5 ms·

It seems more than ridiculous that this is considered a serious issue. Datacenter space, generators (to cover the power needs of the datacenter) and power cost

It seems more than ridiculous that this is considered a serious issue. Datacenter space, generators (to cover the power needs of the datacenter) and power cost real money and thus there is plenty of incentive to reduce it. Already we are seeing both hardware alternatives and software optimizations; human ingenuity will eventually take care of it.

croes 2y ago
Eventually. And if not we accelerate climate change because we built fast solutions for the energy problems instead of sustainable ones.
- XorNot 2y ago
  The solution to climate change is not reducing energy usage. That can, at best, delay it. The solution is to generate energy without emitting CO2. We already have the technology to do this. The conflation of supposed moral virtues versus actual, practical solutions is a huge problem with messaging in this space: there are a lot of people who don't see climate change as a problem to be solved, but as the stepping stone to bringing down capitalism or some other goal they want to use it as a bandwagon for.
  - Yoric 2y ago
    Delay is good. Delay is necessary. While we have (some bricks of) the technology to generate/store/consume energy without emitting CO2, we haven't nearly rolled it out and we need time to do so. So, yes, by all means, any month we win by decreasing energy usage might save countless lives. > The conflation of supposed moral virtues versus actual, practical solutions is a huge problem with messaging in this space: there are a lot of people who don't see climate change as a problem to be solved, but as the stepping stone to bringing down capitalism or some other goal they want to use it as a bandwagon for. Fair point.
  - croes 2y ago
    >The solution is to generate energy without emitting CO2. These take time and emit CO2 themselves at least in the bulding phase. And LLMs like GPT won't help us fighting climate change so we raise our energy consumption without later benefit on that matter. Maybe we first should buld the energy sources and then use them for convenience tools. Or do you want to explain our children we fucked up because we "needed" AI to write emails that are read by AI to generate a response? Much of AIs use is still just for unnecessary things just more convenient. It's still unclear if they have a productivity benefit or just changed the way we spend the same amount of time. Intel's study found no such benefit and suggests teaching the people how to use LLMs means how to tell the machine what it should do. Sounds like programming with extra steps and more ambiguity.
    - XorNot 2y ago
      No matter how you slice it though, not building LLMs doesn't change the underlying equation: you're emitting CO2 to make electricity. Every new human being born once they turn on electric light is. Every new industry. Every bit of economic growth, basically. Any action which isn't the reduction of CO2 to generate electricity, or elimination, is just a delaying tactic. Even substantial population reduction would buy you what, maybe a few hundred years before atmospheric CO2 hits the same levels again? There is exactly one problem to solve with anthropogenic climate change and it's CO2 emissions. Any action which isn't directly attacking that issue is a waste of time and political resources, and demonstrably*has been a complete waste of time and effort - every victory was won on the back of superior technology (i.e. efficiency didn't just happen - LEDs happened and they're just plain better and cheaper).
llm_trw 2y ago
The number one predictor of human wealth is energy consumption. Since the stagnation of the west in the 1980s we have seen the same stagnation in energy consumption there.
- benchmarkist 2y ago
  What have you seen in terms of pollution?
  - llm_trw 2y ago
    The desperate need for clean nuclear power the world over.
    - benchmarkist 2y ago
      I suppose then you have spent a lot of effort advocating for this cause or maybe you know a few advocates off the top of your head you could point me to so I could learn more about this desperate effort to deploy nuclear power plants.
      - throwaway314155 2y ago
        https://i.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3lyeG9nNHBlanV5YTFwMXdoM2MwajRwMGFhZ3FqamM5M2U2ZWt0MyZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/GbPru204dTDfG/giphy.mp4 https://i.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3lyeG9nNHBlanV...
- jncfhnb 2y ago
  In what way has western energy consumption stagnated since the 1980s
  - consp 2y ago
    It hasn't, it has been linear for a while and is not exponential by the looks of it. Still climbing though.
ben_w 2y ago
On the vendor's side, there's incentive to reduce the costs per token, but also an incentive to sell more tokens. On the consumer's side, it's more about what the alternative would be, and the alternative is getting a human to do it. Right now the general quality of the LLMs is good enough to be a tool, but not good enough to fully replace humans using it — I've seen GPT-4 called an intern, I'd agree having seen student code not too long before ChatGPT came out, and you don't want to let students do a whole project unsupervised either. Even though o1 is better, I think it's still at the "interesting, but not good enough to leave unsupervised" level*. For tasks where the AI is competent enough to be interesting for whatever task, even if it took 10 kW to run the compute, at $0.1/kWh that's often cheaper than offshoring to a poor country. As the world's entire electricity supply is 2 TW, which is 250 watts per capita, I think it's not at all implausible that enough people decide that the output of various AI (not just LLMs, also vision, audio, robots etc.) are valuable enough to demand enough of that electricity to make the prices rise. * and I'm not just saying that because when it gets that good none of us will be employable any more

croes 2y ago

Eventually. And if not we accelerate climate change because we built fast solutions for the energy problems instead of sustainable ones.

XorNot 2y ago
The solution to climate change is not reducing energy usage. That can, at best, delay it. The solution is to generate energy without emitting CO2. We already have the technology to do this. The conflation of supposed moral virtues versus actual, practical solutions is a huge problem with messaging in this space: there are a lot of people who don't see climate change as a problem to be solved, but as the stepping stone to bringing down capitalism or some other goal they want to use it as a bandwagon for.
- Yoric 2y ago
  Delay is good. Delay is necessary. While we have (some bricks of) the technology to generate/store/consume energy without emitting CO2, we haven't nearly rolled it out and we need time to do so. So, yes, by all means, any month we win by decreasing energy usage might save countless lives. > The conflation of supposed moral virtues versus actual, practical solutions is a huge problem with messaging in this space: there are a lot of people who don't see climate change as a problem to be solved, but as the stepping stone to bringing down capitalism or some other goal they want to use it as a bandwagon for. Fair point.
- croes 2y ago
  >The solution is to generate energy without emitting CO2. These take time and emit CO2 themselves at least in the bulding phase. And LLMs like GPT won't help us fighting climate change so we raise our energy consumption without later benefit on that matter. Maybe we first should buld the energy sources and then use them for convenience tools. Or do you want to explain our children we fucked up because we "needed" AI to write emails that are read by AI to generate a response? Much of AIs use is still just for unnecessary things just more convenient. It's still unclear if they have a productivity benefit or just changed the way we spend the same amount of time. Intel's study found no such benefit and suggests teaching the people how to use LLMs means how to tell the machine what it should do. Sounds like programming with extra steps and more ambiguity.
  - XorNot 2y ago
    No matter how you slice it though, not building LLMs doesn't change the underlying equation: you're emitting CO2 to make electricity. Every new human being born once they turn on electric light is. Every new industry. Every bit of economic growth, basically. Any action which isn't the reduction of CO2 to generate electricity, or elimination, is just a delaying tactic. Even substantial population reduction would buy you what, maybe a few hundred years before atmospheric CO2 hits the same levels again? There is exactly one problem to solve with anthropogenic climate change and it's CO2 emissions. Any action which isn't directly attacking that issue is a waste of time and political resources, and demonstrably*has been a complete waste of time and effort - every victory was won on the back of superior technology (i.e. efficiency didn't just happen - LEDs happened and they're just plain better and cheaper).

llm_trw 2y ago

The number one predictor of human wealth is energy consumption. Since the stagnation of the west in the 1980s we have seen the same stagnation in energy consumption there.

benchmarkist 2y ago
What have you seen in terms of pollution?
- llm_trw 2y ago
  The desperate need for clean nuclear power the world over.
  - benchmarkist 2y ago
    I suppose then you have spent a lot of effort advocating for this cause or maybe you know a few advocates off the top of your head you could point me to so I could learn more about this desperate effort to deploy nuclear power plants.
    - throwaway314155 2y ago
      https://i.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3lyeG9nNHBlanV5YTFwMXdoM2MwajRwMGFhZ3FqamM5M2U2ZWt0MyZlcD12MV9pbnRlcm5hbF9naWZfYnlfaWQmY3Q9Zw/GbPru204dTDfG/giphy.mp4 https://i.giphy.com/media/v1.Y2lkPTc5MGI3NjExc3lyeG9nNHBlanV...
jncfhnb 2y ago
In what way has western energy consumption stagnated since the 1980s
- consp 2y ago
  It hasn't, it has been linear for a while and is not exponential by the looks of it. Still climbing though.

ben_w 2y ago

On the vendor's side, there's incentive to reduce the costs per token, but also an incentive to sell more tokens. On the consumer's side, it's more about what the alternative would be, and the alternative is getting a human to do it. Right now the general quality of the LLMs is good enough to be a tool, but not good enough to fully replace humans using it — I've seen GPT-4 called an intern, I'd agree having seen student code not too long before ChatGPT came out, and you don't want to let students do a whole project unsupervised either. Even though o1 is better, I think it's still at the "interesting, but not good enough to leave unsupervised" level*. For tasks where the AI is competent enough to be interesting for whatever task, even if it took 10 kW to run the compute, at $0.1/kWh that's often cheaper than offshoring to a poor country. As the world's entire electricity supply is 2 TW, which is 250 watts per capita, I think it's not at all implausible that enough people decide that the output of various AI (not just LLMs, also vision, audio, robots etc.) are valuable enough to demand enough of that electricity to make the prices rise. * and I'm not just saying that because when it gets that good none of us will be employable any more