Blinky approves.
Blinky approves.
In reply to Noddaz :
If anything is going to really kill AI, it's going to be it's huge massive monstorous power demand to run it. Which kind of confirms my initial observation of it that the code is incredibly inefficient back when we worked with the bones of it.
Yup. All that power for AI and data centers has to come from somewhere, and it damn sure ain't coming from windmills and cow farts.
I don't know why I read that at first as "AI will run Three Mile Island," I was really concerned for a second.
Colin Wood said:I don't know why I read that at first as "AI will run Three Mile Island," I was really concerned for a second.
Indeed. Far better that we get Homer Simpson to do it! 
I'm a TRIZ (Theory of Inventive Problem Solving) enthusiast and one of its tenets is to solve big problems by imagining them as massive problems.
The question "how do we provide enough energy to power the world" may best be answered by asking "how do we provide 10 or 100 or 1,000 times more power than the earth needs".
Assuming AI (and quantum computing for that matter) will be the new high ground, it is absolutely imperative that we remain in the lead and that means that we must either reduce the power requirements or increase the power availability or both.
The consequence of a nuclear reactor event is trivial compared to the consequence of losing the AI / quantum computing race.
In reply to RX Reven' :
Quantum computing doesn't need massive amounts of electricity - not in general or necessarily per-computer, although they do have cryogenic cooling systems and deep-vacuum systems the computers themselves use very little electricity.
"Winning" the AI race will most likely have very negative consequences too. It's far less likely to produce something like Star Trek than something like Elysium or even The Terminator. Regardless of whichever country produces any such outcome, approximately everyone loses massively.
Edit: Even producing today's LLMs with massive amounts of electricity is the easy-button ham-fisted approach that companies take because energy is too cheap to deter them from taking it. It could be done with vastly less energy using more specialized hardware than GPUs, which are the second least-specialized option possible next to an ordinary CPU.
In reply to GameboyRMH :
Agreed...we're on the eve of opening a massive can of unintended consequences.
If the free world is in the lead, there's a 99% chance that this movie ends badly...if some other entity is in the lead, it's game over.
BTW, I only included quantum computing because it's part of the information race.
That's a boat ton of power. 835mw per reactor. Wonder if they'll scale it back a bit. Then again, AI is incredibly power hungry.
I'm also stoked on the idea of private companies have private sources of energy that still get Federal oversight, but it's sorta like using public money for private schools - we can't ride on the back of big funders for the benefit of greater good.
I love to see the return of nuclear power, but I would also love to see it done with something that wasn't designed in the 60's.
We have a nuke in CA (near Santa Barbara) that is supposed to be shut down, but has been extended because it is too useful. I doubt it has anything to do with AI power needs since you would have to be an idiot to put a server farm in CA with it's electricity prices. Electric cars though... that might be increasing the draw a bit.
Feds Give OK for Diablo Canyon Nuclear Power Plant to Operate Past 2025
Decision by the U.S. Nuclear Regulatory Commission means the plant in San Luis Obispo County could remain in operation as much as another 20 years
https://www.noozhawk.com/feds-give-ok-for-diablo-canyon-nuclear-power-plant-to-operate-past-2025/
It's in a pretty impressive location (if you are wondering, the San Andreas fault is well to the east of it, and I believe it's pretty earthquake safe):
In reply to aircooled :
How tsunami proof is it?
aircooled said:I love to see the return of nuclear power, but I would also love to see it done with something that wasn't designed in the 60's.
They have just that at Vogtle in Georgia. AP1000 design that's from much later than the 1960s. Just came online in the last year or so.
Just don't ask how far over budget it was...
In reply to aircooled :
For some reason, I prefer the aesthetics of the San Onofre reactor
Colin Wood said:I don't know why I read that at first as "AI will run Three Mile Island," I was really concerned for a second.
I did the same thing. And that would be part of the reason I posted it. AI controlling it's own power. What could go wrong?
Noddaz said:
AI controlling it's own power. What could go wrong?
I for one would like to welcome our new man-made silicon overlords
RX Reven' said:In reply to aircooled :
For some reason, I prefer the aesthetics of the San Onofre reactor
Lt Drebin agrees
We could, you know, just not.
I don't have a problem with repowering the existing infrastructure, it's more the use of the energy.
This is exactly like all the crypto dweebs claiming it's ok if they used solar/wind/hydro. No it really isn't. That clean power installation has an extraction cost to build out and utilizes an opportunity cost that would otherwise be used to offset hydrocarbon based plants and peakers.
Same thing here.
Microsoft also proposed a data center utilizing 6 of the NuScale modular reactors. Again, if you're going to consume that opportunity cost (nuclear is expensive), what benefit does AI bring vs replacing dirty sources? Besides speculative stock prices.
The0retical said:We could, you know, just not.
I'm not certain AI is the humanity changing tech it's claimed to be but I think it's probable that being behind will put us in a position that's somewhere between highly undesirable and disastrous.
We experienced the same conundrum with nuclear weapons..."it'd be far better to not build these things but if we don't, someone else will and then we'll totally be their b**ch so staaaaart building."
Whatever happened to those new age Nukes being constructed in Georgia and South Carolina, billions over budget and/or canceled?
In reply to RX Reven' :
I'll be honest I'm pretty cynical about that kind of language as well. The US is always experiencing some kind of technology gap according to think tanks, and it often enables the worst kind of behaviors. Much like the missile gap you noted, there's also the small drone gap, the stand off missile gap, the fighter gap, the satellite weapon gap, the aircraft carrier gap, etc etc and so forth.
This is very much a "If we don't build the Torment Nexus then [insert adversary] will build the Torment Nexus"
Maybe I've been reading too much Noam Chomsky and Gary Wills lately, but it's hard to look at a superpower like the US and claim that it's stumbling and bumbling its way through history always at the disadvantage.
I'm certainly not against specialist AI's, they're pretty awesome in the medical and certain engineering fields, but I fail to see how today's LLM's are going to make it to AGI if we just feed them more power and data.
My experience in the tech world is the current phase of AI causes more problems than it's worth. Sure, it can write ok small snippets of code, but ask it for a project of even low complexity, and it'll produce something that, by all appearances, should work but doesn't. You spend more time figuring out what the hell it was doing than you would if you just wrote the damn thing yourself. God help you if you ask it to work within an existing framework.
You find the same thing with having it write essays. Sure, it'll produce something passable, but it doesn't produce anything worth putting in front of a human. It's just a fancy word prediction algorithm.
It has its merits for some general uses. I use AI tools to augment my writing, brainstorm, and identify gaps but I'm really not convinced throwing more resources at it would make it "better."
Puddy46 said:aircooled said:I love to see the return of nuclear power, but I would also love to see it done with something that wasn't designed in the 60's.
They have just that at Vogtle in Georgia. AP1000 design that's from much later than the 1960s. Just came online in the last year or so.
Just don't ask how far over budget it was...
all the over budget
In reply to RX Reven' :
If you are talking about truly advanced computing, where super efficient code is able to do things that we never thought about in a way that takes less power than we do now, sure. But I'm not so sure about what we currently call "AI". That's just a boat load of statistical analysis on data done in a manner that may or many not be useful. Listen to the ads of what they are projecting- it's just more of the same statistical process that early machine learning was using. Seeing that it uses a massive amount of energy for what it delivers, I'm not so sure it's really that important.
Again, I am VERY much for advanced computing and new processes. And I really think the current auto coding things we use is a massive step back in computing. Instead of coming up with actual code that needs compiled, "coders" use graphical methods to connect ideas together and let the system make the code. Right now, I really think this is how "AI" code is written- which is a huge reason why it's so power hungry.
But that's my opinion based on how I saw automotive code change over the last 15 years on top of using some machine learning at the end of my career. I really wish I saw some indicator that things have changed for the better- but I just don't see it, yet.
The closest thing I've seen to code progress are the ARM M* chipped Apple computers, especially compared to windows on ARM. Modern code should be able to do more with less power given chip capabilities. Again, as I see it.
VolvoHeretic said:Whatever happened to those new age Nukes being constructed in Georgia and South Carolina, billions over budget and/or canceled?
Vogtle 3 came online last year. It was 7 years late and $17 billion over budget. Vogtle 4 came online in April of this year and was another $18 billion over budget. The reactors cost $10,784 per KWh of capacity, making them the most expensive installations per KWh ever brought online.
https://www.reuters.com/business/energy/georgias-new-nuclear-plants-drive-us-power-sector-clean-up-maguire-2024-09-19/
$35 billion buys a lot of batteries, turbines, and panels.
Edit: Actually, the article states that solar or wind would have bought 10x the capacity (for just the over run) at roughly $1000 to $1500 per installed KWh
The0retical said:...Edit: Actually, the article states that solar or wind would have bought 10x the capacity (for just the over run) at roughly $1000 to $1500 per installed KWh
Not an entirely relevant comparison. I am pretty certain that number does not include the massive batteries and infrastructure that would be require to provide the same 24hr adjustable base load a nuke does.
I am sure they were still VERY expensive (even if you throw the battery cost in), but a bit of an apples to oranges comparison. Similar emissions impacts, very different application.
The article seems to give little explanation of why the cost over runs, which I think would be super relevant E.g. if someone else built the same thing now, what would the expected cost be? The planned number, or what it ended up being? Who F'd up?
In reply to aircooled :
That's correct. Batteries are not in the cost in the Reuters comparison. Lfp has crossed the $100 per kWh, though. It looks like the initial budget was $1 billion per reactor in 2006. Not adjusting for inflation, that would have purchased 20 million kWh of batteries or 20 MW. In Georgia, that would power 2,346 homes on average, assuming the worst possible no-power situation, just off the battery bank for an entire day.
You'll need to log in to post.
