😺 🎙️ Watch: Opening AI’s black box

Eric Ho of Goodfire explains why models may think in shapes, how interpretability can reduce hallucinations, and what Goodfire is building.