Algorithms of Empire (Episode 10); Special Guest: Dr. Ashique KhudaBukhsh
In this episode of Algorithms of Empire, we speak with Dr. Ashique KhudaBukhsh about his team's startling discovery: advanced AI models can be persuaded to accept falsehoods as truth through nothing more than conversational pressure. What began as a casual observation—ChatGPT confidently inventing a non-existent Hitler reference in Good Will Hunting—turned into a systematic study of five leading models across 2,000 movies and novels. Dr. KhudaBukhsh and his team found that soft nudges like "I'm pretty sure…" or "I remember it this way…" could make AI systems abandon facts they initially…
The skinny
The skinny isn't ready yet — notes appear once the transcript is processed.