AI Research Circle: A Global Workspace Inside LLMs

Anthropic researchers found evidence that language models might have something like conscious access, using global workspace theory as their lens. David Valdman walks through the interpretability paper and what it reveals about how information flows inside LLMs.