What is AI thinking? Anthropic researchers are starting to figure it out

New research sheds light on how LLMs reason with probabilities, creatively plan ahead, and think in universal concepts—revealing their potential and their risks.