Methinks the bubble will burst, and it's only a matter of how long it will take for company directors, vulture capitalists and other gullibles* to notice it's happened, and how much of the world economy it'll take with it. Two simple words: model collapse. That's what happens when you use the output of an LLM to train another LLM, or worse still itself; basically the noise drives out the signal, the LLM gets stuck on its own hallucinations, and the poor thing ends up simulating a conspiracy theorist.

This has been demonstrated directly in the lab. Now that much of the Web's content consists of what's called "AI slop" in the trade, scraping the Web for input for other LLMS has closed the loop, and we'll be in for a reproduction of the lab results. "Boom," says Ivanova.

I look forward to the expression "Artificial Incompetence" being used in anger by the Press. In the meantime, Real People™ will have to put up with ever-increasing denial-of-service attacks on genuine Web sites like ours from scraper bots which actively ignore the "get knotted" notices. Buckle in, folks. End of rant.

* Sign of the times: you know something's in trouble when politicians start backing it.