Relying On Chain-Of-Thought For Boosting AI Safety Gets Sideswiped By This Split-Brain Discovery

A discovery in AI safety research shows that chain-of-thought methods have limitations, challenging current beliefs in the field.