Debate Questions Global Workspace Theory Link To Anthropic Claude Research
Skeptics say standard transformer training naturally mimics these brain-like activations.
On X, Carnegie Mellon researcher Aran Nayebi pushed back on a reading of recent Claude interpretability work as a meaningful match for Global Workspace Theory, arguing that the reported alignment may be much broader than that. In a later quote-post on X, he said Dehaene and colleagues did not simply endorse the comparison and argued that a subspace of activations aligned with future output could be expected in trained autoregressive systems more generally.
My read: (1) Dehaene et al.'s article doesn't purely endorse it—they note both similarities & differences; (2) it's natural for a theory's creator to get excited by some match. But that's an illusion because the match is so broad many things fit.
Source Posts on X
Cluster sources@rgblong My read: (1) Dehaene et al.'s article doesn't purely endorse it—they note both similarities & differences; (2) it's natural for a theory's creator to get excited by *some* match. But that's an illusion b/c the match is so broad many things fit: https://x.com/aran_nayebi/status/2074221761281888330
Combined views
350
2 posts, first seen 7h ago