Posts Introspective interpretability Language models, world models, and human model-building Notes on teaching GPT-3 adding numbers