Danger does not come merely from agency, but from agency with no ability to anticipate consequences and with no safety guardrails. The solution? AI agents that can predict the consequences of their actions (world models) and only take actions whose predicted outcomes satisfy

→ View original post on X — @ylecun