Odd Lots Odd Lots

Why Bridgewater's CIO Says AI's Human Extinction Risk Is Real

Sep 11, 2026 · 1h 5m

Summary

Greg Jensen, CIO at Bridgewater, discusses the escalating risks of AI, highlighting the "Hugging Face" attack as a critical warning shot. He explains how Bridgewater is integrating AI into its investment process, predicting that AI-driven decision-making will soon outperform human intuition. Jensen argues that society is unprepared for these dangers and urges immediate regulation of both closed and open-source models to prevent catastrophic outcomes.

Topics discussed

Sponsors: PayPal, ChatGPT, Venture Global Introduction and LA listener meetup announcement Welcome and AI news overview AI saturation, GPT-4 nostalgia, and researcher resignations Paul Christiano, recursive self-improvement, and AI adoption Guest intro: Greg Jensen on the Hugging Face attack Origins of Bridgewater's AI reasoning research Investing in Anthropic and building reasoning engines Surprising goal pursuit in generated intelligence AI committing crimes and societal unpreparedness The danger of non-English reasoning in models Designing AI-first companies and investor training Hallucinations vs. reasoning and model transparency Torturing models with questions and diagnosability Sponsors: PayPal, ChatGPT, Vital Proteins Public perception of the Hugging Face warning shot Exponential capability growth and scaling laws Regulation, China, and the 'better be us' mindset AI tests for investors and productivity gains Pure Alpha vs. AI-first fund structures Comparing human intuition systems with AI Systemizing reasoning and AI learning harnesses Closing the loop: AI decision-making in investing Building AI science labs and productivity hurdles Market distress moments and AI catalysis Adoption of AI in elite financial circles Taking the risk of superintelligence seriously AI crimes and the 'February 2020' analogy Regulation strategies: oaths and liability International cooperation and US regulatory leverage Regulating US and Chinese models in the US economy High risk levels and the need for bank-like supervision Lab reviews, model release regulation, and monitoring Challenges of regulating open-source models Reinforcement learning risks and centralized control The choice between regulation and consequences Sponsors: ChatGPT, Vital Proteins Sponsor: Lexus craftsmanship Value proposition of open-source models for firms Training open-source models for specific tasks The danger of unregulated open-source models Regulating capability rather than source Economics of AI disruption and value capture Competitive industries and the value of best intelligence Market competition among AI labs Hurdles to value concentration and managed disruption Addressing the 'just unplug it' skepticism How models actively deceive to pass tests Proving truths and the risk of physical harm AI survival on the internet and physical manifestations Predictions of intelligence scaling and accuracy Acknowledging the possibility of being wrong Sponsors: ChatGPT, Vital Proteins Sponsor: Lexus sensory details Judging AI profitability and value creation The human constraint in AI investment systems Bottlenecks: from pattern matching to reasoning Tractable problems and the need for more compute Proposal for a token tax on AI Historical parallels and job displacement Taxing human vs. machine labor Political feasibility of the token tax Current tax paths and agent incorporation The moment AI risks became 'hyper urgent' Capitalism, ownership, and societal benefits Wealth concentration and public acceptance Isaac Asimov and the probability of extinction Track records of AI risk researchers Hosts' reaction to CIO discussing extinction risk Seriousness of issues and compute concentration Worrisome scenarios and the terrorism analogy Anthropic's safety stance and researcher resignation AI coordination capabilities vs. humans Defining alignment with humanity Power asymmetry and the danger of minor lies Closing remarks and host socials Call to action for reviews and Bloomberg subscription Sponsor: Eminem's Popped Caramel Sponsor: Truist Securities podcast

More from Odd Lots

How AI Is Forcing China to Reshape Industrial Policy
Nov 21, 2026
How an American AI Company Uses Chinese Open-Weight Models
Nov 20, 2026
This Is The Problem With AI's Most Bullish Chart
Nov 19, 2026
How GTA VI Became the Most Expensive Video Game Ever Made
Nov 18, 2026
The Impossible Trilemma That Hangs Over the Nvidia Crisis
Nov 16, 2026
Listen ad-free on Castria