“Anthropic's interpretability research, led by Chris Ola, identified a specific neuron that activates when a language model processes text about the Golden Gate Bridge.”