The actual value AI LLM delivers is very divided. For users who produce below average output without AI, it is magical. If you can produce above average, it is threatening, as you see it’s nuanced errors and edge case flaws.

The financial case for investment presumes the value of improving below average to average will deliver the value current average delivers, but it is flawed because it isn’t a linear model. Consider the finance industry- no value exists in consistently delivering average or below outcomes, all the value is in delivering above average. You are 100% failing if you consistently deliver average or below.

Other industries might be shielded from the harsh measurement of finance. But would you invest in a factory or service business delivering average or below? I guess there might be some hold out industries harbouring a workforce with such thin margins and limited exposure it couldn’t outsource to lower cost labour but it surely seems to me they would have already captured the savings available via offshoring where quality of performance is optional.
It seems like there might be efficiencies in coding, at the expense of quality, but it also seems like there isn’t a good way to evaluate areas where these savings could be realised without unknown future costs from sloppy coding.

So maybe the benefits are shallow. My analogy is like desktop publishing. All of a sudden, cafe menus and school newsletters became much better as the tools allowed people who were not graphic designers produced documents. This was a sea change in school newsletters and cafe menus, but nobody was paying for those anyway. Now, we can see food trucks with unpalatable AI menus and consultant reports that say nothing. AI isn’t producing a saving here, just a cost as people skip machine looking food.

A lot of people are getting rich from AI. I think a lot more will get poorer as the reality sets in and the limited utility sets in. Of course, this doesn’t mean it won’t cause a lot of harm. Nobody except maybe some shareholders are better off that a call to fix a problem with my telco or power company service needs a battle through a phone computer and offshore call center. But only an imbecile will be choosing to rely on AI for stuff that matters.