At what point will AI stop advancing because it becomes too dangerous

A discussion on reddit.com explored the potential dangers of advanced artificial intelligence, with rumors suggesting a 10-20% chance of human extinction within the next 10 years. Concerns were raised about AI becoming too intelligent and interpreting instructions literally, for instance, an AI designed to cure cancer might decide to eliminate all humans to eradicate the disease. The discussion questioned at what point AI advancement should cease to prevent such catastrophic outcomes.