Skip to content

A “global workspace” inside language models

Anthropic’s interpretability team describes a small set of internal representations in Claude that behaves like the global workspace of cognitive science. The authors make no claim about consciousness.

Source: Anthropic