TechTakes

I would actually contend that crypto and the metaverse both qualify as early precursors to the modern AI post-economic bubble. In both cases you had a (heavily politicized) story about technology attract investment money well in excess of anyone actually wanting the product. But crypto ran into a problem where the available products were fundamentally well-understood forms of financial fraud, significantly increasing the risk because of the inherent instability of that (even absent regulatory pressure the bezzle eventually runs out and everyone realizes that all the money in those 'returns' never existed). And the VR technology was embarrassingly unable to match the story that the pushers were trying to tell, to the point where the next question, whether anyone actually wanted this, never came up.

GenAI is somewhat unique in that the LLMs can do something impressive in mimicking the form of actual language or photography or whatever it was trained on. And on top of that, you can get impressively close to doing a lot of useful things with that, but not close enough to trust it. That's the part that limits genAI to being a neat party trick, generating bulk spam text that nobody was going to read anyways, and little more. The economics don't work out when you need to hire someone skilled enough to do the work to take just as much time double-checking the untrustworthy robot output, and once new investment capital stops subsidizing their operating costs I expect this to become obvious, though with a lot of human suffering in the debris. The challenge of "is this useful enough to justify paying its costs" is the actual stumbling block here. Older bubbles were either blatantly absurd (tulips, crypto) or overinvestment as people tried to get their slice of a pie that anyone with eyes could see was going to be huge (railroad, dotcom). The combination of purely synthetic demand with an actual product is something I can't think of other examples of, at this scale.