It says much more than that. It says that all custom architectures and approaches are become irrelevant if compute increases, which was proven again by the large latent models recently solving RL tasks better than any classical RL technique.