Anthropic research finds Claude uses a global workspace architecture mirroring human access consciousness
Anthropic says a small internal "J-space" inside Claude seems to carry the model's reportable, reasoning-ready thoughts, and the claim is already splitting researchers and X commenters.
1.1K18.6K1.3K7.2K2.1M
“I thought this was an excellent paper! I've long suspected that models have some kind of "working memory" to store intermediate variables during a forward pass and IMO this paper has the best evidence yet”
Neel Nanda@NeelNanda5TECH#254Story Brief
Anthropic says new research on Claude found a small internal “J-space” that acts like a global workspace: a limited set of representations the model can report on, hold in mind, and use for deliberate reasoning. In its announcement on X, the company compared that setup to neuroscience theories of human access consciousness, while the full paper argues that most model activity stays outside that privileged channel.
“One of the only times I remind people I have a PhD in computational neuroscience is when people without a neuroscience background say their model works "like the brain.”
Ziv Ravid@ziv_ravid
