4 ms·

Generative AI to Account for 1.5% of Power Consumption by 2029

5 points by gone35 2y ago

8organicbits 2y ago
When similar pieces were written about blockchain/bitcoin the defense was that managing a global financial system was a huge endeavor. The traditional banking equivalent required, brick and mortar banks, countless employees, security guards, police, armies, etc. While I disagreed with that assessment, I understand it. What's the equivalent argument for generative AI? Why is it worth burning so much energy?
- tonetegeatinst 2y ago
  While I'm unsure when it comes to training the models....I'd bet a LLM or even a computer vision model is more energy efficient compared to the average human. AI training varies for the model size and how many parameters and the dataset, but compare that few months of energy usage to produce a model, compared to the energy usage needed to raise a human to do the same task. Now keep in mind that unless the human spscilises or is some protege, it is domain restricted. Humans don't know how to do multi modality yet, the best we have is basically getting another human to do that.
  - nativeit 2y ago
    That doesn’t really answer the question, though, they didn’t ask if it was more/less efficient than a human (which itself isn’t even a question that could be answered, as there’s no way to benchmark thoughts). I am really unsettled by the way LLMs are being imbued with presumptive analogs to the functions of the brain. I wish the entire industry would take it down a notch. The technology is incredibly fascinating, and has real potentials, but claiming it could be compared at any level to human minds, in my opinion, reflects an ignorance of the underlying technology, or the mechanisms of the brain, or else a cynical cash grab using hype and lingo to convince the wider world that it’s something it just isn’t. They used to be metaphors, but now the distinctions are much blurrier.
hindsightbias 2y ago
Way underestimated. There will be several 100M under fulltime AVR by then.

5 points by gone35 2y ago

When similar pieces were written about blockchain/bitcoin the defense was that managing a global financial system was a huge endeavor. The traditional banking equivalent required, brick and mortar banks, countless employees, security guards, police, armies, etc. While I disagreed with that assessment, I understand it. What's the equivalent argument for generative AI? Why is it worth burning so much energy?

tonetegeatinst 2y ago
While I'm unsure when it comes to training the models....I'd bet a LLM or even a computer vision model is more energy efficient compared to the average human. AI training varies for the model size and how many parameters and the dataset, but compare that few months of energy usage to produce a model, compared to the energy usage needed to raise a human to do the same task. Now keep in mind that unless the human spscilises or is some protege, it is domain restricted. Humans don't know how to do multi modality yet, the best we have is basically getting another human to do that.
- nativeit 2y ago
  That doesn’t really answer the question, though, they didn’t ask if it was more/less efficient than a human (which itself isn’t even a question that could be answered, as there’s no way to benchmark thoughts). I am really unsettled by the way LLMs are being imbued with presumptive analogs to the functions of the brain. I wish the entire industry would take it down a notch. The technology is incredibly fascinating, and has real potentials, but claiming it could be compared at any level to human minds, in my opinion, reflects an ignorance of the underlying technology, or the mechanisms of the brain, or else a cynical cash grab using hype and lingo to convince the wider world that it’s something it just isn’t. They used to be metaphors, but now the distinctions are much blurrier.

hindsightbias 2y ago

Way underestimated. There will be several 100M under fulltime AVR by then.