04/09/2025 / 最終更新日 : 04/09/2025 kawahara AI in a vat: Fundamental limits of efficient world modelling for agent sandboxing and interpretability
