I think we will be able to get the power and chips we need for the next three years or so. After that, it's possible that the big advances driven by hordes of AI-engineers on a server might come from or involve big efficiency gains. My intelligence, for example, runs on about 20W. It's facile to point to biology as what's possible in many cases, but it shouldn't be discounted.
Thanks, Paul, thoughtful comment. I agree that efficiency gains may become a bigger driver eventually, and it’s wise not to dismiss biological benchmarks entirely. But I’d argue that whatever breakthroughs arrive in the medium term, whether algorithmic or architectural, still ride on a substrate that’s brutally hard to scale: land, power, cooling, chips, and logistics. Even recursive self-improvement loops consume vast energy and compute at the outset.
So yes, intelligence may one day become cheap, but right now, it’s expensive, and getting more so. The gating function isn’t what’s theoretically possible. It’s what’s physically buildable in the next three years. That’s where the misallocation lies.
My read on this is maybe.
I think we will be able to get the power and chips we need for the next three years or so. After that, it's possible that the big advances driven by hordes of AI-engineers on a server might come from or involve big efficiency gains. My intelligence, for example, runs on about 20W. It's facile to point to biology as what's possible in many cases, but it shouldn't be discounted.
Thanks, Paul, thoughtful comment. I agree that efficiency gains may become a bigger driver eventually, and it’s wise not to dismiss biological benchmarks entirely. But I’d argue that whatever breakthroughs arrive in the medium term, whether algorithmic or architectural, still ride on a substrate that’s brutally hard to scale: land, power, cooling, chips, and logistics. Even recursive self-improvement loops consume vast energy and compute at the outset.
So yes, intelligence may one day become cheap, but right now, it’s expensive, and getting more so. The gating function isn’t what’s theoretically possible. It’s what’s physically buildable in the next three years. That’s where the misallocation lies.