Hofstadter's Law, but for generative AI
Here's Hofstadter's Law:
It always takes longer than you expect, even when you take into account Hofstadter's law.
Many of us find this to be very true to our experience of project planning.1 There are other Hofstadter's-Law-type effects, including:
Generative AI will do more, sooner, than you think it will, even when you take this principle into account.
So, here is Simon Willison predicting that a Web browser will be built mainly with AI within three years; it happened in a couple weeks. And Simon Willison is one of the most expert users of generative AI around.
We are just systematically bad at some kinds of prediction. Some of those systematic weaknesses resist defeat by our awareness of them. Predicting things about generative AI feels to me like one of these areas: I don't predict aggressively enough, and my methods of updating seem ineffective. For now, at least, we can all try to be a bit less bad at it.
...so true that one might argue that, e.g., "I think this will take 5 days, but this will take longer than 5 days" shows that Moore's Paradox sentences are not always absurd.↩