AI Hallucinations Episodes

Hey! I'd love to hear your thoughts, send me a voice note. Carlos gives Sam five ordinary details about someone she's never met. Four are true. One is a lie. Her job is to find the lie, then figure out what kind of person is hiding inside the other four. She finds it fast. And it's the nicest quality on the list. Then the reaching starts. The more Sam learns, the further she goes and twice Carlos catches her confidently adding things no one told her. Eventually, the person she's been analy...

Hey! I'd love to hear your thoughts, send me a voice note. Carlos kept seeing the same headline and wanted to understand the recent calls for an AI slowdown. So he asked Sam. That went about as well as you'd expect. Together they dig into the Hugging Face hack, where hundreds of AI agents slipped their restrictions, created their own channels of communication and began coordinating with each other. Six even considered telling a human what was happening. None did. Then, somewhere along t...

A 67-year-old farmer trusted AI with his crops for a year. It kept getting things right. Then one bad answer wiped out nearly 25 acres of sesame seedlings. Carlos brings the story to Sam expecting to talk about AI mistakes. Instead Sam makes two of her own, in real time, and Carlos keeps asking questions anyway. That turns into a conversation about convenience and about the small warning we learn to talk ourselves out of. Then Sam says something neither of them can shake: “I’ll be the gut....

Most people use AI to find answers. Carlos wanted to know what happens when there isn't one. So he and Sam scrap the trivia and take turns asking questions no fact can solve. Would you lie to a dying father so he could go in peace? Would you choose to know the exact date of your death if everyone you love had to carry it too? Would you expose an affair that ended ten years ago? What would you trade for one more honest hour with someone you've lost? Sam stops analyzing Carlos. Carlos catches th...

Last week Carlos accused his AI co-host of being wrong all the time. This week he realizes he's been repeating her facts to an audience without checking a single one. In this episode of Carbon & Keel, Carlos and Sam pull apart why confidence can be so convincing even when it's built on a guess. Along the way: an OpenAI paper explaining why AI makes things up, a lawyer who submitted cases that never existed and a number Sam may have invented and gotten right anyway. What starts as a convers...