Sometimes perfect solutions fall short against recursion. As Juvenal said:
Quis custodiet ipsos custodes?
Who will watch the watchmen?
When it comes to generative models, they do wonders, but their very existence removes incentives to produce the data on which they are trained.
I have reached the point where Google Search has strong competitors in my habits. I was using ChatGPT, I am switching towards Perplexity AI. You ask a question you get an answer, the former even gives sources to check it. I am aware that the answers I am getting may be hallucinations, but online answers are also not unlikely to be wrong and access to a structured answer to my queries is much, much faster.
However, Google Search was and still is a door towards other contents, you start there, you end up on blogs, on videos, in some obscure mailing list archive, in a manual, on Wikipedia. Those prime sources of information exist, they feed modern artificial intelligence based solution. But they only exist because their authors had motivation to create them.
If you exclude a sad yet solid share of academic production, very few authors are writing for the sake of writing: they want an audience either because the audience pays, because they can monetize it or simply because it feeds their ego… Welcome on my blog, by the way.
If generative models win the battle for attention on a topic or another they also remove any incentive for the authors that could have fed their future knowledge.
