Interpretable Features in Large Language Models | Towards Data Science

And other interesting tidbits from the new Anthropic Paper