Two Minute Papers unpacks how Claude's latest quirks lay bare AI's deepest flaw: brittle, pattern-bound reasoning that fails hard on anything truly novel. This hits hard because it proves endless scaling won't birth reliable intelligence. My take? We're staring at a paradigm wall, and fresh architectures are the only way forward. What's the one limitation you think will break first? Full video link https://youtu.be/axOcn--n_lM