Descriptions:
Episode 248 of the Peter H. Diamandis podcast opens with a news item that landed just before recording: a 20-year-old Texan threw a Molotov cocktail at Sam Altman’s San Francisco home, with the suspect reportedly linked to the “official pause AI” Discord server. The hosts connect this to broader social unrest data — Maine has passed the first statewide data center ban in U.S. history, and polling shows only 23% of the public is optimistic about AI — framing these events as early signals of a widening gap between public anxiety and industry momentum.
A significant portion of the episode is devoted to Anthropic’s same-day release of Claude Opus 4.7. Resident benchmarking analyst Alex Wezer Gross calls it “moderately interesting” — solid on evals, particularly in biological benchmarks and computer use, but underwhelming relative to the leaked “Mythos” model. His technical read: the removal of temperature controls, explicit reasoning-token budgets, and other granular hyperparameters marks an architectural shift toward prompts-as-the-only-dial. Fellow guest Dave reports that Opus 4.7 successfully used computer use to self-install a new terminal and Node version, something he doubts 4.6 would have managed.
The episode also covers Amazon and Apple teaming up to build a satellite internet network to compete with SpaceX Starlink, and a discussion of Indian factory workers wearing camera-mounted headsets to train robotic replacements — touching on the ethics of using human labor to automate itself out of existence.
📺 Source: Peter H. Diamandis · Published April 18, 2026
🏷️ Format: Podcast
