succinctness

This paper argues that transformer architectures are inherently succinct, meaning they can represent certain functions more efficiently than other models. It presents theoretical analysis and proofs.