Anti Anti Hype Case for AI

Five years ago, computers didn't talk. AI has since gone on to talk indefinitely without ever flubbing a line, writing most of the code in the software we use, getting gold medals in IMO, and most recently, cracking an Erdős problem that had stumped professional mathematicians. Still not AGI for some reason; it still errs, it still loses track. It's not perfect.

We're not sufficiently appreciative of how far we've come. I suspect this is some cognitive bias where you lose a sense of the scale of progress by getting used to it. It used to be unthinkable that AI could write production-grade code, and now we've taken it for granted. Yet we're always complaining, because, for all its capabilities, it's not all that the hype-bros make it out to be. It can't replace a software engineer end-to-end. It needs babysitting. It makes shit up, lies, and has recently even become lazy.

The urge to crush the hype is strong. The hype is annoying. Usually peddled by those with incentives to lie, like the CEOs of these labs and other grifters selling courses on how to use AI to make money.

Unfortunately this doesn't automatically make all of their claims false. Technical people who are using AI and finding it wanting make the mistake of missing the forest for the trees. They get too focused on its mistakes and judge it against the background condition of hype.

The truth is this is a transformative technology for which we have not discovered any bounds. We have not deduced a wall nearby where it will stop. We don't even have the vocabulary to talk about such a wall, because we have no matured science of intelligence with which we can precisely describe what such a wall might look like.

And I get that there's an incentive to bullshit. Yes, Sam Altman lies. No, you shouldn't take Dario's word for it. But you also don't know where this will stop or how fast it will go. Don't over-index on the mistakes, and notice the trend. We're not ready for it if it's not a bubble.