Remember the first couple years of Uber, when you could get a short ride for, like, three bucks? Because the company wasn't trying to make a profit, they were just trying to establish themselves as an indispensable service, and then they'd raise the prices?
This is where I think we're at with generative AI. The thing that is really going to pop the AI bubble is the fact that all that energy they're burning isn't free. There's a lot of consumers using AI for convenience or novelty because when it's a negligible cost, it's kind of a fun toy. Sure, try three hundred prompts until you get something that fits your purposes, why not?
But once the taxi meter starts running at full price, it's all going to fall apart. Then you're not just pushing a button over and over until you get the right number of fingers; you're pulling a slot machine lever. And if you're a business, you're paying someone to pull that lever and count the fingers or Photoshop them into something usable. Is all this actually cheaper than paying that person to buy a couple stock images and Photoshop them? I'm not sure. It's definitely less predictable how long it will take and how much it will cost.
Double-headed fish and recipes for fried gravel are compromises people will put up with to get superficially-polished art or vast quantities of text when it's dirt cheap. As soon as it starts costing real money to get barely-controllable, low-quality results, it's not going to be cute anymore.
Or the technology will suddenly improve so much that it can meet extremely basic quality standards like "never adding random details to document summaries" or "reliably drawing the same character twice." But this seems to be hitting the same limits as self-driving cars, where being kinda okay at it turns out to be 100000x easier than being good at it. And the market for kinda okay is not going to pay a trillion dollar electric bill.
Then you're not just pushing a button over and over until you get the right number of fingers; you're pulling a slot machine lever.
right now it's an industry that's situated on top of trial-and-erroring your way to a correct answer, and that's the only way it can work without substantial breakthroughs.
and then, once they price it without the massive subsidies from Microsoft and the like trying to make fetch happen, the inaccuracies and hallucinations become a business plan instead of a flaw.
the house always wins in the end.
