LLMs are not the black box you were promised

The article discusses the progress made in understanding the inner workings of large language models (LLMs) through mechanistic interpretability, a technique that breaks down the model's reasoning into human-interpretable features.