How a Simple Puzzle Reveals the Jagged Edge of AI’s Abilities
Sept. 23, 2026
It’s undeniable that AI models have gained many powerful abilities. And yet, for all of their advances, they can still make basic blunders.
Researchers call this “jagged intelligence,” a term suggesting that there’s a jagged boundary separating what AI can and can’t do.
In this article, I’ll teach a large language model how to solve a simple puzzle, starting from scratch, and we’ll see exactly how this jagged intelligence emerges.
First, try the puzzle yourself: