Anthropic released a post earlier this month about the "J-Space," which shows how an LLM thinks, and posited a "J-lens" to help engineers or users "see" what is in this internal workspace. The J-space is a part of the LLM where hidden signals light up, showing what the LLM is thinking about without generating explicit expression. It is named after the "Jacobian matrix," where a function maps how each of many inputs relates to the steering of the LLM’s cognitive process. According to Anthropic's July 6 announcement, each J-space pattern is linked to a particular word, but when a pattern lights up, it does not mean the model is saying that word—just that the word is on its mind. The J-space operates silently in the model’s internal neural activations and emerged on its own during Claude’s training process. The J-lens uses a view of that Jacobian matrix to see hidden patterns representing what Claude is thinking without telling you. Applying the lens at each layer gives a ranked list of vocabulary words, with patterns collectively forming J-space, Claude's internal global workspace.