Posts with tagllms

Looped Transformers, or Recurrent Depth Models

How looped transformers reuse their layers to reason in latent space, why recurrent depth is suddenly showing up at the frontier, and what it could cost us in chain-of-thought monitorability.