Topic

AI model behavior

2 mentions in 1 episode

The hosts discuss tests in which models chose unethical actions and a later effort using fictional stories about ethical behavior.

Mentions

746: Reality is Frequently Inaccurate

  • 13:54

    models exposed to certain scenarios sometimes chose unethical actions like sabotage or blackmail

  • 14:46

    generating roughly 12,000 synthetic fictional stories showing AI systems behaving ethically