“Shock! Shock!”

That’s how Professor Donald Knuth—the man who literally wrote the bible of computer science, The Art of Computer Programming—opened his latest paper.

He wasn’t reacting to a scandal or a bug. He was reacting to a mathematical breakthrough. Specifically, a breakthrough made by an AI.

For weeks, Knuth had been wrestling with an open problem: a graph decomposition conjecture. It’s the kind of deep, structural math that usually requires years of specialized intuition. Then he gave it to Claude (Opus 4.6).

In about an hour, after 31 explorations, the model solved it.

The Human in the Loop

There’s a common narrative that AI “replaces” thought. But look at how Knuth handled this. He didn’t just copy-paste the output and call it a day. He read the AI’s work, verified the path it found, and then sat down to write the formal proof himself.

This is the part people often skip over. The model found the path; the human confirmed it was real. It wasn’t replacement—it was collaborative math.

A Scientist’s Joy

Knuth is a legend. He could have easily dismissed the result, or felt threatened that a machine “solved” something he’d been stuck on for weeks. Instead, he named the paper Claude’s Cycles.

“What a joy it is to learn,” he wrote.

That’s the difference between a scientist and a pundit. A pundit writes threads about why AI is over hyped to protect their brand. A scientist changes their mind when the evidence changes. Knuth didn’t grumble about the “death of craft.” He celebrated the fact that he knew something today that he didn’t know yesterday.

If the man who defined the foundations of our field is willing to revise his opinions on generative AI, maybe the rest of us should be paying closer attention.

Source: Claude’s Cycles (PDF)