paper

Deep Learning and the Global Workspace Theory

Rufin VanRullen, Ryota Kanai · May 14, 2021

The paper that argued the workspace could be built, not just found: a roadmap for a "global latent workspace" shared between neural modules via unsupervised translation. Anthropic's result is the mirror image: the workspace appearing unbidden in a model that was never designed for it. Read the two together.

← Back to sources