When I started training models in 2021 I joked to my wife that I had found something more computationally expensive than crypto mining. Whenever I was asked about AI and the risk to humanity (2023), I would point out that there were very real physical limits to what was going on. AI training is eye-bleedinglying expensive in power. This is a key limiting factor1. There is a very real reason that Microsoft is investing in nuclear energy. Nature always bats last.
What does nag at me in this paper, and many other discussions of the future of AI, is the failure to give adequate consideration to some of the impediments to adoption.
- Thou we all expect that John Carmack has a fast-inverse type hack on the go ↩︎
Dave5 
Leave a comment