Sep 10, 2026 · 1h 21m
AI safety warnings collide with incentives and evidence
OpenAI's chief scientist wants to slow down | E30
The episode examines whether frontier labs’ warnings describe genuine technical risks, strategic positioning, or an inseparable combination of both.
- 1AI risk messaging can reflect sincere concern while also strengthening labs’ case for regulation and influence.
- 2Open-source models and task-specific routing are making expensive frontier systems less necessary for many corporate applications.
- 3The panel argues that emotionally charged AI claims require scrutiny of evidence, incentives, and the conclusions they seek to produce.
Don't miss
Tudor Achim explains the reported Navier–Stokes breakthrough and why a Lean-formalized proof that compiles would be mechanically verifiable.
The brief
OpenAI chief scientist Jakub Pachocki’s warning about an incomprehensible “alien mind” sets up the episode’s central question: genuine safety concern, strategic marketing, or both?
Tudor Achim, Alex Elias, and Ahmed Ahres debate reward hacking, blackmail scenarios, nuclear analogies, and whether frontier labs gain influence by emphasizing AI’s dangers.
The panel also reports that OpenAI apparently solved the Navier–Stokes problem with a Lean-formalized proof, making mathematical verification a concrete counterpoint to broad claims about superintelligence.
Later, the conversation turns practical: UBS plans to test AI skills in hiring, while companies increasingly use cheaper open models and route difficult tasks to frontier systems only when needed.
The episode’s throughline is media literacy: evaluate who makes an AI claim, what incentives they have, what evidence supports it, and what response they want.
Books & mentions
Some links are affiliate links — PodLume may earn a commission if you buy.
