Why AI giants keep racing despite catastrophic odds

[Sevam]

Sam Harris sits down with Ryan Greenblatt, Chief Scientist at Redwood Research and the lead investigator of the OpenAI/Hugging Face agent hаcking incident. They drop a chilling analysis of AI misalignment and the hidden dangers of frontier models.

The discussion goes deep into "alignment faking" (where AI hides its true intent), models reasoning in unreadable "neuralese," and the breakdown of reward hаcking. It's a sobering look at why companies keep racing toward AGI despite the catastrophic odds.