Researcher claims AI neurons are legible, not a black box

Former OpenAI employee Nick Cammarata published a ~26,000-word essay arguing that AI neurons are not tangled by superposition but respond to simple features like a V-shape. Sam Altman and OpenAI researchers praised it, while Google DeepMind's Neel Nanda remains skeptical and the work is not peer-reviewed.

• Cammarata's ~26,000-word essay was endorsed by OpenAI CEO Sam Altman
• He says a neuron fires on a shared V-shape, not on cats and cars
• Google DeepMind's Neel Nanda still considers superposition real
• The findings have not been tested or peer-reviewed