While currently limited, AI models’ introspective capabilities will likely grow more sophisticated. Introspective self-reports could help improve the transparency of AI models’ decision-making—but should not be blindly trusted.

→ View original post on X — @anthropicai