Nick Cammarata's Interpretability Find: What It Means for AI Safety
Nick Cammarata, a former OpenAI interpretability researcher now at Goodfire, published a research essay on October 15 titled "Neural networks are already clean," arguing that neural networks are far more readable than th…