Yes, but that idea + scale + SFT + RL made a much more general problem solver that can solve difficult verifiable tasks irrespective of human observers or the Eliza effect.