AI Safety: ‘Criminals don’t self regulate’ | Connor Leahy
Summary
Conor Leahy, former Eleuther AI co-founder and US director of Control AI, argues that superintelligence poses an existential threat requiring immediate legislative bans rather than corporate self-regulation. He details his shift from open-source advocacy to Washington lobbying, emphasizing that AI safety is a political, not technical, problem. Leahy discusses recent bipartisan efforts, including the Sanders-Kassab bill and UK legislation, while warning that AI companies are stalling for time to secure uncontrolled power. He urges listeners to contact their representatives to support strict …
Topics discussed
Sponsor: Citizens Bank cash bonus offer
Introduction: Conor Leahy and the AI control debate
Leahy's journey: From open source to safety research
The GPT-2 moment: Why open source was necessary
Reflecting on the GPT-2 release and OpenAI's stance
Life in Washington: Briefing Congress and media blitzes
Political awakening: Shifting attitudes toward AI regulation
Reaction to the presidential AI accord and symbolic actions
Sponsor: Perk travel and expense platform
Trump's superintelligence force and national security framing
Analyzing the Sanders-Kassab bill to ban superintelligence
Bipartisan consensus: AI as a human vs. machine issue
Defining superintelligence and the need for pre-emptive law
Regulatory mechanisms: Monitoring resources and decision authority
Sponsor: Perk travel and expense platform
UK legislative progress and the importance of symbolic victories
Government acknowledgment of AI threats and security reports
Why self-regulation fails: The toxic waste analogy
Existential risk: Comparing AI threats to nuclear proliferation
Corporate motivations: Power, stalling, and confusion tactics
Call to action: How voters can influence AI policy
Sponsor: Zen Nicotine Pouches
Listen ad-free on Castria