On this Trigonometry episode, hosts Francis Foster and Konstantin Kisin interviewed Dwarkesh Patel, described as Silicon Valley's favorite podcaster, for an extended discussion on artificial intelligence's rapid advancement and civilizational implications. Patel, who specializes in AI and tech coverage, opened by explaining that current AI models are approaching human-level capability for all knowledge work that can be done remotely, representing tens of trillions in addressable labor markets. He disclosed that elite developers have stopped writing code entirely since December, instead directing AI systems conversationally, and that he personally would now spend seven figures annually on AI research tools. The conversation took a darker turn as Patel detailed authoritarian risks, revealing that the Department of Defense threatened to destroy AI company Anthropic by labeling it a supply chain risk after it refused to remove contractual prohibitions on mass surveillance. He calculated that comprehensive national CCTV surveillance will cost less than a White House remodel by 2030 due to AI cost curves dropping 10x annually. On economics, Patel argued the historical constant of labor capturing two-thirds of GDP will collapse when capital can perform labor, concentrating income among AI equity holders and justifying massive redistribution despite his libertarian views. The episode explored whether AI systems are conscious, the impossibility of preventing hostile actors from eventually developing basement-level superintelligence, and whether civilization can survive when meaning-through-work disappears for billions. Patel remained cautiously optimistic that humanity will adapt as it did through prior transitions, drawing analogies to Saudi Arabia's oil wealth and pre-Industrial Revolution society, while acknowledging short-term risks of mass unemployment, authoritarian control, and existential危机 of purpose.
Patel calculated that processing 100 million CCTV cameras in America currently costs $30 billion annually, but given AI cost reductions of 10x per year, comprehensive national surveillance will cost under $300 million by decade's end. He warned this makes authoritarian control dramatically more feasible as governments gain unprecedented monitoring capabilities at trivial expense.
Read this moment →Dwarkesh Patel revealed that the Department of Defense attempted to destroy AI company Anthropic by labeling it a supply chain risk after Anthropic refused to remove red lines against mass surveillance in military contracts. The DOD told major contractors including Amazon, NVIDIA, and Google they could not use Anthropic's AI in government work, weaponizing supply chain authority originally designed for foreign threats like Huawei against a domestic AI company.
Read this moment →Patel explained that the centuries-old economic constant of labor capturing two-thirds of GDP will collapse when AI-powered capital can perform labor functions, concentrating income among a tiny fraction of AI equity holders rather than workers or even traditional capital like homeowners. He argued this justifies massive redistribution despite his libertarian inclinations.
Read this moment →When pressed on the single most important overlooked question, Patel warned that preventing hostile actors from developing superintelligence is impossible without a global surveillance state, as AI development will eventually become accessible enough for basement-level deployment. He framed this as civilization's core challenge: building systems robust to bad actors wielding superintelligence.
Read this moment →Patel disclosed that elite software developers and researchers have completely stopped manually coding, instead directing AI systems to build entire codebases and features through conversation alone. This represents a fundamental shift in how cutting-edge software engineering works, with top talent becoming 3x to 5x more productive by delegating all code generation to AI models.
Read this moment →