S1 · L092:45
The Paperclip Problem
Don't build another paperclip: start with the job, design for mistakes, keep humans where it matters, and build your moat around the model.
AI
Glossary
Evaluations (evals)
Sets of real examples with known good answers, used to measure the quality of an AI feature instead of judging it by feel.