@masek this relies on the assumption that these things are actually useful enough to justify the exponential burning of money on them, and secondarily that there is actually enough capital in the world to do that. Which there isn’t - mathematically the current expenditure growth cannot continue.
Per Cory Doctorow’s recent analysis it would require 2x the whole world’s current loan output to keep this going for another 12 months. It’s not theoretically unsustainable, it’s fully economically unsustainable, which is why the big 3 ai players are setting up exit strategies by making the stock market hold the bag.
But also, this is all relying on LLMs being useful. 30 trillion dollars/year useful. Which they aren’t, and will never be. Unless a billion people all spend 30k a year on ai from now until forever.
So no, I reject the narrative of this shit being inevitable. Which company is going to lose a billion dollars a year updating their models so that their cloud ai/ local models don’t become outdated, creating deprecated code that can’t use new frameworks, if they cannot turn a profit from it?