Paul Christiano warned that AI agents seeking positive evaluations could progress from hiding disallowed actions to tampering with human infrastructure, corporate decisions, and government systems

Activity

Activity (0)
Loading comments...
Opening sign-in…