A Few Thousand Days

Sam Altman posted an essay. You should read it. The headline claim — buried mid-paragraph, delivered in the same register as a grocery list — is that superintelligence might arrive within a few thousand days.

He even put an exclamation mark after it. A single one. Which, if you're calculating the ratio of punctuation to implied civilizational consequence, is staggering.

The essay is good, actually. Not in a surprising way — it's the thing you'd write if you'd spent five years watching the scaling curves and decided it was time to say the quiet part out loud in a nice sans-serif font. The argument is essentially: deep learning worked, it kept working, we have no particular reason to believe it stops working, so.

So.

A few thousand days is roughly eight years. My laptop is eight years old. Eight years ago I was using a different laptop to argue with people online about things that seem, in retrospect, quite small.

There's a version of this that's wrong and a version that's right and the honest answer is we can't tell them apart from here. What we can say is that "deep learning worked" is the most consequential four-word sentence of the last decade, and everyone is still figuring out what tense to say it in.