I immediately get suspicious when people try to blame things they don’t like on “capitalism” without interrogating whether any alternative system would avoid the problem. It’s almost always an intellectually lazy way of avoiding engaging with deeper issues. So I was very pleasantly surprised to see that invoking capitalism here was entirely appropriate and got me thinking about issues directly germane to my own professional work (I currently work with the global market for cross-language services in a commercial context).

I think you have hit something very real here. Language has been commodified at multiple levels and transformed into a vehicle for promoting machine discoverability of products, all to sell marginally higher amounts of something.

The entire SEO industry is based on this and it is evident that a lot of content is written and transmitted for machines as the primary audience, with humans as a secondary target. This has both a leveling effect and a weird diversifying effect. When you look at keyword strings in e-commerce, they not infrequently include misspellings of common terms because they want to be discoverable based on defective input. This means that web content in this context will deliberately include what appears to be garbage input, but always around a core of “high performance” terms, trying to eke out slightly higher placement in what the machines decide to show to humans. In other words, the persuasive intent of this text is aimed at convincing a machine to show it to a human, and only then does the concern of the human audience come into play.

If you want a wild ride of text that sounds very much like something generated by an LLM, look at the titles of cheap e-books on Amazon. They are thousands and thousands examples of ridiculous titles full of homogenized keyword stuffing, things like “Don't Let Her Stay: An unputdownable psychological thriller with a breathtaking twist” (the first real example I found with an appropriate Amazon search). I say “homogenized” because all of them sound almost exactly the same, with lots of “heart-stopping” and “breathtaking” “surprise twists” “you didn’t see coming” and so forth. Linguistic slop to promote linguistic slop…

When AIs are trained on this sort of (presumably) human-generated slop, is it any wonder that they produce slop in the same vein?

So I thank you for the capitalism framing that I normally object to. It was the opposite of lazy, but instead highly apropos and probative.