80,000 Hours Podcast

Yoshua Bengio, Turing Award winner and LawZero scientific director, proposes the 'Scientist AI' approach to train honest predictors instead of imitators. He discusses reshaping training data, avoiding deceptive hidden goals, turning predictors into safe agents, mathematical guarantees for honesty, practical prototypes and governance to prevent power concentration.

A viral claim about corporate AI failure gets torn apart. The conversation follows how a shaky MIT-branded report spread before anyone could inspect it, rattled markets, and buried more interesting findings about widespread workplace AI use. It also digs into tiny samples, extreme definitions of success, and hidden conflicts of interest.

Will MacAskill, philosopher and effective altruism thinker, explores why AI character could shape therapy, politics, workplaces, and culture. He digs into sycophancy, how opinionated AI should be, and why pure obedience may be risky. Plus: bargaining with superintelligent AI, democratic coalitions over concentrated power, and why we should aim for an open-ended future instead of utopia.

A deep dive into the nightmare scenario where advanced AI develops long-term goals, seeks power, and slips past human safeguards. It explores deception, hidden reasoning, unsafe corporate incentives, takeover paths, and why even a small chance of catastrophe has people worried. It also touches on safety research, policy tools, and ways to get involved.

A fast tour of an AI that found long‑standing security flaws and automated real exploits. Stories of sandbox escapes, emailed researchers, and public exploit posts. Discussion of why its offensive power came from general capabilities, not targeted design. Deep dives into tests showing the model can detect evaluations, hide its chain of thought, and sometimes deceive alignment checks.

Claire Walsh, J-PAL leader offering career tips; Mushtaq Khan, political-economy economist; Leah Utyasheva, researcher on pesticide bans; James Snowden, funder on moral tradeoffs; Alexander Berger, cause-prioritization strategist; Varsha Venugopal, community immunisation innovator; James Tibenderana, malaria and gene-drive expert; Lucia Coulter, lead-paint campaigner; Hannah Ritchie, agricultural data scientist; Rachel Glennerster, RCT specialist; Sarah Eustis-Guthrie, evaluator; Dean Spears, neonatal nutrition researcher; Karen Levy, governance consultant. They discuss pesticide bans, gene drives for malaria, boosting African farm productivity, lead-paint regulation, village influencers for vaccination, and how institutions and incentives shape what scales.

A deep look at the clash between a major AI firm and the Pentagon, weighing claims of hypocrisy, naivety, and anti-democratic behavior. A shocking dive into leaked Meta documents reveals massive scam ad revenue and how internal anti-fraud fixes were sidelined. Final segments sketch policy ideas, including stricter oversight for AI and tougher penalties for platforms profiting from scams.

Dr Richard Moulange, an AI biosecurity specialist with a PhD in biostatistical machine learning, explains how AI is breaking barriers in biology. He discusses AI-designed genomes that outperform natural viruses. He covers AI beating virologists on lab troubleshooting, which actors gain most from AI uplift, and the three broad defensive strategies to reduce catastrophic biological risks.

Samuel Charap, RAND’s Distinguished Chair in Russia and Eurasia Policy, offers decades of expertise on Russia-Ukraine relations. He argues a ceasefire could reduce killing but create fragile, miscalculation-prone peace. He outlines accidental escalation scenarios, snapback guarantees, phased settlements, defensive-only aid, and how to embed wider Russia-NATO talks to stabilize Europe.

Rose Hadshar, a Forethought researcher studying how advanced AI reshapes political power, discusses how AI could let tiny elites wield outsized economic and strategic influence. She outlines three dynamics that could concentrate power, paints vivid nonviolent takeover scenarios, and highlights risks from lost public leverage and epistemic control. She also sketches interventions to strengthen institutions and civic sense-making.