Dreaming (Internal Simulation Scaling)
Definition
A speculative but powerful model: AIs could build internal simulations of environments and train against these simulations, dramatically multiplying the available training signal without more real-world interaction. This would add a new scaling axis (test-time compute/dreaming) alongside pretraining, RL, and inference compute. Historical precedent: EfficientZero used internal Monte Carlo tree search over learned models to multiply effective samples.
Cast within
No topic-region cast yet — this appears once Dreaming (Internal Simulation Scaling)'s compiled claims are aligned into a topic region's argument tree.
Claims mentioning Dreaming (Internal Simulation Scaling) (2)
Your notes
Loading notes...