“Researchers at OpenAI discovered a specific neuron in the GPT-2 model that selectively activates in response to Canadian concepts.”