Post

I feel like we need an "Appeal to Future AI" entry in the list of common fallacies. "Eventually AI will be good enough that it will do X as well as a human. Therefore, we should give up now and stop bothering to have humans do X today, because it's too slow and it won't matter if the AI can do it!"

So we take humans out of the loop and let today's AI fail at solving today's problems, in the hope that it'll fail less next year. And if it doesn't? Well, AI can never fail, only be failed. We'll be told to write ever longer and more detailed skill markdown. And that means more memory for DCs!

(Ironically, the X that's on my mind is actually a task I hate doing; it's onerous, repetitive, and thankless. If our current models+skills were good at doing it today, I'd use them! But they're not, and the consequences of taking humans out of the loop _today_ are dire. We're doing it anyways!)

I have a graduate student working for me who admitted to me they have 100% vibe coded their way through their degree and have little to no real understanding how anything they have done works. They are not unique in their peer group and they are concerned about viability of job growth going forward