> I think you'd need to explain to me why adding a tiny bit of additional complexity here is so unreasonable As far as I understand DreamerV3 doesn't employ in

> I think you'd need to explain to me why adding a tiny bit of additional complexity here is so unreasonable As far as I understand DreamerV3 doesn't employ intrinsic rewards (like in novelty-based exploration). It adopts stochastic exploration which makes it practically impossible to get to rewards that require to consistently repeat an action with no intermediate rewards. And finding intrinsic rewards that work good across diverse domains is a complex problem in itself.