Pretty terrible metaphor as realistically even without AI we will never need less compute. Compute is awesome, it’s a good thing to have more. The problem is the environmental damage not anything else.
That makes the metaphor imperfect, but the data centers are being paid with circular money that isn’t there, based on the assumption that AI will make obscene amounts of money to actually pay their bills when currently they’re losing lots of investor money to subsidize the price, and they’re already talking about transitioning to a usage based system, which might push some enterprises to use a FOSS alternative from China.
We’re in a bubble, when it pops it’s likely many data center projects will be abandoned before they’re done.
isn’t that a good thing though? let billionaires lose money then just don’t bail them out. Either way the datacenters will remain useful and might actually benefit real work. This is btw how we got a lot of open technologies where corporate overinvests and then community takes over.
Considering the MIC is in bed with AI corporations I worry they will get bailed out to some extent, and that the gains will remain completely private, with taxpayers being expected to foot the bill (or alternatively, dealing with the inflation while their wages remain stagnant). I hope you’re right though, I really want there to be a positive in the end, I’m just skeptical on this front.
The silver lining to me is that there’s encouraging developments in FOSS AI, developed mainly in China. Maybe after the crash we can get somewhat cheaper GPUs to run them locally.
I think it’ll be really hard to bail AI out on public funds with current reputation if bubble pops.
However I think real risk is not full bail but constant leeching of public funds to private enterprise. This is not new to AI and this datacenter issue is exactly that - have people bear the environmental costs for private gains of datacenters. The other argument that US “doesn’t need datacenters” is just bad - that’s not the issue, everyone will need more datacenters forever basically.
encouraging developments in FOSS AI, developed mainly in China.
I wouldn’t really trust China’s open source commitments though. FOSS is non-existent in chinese culture and it’s directly a competitive response to american corporate models. That being said, a good deed done for bad reasons still can be good!
Honestly if AI would gracefully decline now and we just had open source models we’d still have work for the next 10 years implementing these tools.
Ideally, we wouldn’t build infinite compute, just as much as we are actually using, and using it efficiently allows you to build less. We would still need datacenters even without LLMs, but they wouldn’t need to be so gargantuan, because even the worst, inefficient, nodejs-based, intern-written server you could ever encounter, would be heaps more efficient (or at least less demanding) than any LLM.
This is true even from an economical point of view, or any practical point of view, not just environmental.
To quote Tannenbaum: “You know you have the right computer when you are always using 99% of it. If you are using 100%, you are being limited by the machine. If you are using 98% you have bought more than you need”. If datacenters were always running at 40%, we would build bigger ones.
Pretty terrible metaphor as realistically even without AI we will never need less compute. Compute is awesome, it’s a good thing to have more. The problem is the environmental damage not anything else.
That makes the metaphor imperfect, but the data centers are being paid with circular money that isn’t there, based on the assumption that AI will make obscene amounts of money to actually pay their bills when currently they’re losing lots of investor money to subsidize the price, and they’re already talking about transitioning to a usage based system, which might push some enterprises to use a FOSS alternative from China.
We’re in a bubble, when it pops it’s likely many data center projects will be abandoned before they’re done.
isn’t that a good thing though? let billionaires lose money then just don’t bail them out. Either way the datacenters will remain useful and might actually benefit real work. This is btw how we got a lot of open technologies where corporate overinvests and then community takes over.
Considering the MIC is in bed with AI corporations I worry they will get bailed out to some extent, and that the gains will remain completely private, with taxpayers being expected to foot the bill (or alternatively, dealing with the inflation while their wages remain stagnant). I hope you’re right though, I really want there to be a positive in the end, I’m just skeptical on this front.
The silver lining to me is that there’s encouraging developments in FOSS AI, developed mainly in China. Maybe after the crash we can get somewhat cheaper GPUs to run them locally.
I think it’ll be really hard to bail AI out on public funds with current reputation if bubble pops.
However I think real risk is not full bail but constant leeching of public funds to private enterprise. This is not new to AI and this datacenter issue is exactly that - have people bear the environmental costs for private gains of datacenters. The other argument that US “doesn’t need datacenters” is just bad - that’s not the issue, everyone will need more datacenters forever basically.
I wouldn’t really trust China’s open source commitments though. FOSS is non-existent in chinese culture and it’s directly a competitive response to american corporate models. That being said, a good deed done for bad reasons still can be good!
Honestly if AI would gracefully decline now and we just had open source models we’d still have work for the next 10 years implementing these tools.
Historically speaking it’s most often your average citizen that pays for it in the end.
Ideally, we wouldn’t build infinite compute, just as much as we are actually using, and using it efficiently allows you to build less. We would still need datacenters even without LLMs, but they wouldn’t need to be so gargantuan, because even the worst, inefficient, nodejs-based, intern-written server you could ever encounter, would be heaps more efficient (or at least less demanding) than any LLM. This is true even from an economical point of view, or any practical point of view, not just environmental.
To quote Tannenbaum: “You know you have the right computer when you are always using 99% of it. If you are using 100%, you are being limited by the machine. If you are using 98% you have bought more than you need”. If datacenters were always running at 40%, we would build bigger ones.
It’d be like if the railroad went through the canal