AI agents can now judge their own success
A new method lets AI agents detect whether they succeeded by reading their own internal states, with no extra compute needed.
A new method lets AI agents detect whether they succeeded by reading their own internal states, with no extra compute needed.