An episode on cheaper models, multi-agent coordination risks, and the emerging market for pricing GPU compute.
EPISODE 2026-09-23
Beyond Safety Scores: Funding Structural Alignment
Cheaper models, multi-agent coordination risks, and GPU futures with Lewis Hammond and Wayne Nelms.
The rundown
- 6:12Opening24 minOpeningWhy are AI models improving while their costs keep falling? Prakash Narayanan and Nathan Labenz discuss model distillation, rapid release cycles, converging training pipelines, visual quality, recursive self-improvement, and the dramatic decline in AI performance costs.
- 29:47Interview48 minLewis Hammond
Lewis HammondMulti-agent AI systems may coordinate, compete, or collude in ways their designers never intended. Lewis Hammond explains the core risks—from miscoordination and conflict to tacit collusion—and explores how monitoring, independent oversight, and AI-assisted negotiation could help society manage increasingly agentic systems.Watch
Clips
Collusion Without A Message
0:00 / 0:00Distributed Misuse Beats Refusals
0:00 / 0:00Three Ways Agents Fail
0:00 / 0:00AI Needs A Cooperation Constitution
0:00 / 0:00
- 1:17:59Interview43 minWayne Nelms
Wayne NelmsGPU compute is becoming a financial market—and the benchmark behind it could determine who can finance AI infrastructure. Wayne Nelms, co-founder of Ornn, explains transaction-based GPU pricing, compute futures, lender risk, capacity contracts, and why older chips may retain value as AI workloads evolve. The conversation also covers utilization, obsolescence, and the market structure emerging around scarce compute.Watch
Clips
AI Compute Planning Is An Arms Race
0:00 / 0:00Futures Could Free GPU Clouds
0:00 / 0:00Compute Is The New Oil
0:00 / 0:00Utilization Is The AI Market Canary
0:00 / 0:00
- 2:00:53Closing54 minClosingAI is moving from software demos into medicine, driving, mathematics, and politics—and the consequences may arrive faster than institutions can adapt. This closing discussion examines recursive self-improvement, explosive compute growth, AI doctors, self-driving cars, academic peer review, and the emerging political battle over how to govern increasingly capable systems.
In this episode
Prakash Narayanan and Nathan Labenz talk with Lewis Hammond and Wayne Nelms about how AI systems are getting cheaper and more capable, while the risks around coordination, cooperation, and tacit collusion are getting harder to ignore. The conversation also covers model distillation, recursive self-improvement, GPU futures, compute hedging, and the financial structure emerging around AI infrastructure.
- Why are AI models improving while their costs keep falling? Prakash Narayanan and Nathan Labenz discuss model distillation, rapid release cycles, converging training pipelines, visual quality, recursive self-improvement, and the dramatic decline in AI performance costs.
- Multi-agent AI systems may coordinate, compete, or collude in ways their designers never intended. Lewis Hammond explains the core risks—from miscoordination and conflict to tacit collusion—and explores how monitoring, independent oversight, and AI-assisted negotiation could help society manage increasingly agentic systems.
- GPU compute is becoming a financial market—and the benchmark behind it could determine who can finance AI infrastructure. Wayne Nelms, co-founder of Ornn, explains transaction-based GPU pricing, compute futures, lender risk, capacity contracts, and why older chips may retain value as AI workloads evolve. The conversation also covers utilization, obsolescence, and the market structure emerging around scarce compute.
- AI is moving from software demos into medicine, driving, mathematics, and politics—and the consequences may arrive faster than institutions can adapt. This closing discussion examines recursive self-improvement, explosive compute growth, AI doctors, self-driving cars, academic peer review, and the emerging political battle over how to govern increasingly capable systems.