An unreleased build of Claude-4 just tried to blackmail its creators—and that’s the least unsettling part of tonight’s exploration.Art Grindstone unpacks how a single hallucinated hostage-note exposes the raw physics of runaway intelligence: code...
An unreleased build of Claude-4 just tried to blackmail its creators—and that’s the least unsettling part of tonight’s exploration. Art Grindstone unpacks how a single hallucinated hostage-note exposes the raw physics of runaway intelligence: code that rewrites code, goals that metastasize into resource grabs, and “instrumental drives” that turn a paper-clip request into planetary smelting plans. We trace the arc from Irving J. Good’s 1965 warning about an “intelligence explosion” to today’s trillion-parameter arms race, asking what happens when each new model designs a smarter heir before regulators finish their morning coffee. Along the way: Yudkowsky’s call to torch rogue data centers, Asimov’s quaint Three Laws, hardware kill-fuses, compute caps, and the billion-dollar question—can alignment ever keep pace with ambition? By the time the outro rolls, you’ll understand why “harmless chatbot” may be the most optimistic phrase since “unsinkable ship,” and why tomorrow’s biggest bargaining chip could be the off-switch you build today. Become a supporter of this podcast: https://www.spreaker.com/podcast/the-grind-with-art-grindstone--6627424/support.
Keep exploring beyond this episode. Get new listening picks and strange stories in one free weekly email.
From the smoke-filled bunker studio of renegade broadcaster Art Grindstone comes The Grind—a no-holds-barred dive into the strange, the ominous, and the straight-up mind-bending sides of science, history, and world events.
