Imperial College professor Andrew Davison argues Déjà View challenges scaling laws by matching models ten times larger using a single iterative block

The method replaces stacked layers with shared-weight iterative execution.