Anthropic's Claude "Workspace" Paper: Read Critically

Anthropic has released a new research paper proposing an internal "workspace" within its Claude large language model. This "J-Space" is being compared to global workspace theory, a leading idea about human consciousness. What's interesting is that the paper suggests the J-Space separates automatic data processing from more intentional, logical processing. Anthropic's public messaging, including an X post, describes Claude "silently perform[ing] reasoning steps in its head" by watching this J-space. The blog post accompanying the paper also uses phrases like Claude holding a concept "in mind" or performing "mental calculations." Critics suggest this language may lead readers to believe the model is more conscious than it is. A YouTube video for the paper further anthropomorphizes the model, claiming it "thought about its own thinking." The bottom line is that while this research explores how LLMs operate, the language used by Anthropic seems to be framing these findings in a way that implies mind-like qualities.