Flaws in AI models — from hallucinations to biased training data — often manifest only when systems are live. Experts propose a new framework for third-party reporting to allow external evaluators to identify and flag these issues effectively.

→ View original post on X — @stanfordhai,