitting out guesses based on millions of examples it hoovered up.
Oh no, electricity! If only there were some way to generate more of it.
This "it uses electricity" thing is such a weird objection. Yes, it uses electricity. That's why it costs money to run. People pay that money to run it, and if it wasn't helpful enough to be worth that money they wouldn't pay it.
Yeah, for those of you that don't know your ass from your elbow, these systems are predicted to reach 1 gigawatt per data center up from 50 to 75MW, 100 at the peak. So a 10 to 20 times increase in power, now, I don't know where you think we're going to get 10 to 20 times. More power for every single built data center, but you're smoking crack if you think it's reasonable.
Not only that, but there's this little issue we've been noticing for the past 100 years called climate change. Have you heard of it? It's truly idiotic to consider increasing the demands of these data centers by 10 to 20 times while we're talking about complete global catastrophe within 50 to 100 years. Monumentally stupid shit.
And then, of course, we have the people that don't understand how electronics work. People that might drive by and say will reduce the amount of power these systems need. No, we won't. We will reduce the amount of joules per operation, but will increase the number of operations drastically. Thereby, causing the power demand to increase. These numbers aren't for me, they're from actual industry insiders designing the far future generations of these products.
Nice attempt with a snark, you've proven. You don't know what you're talking about. Thank you for playing.
As I said, yes, it uses electricity. You realize that there are ways to generate electricity that don't contribute to global warming? We're going to need to be switching to those methods anyway.
You seem utterly confused about the scale of the problem I described. Which isn't entirely surprising. But I think you should go look up those sources. Because the output of a good sized nuclear station is about 1GW and we aren't going to be building a nuclear station next to every single datacenter, now are we...
I fail to see any problem here at all.
It's really quite simple. If AI is useful enough that people are willing to pay for the electricity it consumes, then they will pay for that electricity and the generating capacity will be funded by that. If it's not useful enough for people to be willing to pay for the electricity, then the AI won't be run. This is a trivial supply and demand situation. The AIs won't use "too much electricity" because nobody's going to want to pay for that.
So if you point at an AI and exclaim "it's using a kajillion dollars worth of electricity!" I'll shrug and say "it must be providing a kajillion dollars worth of services, otherwise who's paying for it?"
You fail to see a problem with increasing power demands by 10-20x beyond their existing consumption rates? While the world burns around you? Alright Niro, enjoy your fiddle.
Let me change your sentence, then you try it on for size and see how you like it.
"If CFCs are useful enough that people are willing to pay for them, then they will pay for those CFCs and the hole in the ozone will be an acceptable consequence". I could go on with Asbestos, lead in gasoline, literally anything that releases a greenhouse gas.
And again, you clearly can not conceive of what you're talking about. The cost for such generation is beyond reasonable, and you've entirely missed that point. Not a surprise, really, but you've missed it all the same. Guessing the next word isn't useful enough to humanity to burn the world to the ground, but it IS something that companies can sell to simple rubes that have been conned into thinking that the illusion is real magic. And we know what companies will do for money.
It isn't. Because as it is already, these systems are behemoths that consume insane amounts of energy, they are not making enough money to pay for themselves, they are not serving a real utility that provides value, and still the drooling masses use them for their amusement. Either way, you've proven you don't understand the technical aspects of this, the consumption aspects of it, the as-implemented state of the industry, or the scale of demand induced by companies trying to make a buck. So I think the value of this conversation is about the value of any of these rube goldberg guessing machines.
Do you have any sources to back up that asspull? Seriously, you have no idea what you're talking about here. How would we suddenly build ten to twenty times as many power generators as we currently have, even with infinite money being thrown at the problem?
Again, do you not realize there are environmentally friendly ways to produce electricity?
Then how are the people who are running them able to keep running them?
Directly from a keynote given at OpenCompute a few weeks ago. Swing and a miss.
List for me which ones can be built in 1GW installations feasibly and cost effectively, please. And if you choose solar, detail the physical size of the facility. I'll wait here.
Do you forget my initial comment? Go read it again.
I work on AI and it feels to me like you literally don't understand it at all based on your comments in this thread. But you sure do have all the buzzwords down pat.
I would love an example of something I got wrong.
He thinks LLMs could be replaced by a "text template", so yeah, this guy's clearly not actually tried using it for anything meaningful before.
You're right, a template would be more specific to the question and guaranteed accurate, while not taking GPU years to train and untold quantities of stolen content. So, I guess a template would be a much better solution.