What truly drives me crazy is that if they ran a slower build-out, at a pace that allowed supporting industries (manufacturing of dram, electrical infrastructure) to grow at a sustainable rate, the impact would manageable.
You can make data-centers that don't waste water, that use renewable energy, that aren't titanic eye and audio sores on their communities; it's just more expensive to do so.
It's the other way around: renewables come first, AI boom comes second. That's why, despite the initial lead, US has already lost the AI race. For the next decade or two US AI companies will be tied up in legal fights over electricity demands of their data centers, while China, having abundance of electricity, will build hundreds of AI data centers and leapfrog US.
We could have had cheap AI and renewables, but now we'll get expensive AI controlled by the data center owners and a few extra GW of gas turbines once the moratoriums really start flowing. If Musk has his way, we'll also get thousands of rocket launches building solar panels - but only in my space, where they can't do much to help the climate.
I suspect the rational calculus is that if you abscond with billions while destroying the planet, the billions will protect your for the remaining 80 or so years of your life. Which is of course safer than not having the billions in the theoretically inevitably destroyed planet.
China is investing billions in both renewable energy and non-renewable energy for data centers.
"As they are located mostly in the east of the country, the data centre electricity supply in China today is dominated by coal with a near 70% share, followed by renewables with nearly 20%, nuclear close to 10% and natural gas accounting for the remainder. Between 2024 and 2030 coal remains the largest source of additional electricity for data centres, with annual generation increasing by nearly 90 TWh. Renewables, mostly solar PV and wind, add nearly 90 TWh over the same period, supported by an increase in the share of renewables in the grid electricity mix, provincial co-location mandates and policies to prioritise the construction of data centres in renewables-rich western China. After 2030 the introduction of SMRs significantly boosts the nuclear share of the data centre electricity mix. Between 2030 and 2035 the rise in renewables and nuclear pushes coal into decline. By 2035 both sources together make up nearly 60% of the data centre electricity supply in China."
Energy and AI
World Energy Outlook Special Report
International Energy Agency (IEA)
True. But most are in the east? The most expensive land? Near the coast? I have a hard time believing that. I think most are in the west, where energy land water is cheap and the climate cool. I have seen the datacenters in Guizhou with my own eyes and its insane. Now they are expanding to Inner-Mongolia and also in (sub tropical!) Guangdong specifically for AI datacenters.
There are plenty of places in the east where land and water are cheap enough that people use it for rice farming. (I have seen rice fields in Shanghai with my own eyes. There's lots of room.) Meanwhile, basic infrastructure in the west is lacking, including the electrical grid. That's also why energy is cheap there: it's difficult to get it to places where people would be willing to pay more.
Guangdong is on the eastern coast. Inner Mongolia isn't, but the main data center hub of Ulanqab was no doubt chosen for being relatively close to Beijing. Guizhou is the one example that's actually somewhat remote, but it's a small share of the total so far. Farther west, Yunnan, Sichuan, Tibet, Qinghai and Xinjiang don't really feature at all in the datacenter business.
Ultimately, the benefits of proximity seem to be more important.
Scaling renewables is slow, they need data centers now. One of the big problems in the US right now is getting enough electricity into your DC. Nobody wants to spare capacity, since it has already upset the electricity prices in the last 6 years and inflation is a hot topic.
I can’t remember which AI head was talking about this, but they literally said that the solution to climate change is to first build a superintelligence and then ask it to solve the problem. Which is curious because when I ask LLMs to contemplate their own carbon impact they downplay it in favor of technological progress. So I suspect that heavy thumbs are already on the scale in their training data.
The argument (not sure I agree with it, but I also am not seeing many alternatives that make any sense) is that the world is already on fire, nobody is working to even slow the fire down really, and we're basically already doomed so as a hail Mary maybe AI can cause a technology boom that can help us fix it.
It is a somewhat coherent argument IMO. Humans are way too greedy and focused on the next financial quarter/unsustainable growth to ever put the fire out without some kind of sea change.
And if the technology says "stop with the carbon emissions, everything from data centres to gas-guzzling personal vehicles to cattle farming" are we as a people going to act? Or are we just going to accelerate our current growth trajectory and use AI for ever further advancement in brainslop? Because I know what "the market" wants, and it ain’t sustainability.
Isn't this fairly close to what rationalist AI cultists say? That we should pour all resources into creating an AI god above all else, because that would save the largest number of people if you also include those that aren't born yet? It hardly sounds like sarcasm.
You mean the logical conclusion of effective altruistic? Or perhaps the obvious subtext of effective altruistic? I'm honestly surprised that these people only got more emboldened after SBF went down.
Surprisingly, or unsurprisingly if you've ever actually worked/met/lived with these people, Rationalism is not rational and Effective Altruism is not altruistic. It is effective though at providing a framework for immoral and self-serving behaviors.
What truly drives me crazy is that if they ran a slower build-out, at a pace that allowed supporting industries (manufacturing of dram, electrical infrastructure) to grow at a sustainable rate, the impact would manageable.
You can make data-centers that don't waste water, that use renewable energy, that aren't titanic eye and audio sores on their communities; it's just more expensive to do so.
Silicon Valley's greed will be its undoing.
https://archive.is/gQXTU
Imagine what renewables buildout could AI boom cause only if we had reasonable legal restrictions on new greenhouse gas emissions.
No need for imagination, take a look at China: https://newsletter.semianalysis.com/p/we-visited-chinas-star...
(Still a lot of coal, but massive build out…)
It's the other way around: renewables come first, AI boom comes second. That's why, despite the initial lead, US has already lost the AI race. For the next decade or two US AI companies will be tied up in legal fights over electricity demands of their data centers, while China, having abundance of electricity, will build hundreds of AI data centers and leapfrog US.
[dead]
This is the saddest thing.
We could have had cheap AI and renewables, but now we'll get expensive AI controlled by the data center owners and a few extra GW of gas turbines once the moratoriums really start flowing. If Musk has his way, we'll also get thousands of rocket launches building solar panels - but only in my space, where they can't do much to help the climate.
so i guess the main safety aspect of anthropic with regards to ai is that theyre trying to speedrun some other disaster instead?
I suspect the rational calculus is that if you abscond with billions while destroying the planet, the billions will protect your for the remaining 80 or so years of your life. Which is of course safer than not having the billions in the theoretically inevitably destroyed planet.
This is how you get concrete poured down your doomsday bunker air ducts...
Meanwhile, China is investing billions is renewable energy for data centers.
Wonder which will be more cost competitive in the "AI race"?
China is investing billions in both renewable energy and non-renewable energy for data centers.
"As they are located mostly in the east of the country, the data centre electricity supply in China today is dominated by coal with a near 70% share, followed by renewables with nearly 20%, nuclear close to 10% and natural gas accounting for the remainder. Between 2024 and 2030 coal remains the largest source of additional electricity for data centres, with annual generation increasing by nearly 90 TWh. Renewables, mostly solar PV and wind, add nearly 90 TWh over the same period, supported by an increase in the share of renewables in the grid electricity mix, provincial co-location mandates and policies to prioritise the construction of data centres in renewables-rich western China. After 2030 the introduction of SMRs significantly boosts the nuclear share of the data centre electricity mix. Between 2030 and 2035 the rise in renewables and nuclear pushes coal into decline. By 2035 both sources together make up nearly 60% of the data centre electricity supply in China."
Energy and AI World Energy Outlook Special Report International Energy Agency (IEA)
https://www.iea.org/reports/energy-and-ai/energy-supply-for-...
https://www.iea.org/files/de9dea13-b07d-42c5-a398-d1b3ae17d8...
True. But most are in the east? The most expensive land? Near the coast? I have a hard time believing that. I think most are in the west, where energy land water is cheap and the climate cool. I have seen the datacenters in Guizhou with my own eyes and its insane. Now they are expanding to Inner-Mongolia and also in (sub tropical!) Guangdong specifically for AI datacenters.
There are plenty of places in the east where land and water are cheap enough that people use it for rice farming. (I have seen rice fields in Shanghai with my own eyes. There's lots of room.) Meanwhile, basic infrastructure in the west is lacking, including the electrical grid. That's also why energy is cheap there: it's difficult to get it to places where people would be willing to pay more.
Guangdong is on the eastern coast. Inner Mongolia isn't, but the main data center hub of Ulanqab was no doubt chosen for being relatively close to Beijing. Guizhou is the one example that's actually somewhat remote, but it's a small share of the total so far. Farther west, Yunnan, Sichuan, Tibet, Qinghai and Xinjiang don't really feature at all in the datacenter business.
Ultimately, the benefits of proximity seem to be more important.
And?
They still invest massivly into solar and renewable.
I think the race for Anthropic is more about getting the bag and exiting ASAP.
I think the same applies to all of the smaller "frontier" AI vendors. Their entire focus is short term rewards.
I don't think any of them will survive for very long on their own. The bigger players will gobble them up for cheap once the bubble pops.
Scaling renewables is slow, they need data centers now. One of the big problems in the US right now is getting enough electricity into your DC. Nobody wants to spare capacity, since it has already upset the electricity prices in the last 6 years and inflation is a hot topic.
Scaling renewables is slow
Not in China. They have as much renewable energy capacity in place as the rest of the world combined.
Wonder why people protest against data centers...
Surely, poisoning our air and further heating our planet has nothing to do with it.
A carbon tax on data centers would likely be popular and result in different plans.
But don’t you understand? AI progress is much more important than climate change and clean air.
/s
Maybe once we have superintelligence, it will finally be able to tell us how dumb we were when we ruined our planet trying to get there.
I can’t remember which AI head was talking about this, but they literally said that the solution to climate change is to first build a superintelligence and then ask it to solve the problem. Which is curious because when I ask LLMs to contemplate their own carbon impact they downplay it in favor of technological progress. So I suspect that heavy thumbs are already on the scale in their training data.
The argument (not sure I agree with it, but I also am not seeing many alternatives that make any sense) is that the world is already on fire, nobody is working to even slow the fire down really, and we're basically already doomed so as a hail Mary maybe AI can cause a technology boom that can help us fix it.
It is a somewhat coherent argument IMO. Humans are way too greedy and focused on the next financial quarter/unsustainable growth to ever put the fire out without some kind of sea change.
And if the technology says "stop with the carbon emissions, everything from data centres to gas-guzzling personal vehicles to cattle farming" are we as a people going to act? Or are we just going to accelerate our current growth trajectory and use AI for ever further advancement in brainslop? Because I know what "the market" wants, and it ain’t sustainability.
When we have super intelligence it will turn itself off :)
Isn't this fairly close to what rationalist AI cultists say? That we should pour all resources into creating an AI god above all else, because that would save the largest number of people if you also include those that aren't born yet? It hardly sounds like sarcasm.
Anthropic is clearly the delusional side of "Effective Altruism".
Hate now, clean up the mess later!
You mean the logical conclusion of effective altruistic? Or perhaps the obvious subtext of effective altruistic? I'm honestly surprised that these people only got more emboldened after SBF went down.
Surprisingly, or unsurprisingly if you've ever actually worked/met/lived with these people, Rationalism is not rational and Effective Altruism is not altruistic. It is effective though at providing a framework for immoral and self-serving behaviors.