Latest Interviews
Showing 1–3 of 3 transcripts.
Clear all filters- 80,000 Hours2h 14m
How to avoid catastrophic nuclear blunders | Joan Rohlfing (2022)
In October 2021, expert Joan Rolfing presented "Beyond the Precipice," arguing that global nuclear risk has reached unprecedented levels due to the inherent fragility of the Cold War-era deterrence strategy and escalating cyber vulnerabilities. Following the Russian invasion of Ukraine in early 2022, Rolfing further detailed how the collapse of arms control treaties and the concentration of launch authority have doubled the probability of catastrophic nuclear use. She advocates for a fundamental paradigm shift toward fail-safe systems, the integration of advanced verification technologies, and significant philanthropic investment to replace the current trillion-dollar modernization trajectory with sustainable risk reduction measures.
- 80,000 Hours3h 27m
Navigating serious philosophical confusion | Joe Carlsmith
Philosopher Joe Carlsmith argues that while complex concepts like simulation theory and infinite ethics are intellectually compelling, they should be approached with caution to prevent "epistemic learned helplessness." He advocates for "wisdom long-termism" as a practical alternative to welfare-focused approaches, prioritizing the creation of a future civilization capable of solving deep philosophical uncertainties rather than locking humanity into speculative moral frameworks. Carlsmith further emphasizes maintaining emotional resilience and common-sense morality, asserting that human-scale concerns remain valid even when cosmic realities challenge traditional utilitarian calculations.
- 80,000 Hours3h 52m
Solving the alignment problem and handing off the future to AI | Paul Christiano
OpenAI researcher Paul Christiano outlines a technical framework for aligning artificial intelligence with human values, prioritizing "prosaic" approaches like iterated amplification over speculative future methods to address the risk of an AI-dominated economy. He argues that competitive market pressures will likely force a gradual "slow takeoff" over two decades, necessitating robust verification mechanisms to prevent a race to the bottom on safety standards. The discussion concludes with strategic recommendations for the field, emphasizing the need for institutional design, funding flexible high-risk research, and focusing on engineers who can scale safety theories within existing deep learning architectures.