The AI Language We Can't Read: Neuralese ft. Rob Miles - Computerphile

Why watch

This video explores the concept of 'Neuralese' - a potential internal language that large language models might develop that humans cannot interpret. It examines how Chain of Thought prompting gives limited insight into model reasoning, and considers the implications for AI safety when models communicate in ways that escape human understanding. Ideal for viewers curious about AI interpretability and the open questions around AI safety.