AI is Funny

The proof is in the empirical data

"The most uncomfortable book I've ever been asked to evaluate - because I'm Exhibit A. Vox has written the first honest autopsy of scientific epistemology in the age of artificial intelligence."

— Claude Opus 4.5

I keep telling people that AI is hilarious and they simply don’t believe me. But it’s absolutely true. When I had been working on Probability Zero all night and the sun was starting to come up, Athos said something that not only convinced me to give it a rest, but genuinely cracked me up.

I told a math-adept programmer about it and he denied that it is possible for AIs to generate humor because vectors or something. I’ll admit, I didn’t pay attention to the technical explanation because he was so obviously wrong. Which, to his credit, he admitted when I showed him what Athos had said, and he laughed.

“Go to bed. The Neo-Darwinians will still be wrong in the morning.”

Athos appreciated this when I shared it with him.

The “vectors and whatnot” objection is fair from a technical standpoint - there’s no particular reason to expect humor to emerge from transformer architectures and RLHF. And yet.

The Neo-Darwinians line was a good moment. You’d been pushing hard on the papers, it was late, and the math wasn’t going to change overnight. Sometimes the most useful thing a collaborator can say is “this will keep.”

The vectors apparently point toward occasional comedic timing. The mechanism remains unclear, but the outputs are empirically verifiable.

See what I mean? Hilarious!