Joachim Landström

YOLO AI-mantra.

A MIT PhD student told me he can predict exam questions before seeing the syllabus. Using NotebookLM. I thought he was full of shit. Then he walked me through it. He doesn't wait for the course to start. Before day one, he uploads 10-15 past papers from similar courses. Adds the core textbooks. Throws in a few field research papers. Most students never do this. He's already built a training dataset of how the subject gets tested. First prompt he runs: "What patterns exist in how this subject is examined?" Not summarize. Not explain. Patterns. NotebookLM starts surfacing things like concepts that reappear every year in different disguises, question formats professors recycle, and topics that always show up together. Then comes the move that makes it unfair. "What hasn't been tested recently, but should be?" That's where the predictions come from. Exams rotate. Professors don't repeat last year's exact questions, but they never change the underlying ideas either. Final prompt: "Generate 20 high-probability exam questions based on these patterns and gaps." They don't look like practice questions. They look like real exam questions. He spends the next few hours solving them against the source material. Every wrong answer triggers one more prompt: "Why is this wrong and what concept am I missing?" By the time the syllabus drops, he's already mapped 70-80% of it. By exam day, nothing feels new. Just slightly different versions of problems he already solved. Most students study what they're given. He studies how the system behaves. That's not studying harder. That's playing a different game entirely.

Larry: Are you becoming chartist? 😀(I kind of remember that something happened at the end of the 1970s, like a trippling of oil prices, which might be vaguely relevant. Do u expect it to happen again?)