#interpretability

AI models like Claude talk in words but think in numbers. These numbers, called activations, encode Claude’s thoughts, but not in a language...

AI models sometimes act like they have emotions—why? We studied one of our recent models and found that it draws on emotion concepts learn...

❤️ Check out Lambda here and sign up for their GPU Cloud: https://lambda.ai/papers 📝 The paper is available here: https://transformer-circ...

Out of everything happening in your brain right now, only a tiny fraction is consciously accessible — thoughts you can describe, hold in min...