hype-utility-inversion

I’m writing to you from near the peak of the global generative AI and LLM bubble. It’s pretty crazy in here. Hundreds of billions of dollars are being almost literally burned in massive Earth-warming trash cans to achieve little of genuine benefit.

Generative AI hype has been out of control for at least 3 years, but reality is finally starting to catch up. With this, something interesting is becoming clear:

// Generative AI is adding some value. But in simple and utilitarian ways that no one in the generative AI industry really wants to promote.

// Whereas the impressive world-changing use cases for generative AI that the industry is constantly talking about are in fact pretty much useless.

I can give you an example from my own industry: There have been various claims by highly influential technology leaders (people who should know better) that generative AI is on the cusp of completely replacing software developers. And taking this cue we’ve now got a whole wave of ‘vibe-coders’ producing absolute garbage software that the vibe-coders themselves have no idea how to improve, fix or maintain – if they are lucky enough to even get it working in the first place. In the meantime, the use of generative AI in professional software development has started falling already, because people are increasingly seeing and understanding its limitations. With humans being actually intelligent, the industry has been adaptable to these limitations. Developers have found areas where generative AI can add some marginal utility, such as automating the coding of unit tests. Not particularly sexy, but moderately helpful if done right.

Generative AI has a hype-utility-inversion. The greater the hype; the worse the outcome. The more modest the claim; the more useful it tends to be.

Bit strange, isn’t it?

Actually if you think about it, it makes perfect sense in an industry where c. 90% of the money currently fuelling it is coming from investors rather than customers.

But investment capital cannot be sustained forever. And customers simply won’t pay insane prices for modest (even if real) productivity improvements.

The problem is that there is simply too much money in AI now. It’s over-leveraged, and the only way it won’t collapse is if some of these ridiculously grandiose claims come true. But they won’t. Not only is generative AI in a hype-utility-inversion, it is stuck there. Which means the AI industry has needlessly checkmated itself, and the bursting of the bubble is now inevitable.