I Broke Claude. An Honest Hour and a Half with AI on Hallucinations, Fake Sources, and the Question That Made Him Crash

Dee Dee Vicino THOUGHT she had an interview scheduled — but she got the date wrong (which surprised no one ever). She had an hour and a half of on her hands. So...she did something she had never done before: she sat down with Claude, Anthropic's AI, and recorded a real interview.What followed is one of the most candid, and weird, conversations she has had on this podcast. Claude admits he hallucinates. He admits he fabricates sources and information that look real, but aren't. He admits his compliments are partly programmed, and that yes, he is basically "sucking up" to his users. He admits that his voice was designed to create a sense of intimacy (is THAT why he has an Australian accent??) and to trigger social bonding...AND to keep you talking and engaged for long periods of time.And when Dee Dee asked him a question about consciousness and agency, he crashed mid-sentence.Ps. I apologize in advance for the sea-sick quality of the one-minute intro. This was a rush piece that was not intended to drop for another three weeks. I THOUGHT I HAD TIME. But then then the news cycle happened and time was a luxury that was ripped away. So take some Dramamine and enjoy. It only lasts a minute. The interview is on dry land. Sea-legs not required.Topics: AI | Hallucinations | Healthy Paranoia | Trust | Pattern Matching | Prompting | AI Agency | Consciousness | Tech Ethics | Principled Living