The people who know the most often sound the least certain

People who know the most often sound the least certain

Dwarkesh Patel hosted a 90-minute conversation with John Schulman, Beren Millidge, and Charlie O'Neill on whether AI can automate research and whether progress hits a reward-function wall. Viewers fixated on filler words like 'like.' The author argues that genuine experts hedge, correct themselves, and sound uncertain, while polished speakers compress complexity into slogans that travel further. In AI, the gap between technical truth and public perception is dangerous. Researchers should practice explaining ideas to a five-year-old, use pauses instead of fillers, and make expertise visible as a process—not a political pitch.