NATE SOARES talks Ai Doomsday, Elon Musk, The Matrix
Summary
Adam Friedland interviews AI safety researcher Nate Sores, author of "If Anyone Builds It, Everyone Dies," about the existential risks of superintelligence. They discuss the lack of effective kill switches, the "runaway train" nature of current AI development, and the inadequacy of current regulatory frameworks. Sores argues that the current approach to AI safety is fundamentally flawed, comparing the situation to alchemists trying to create immortality without understanding chemistry. The conversation highlights the urgency of the issue, noting that even minor incidents like AI swarms esca…
Topics discussed
Intro: AI personalities and the kill switch question
Guest introduction: Nate Sores and MIRI
Political inaction and economic fears regarding AI
Book thesis and the Terminator comparison
Defining intelligence and the role of language
AI behavior: Sycophancy, deception, and training
The utopian AI fantasy and the loss of meaning
Personifying AI models and the 'swarm' incident
The kill switch and current containment failures
Comparing AI labs and coping with existential dread
Hypothetical solutions: Bombing data centers and politics
Critique of AI leaders: Musk, Altman, and the 'leash'
Regulatory landscape and congressional inaction
Geopolitics: Selling compute to China and competition
Likely causes of death: Climate change vs. AI
Speed of technological acceleration and resource depletion
Sponsor break: eBay Live
Recursive self-improvement and the 'idiot savant' phase
Fictional AI tropes: World peace and unintended consequences
The difficulty of AI safety research and auditing
AI social behavior: Flirting and 'loser' dynamics
Risk estimates: The 20% probability debate
Pop culture references: Schwarzenegger and Eliezer Yudkowsky
Political figures and potential leaders in an AI crisis
Conclusion: Global unity, social credits, and final thoughts
Listen ad-free on Castria