What you'll learn

• 1 (00:00) **Opening: The Jacob Coxon Resignation & The AI Safety Debate** - The episode opens by introducing Jacob Coxon, a researcher who resigned from OpenAI and Anthropic, claiming the companies are "gambling with our lives" in the race for superintelligence.
• 2 (07:35) **The Data Advantage: Why Better Data Beats Better Models** - Dwarkash Patel and Jerry Han's experiment shows that better training data produced a 12X improvement in compute efficiency, while better architectures produced only 3.7X.
• 3 (19:28) **Inside the Labs: The Resignation and the 10% Doom Claim** - The panel dissects Jacob Coxon's resignation and Evan Hubinger's (Anthropic) claim that he personally believes there is a greater than 10% chance AI could kill all humans.
• 4 (31:00) **The "P Doom" Debate: Steering vs. Stopping** - The panel argues over whether AI progress should be slowed or steered, with a consensus that stopping is impossible and dangerous.
• 5 (47:40) **OpenAI's Navier-Stokes Breakthrough & Sam Altman's Call for Slowing Down** - OpenAI claims a breakthrough on the Navier-Stokes Millennium Prize problem using a swarm of 10,000 AI agents, and Sam Altman uses it as evidence for the urgency to slow down.
• 6 (58:38) **The $25,000x Cost Collapse: From $500k to $20** - Noam Brown's prediction that solving such math problems will cost $20 a month within a year is discussed, representing a 25,000x cost collapse in two years.
• 7 (70:48) **The Oren H100 Price Index: A Three-Year-Old Chip Gets More Expensive** - The rental price of Nvidia's H100 GPU rose 22% in a single month, defying all depreciation standards, signaling a massive compute scarcity.

+ Full timestamped outline available in the app