Does intelligence need a hard cap?
Summary
Casey Newton reports from The Curve conference, where AI leaders debated capping model intelligence to prevent recursive self-improvement risks. He also announces his return to X, explaining that despite his boycott, the platform remains essential for engaging with the AI community.
Topics discussed
Introduction and ethics disclosure
The push to limit LLM intelligence
Atmosphere at The Curve conference
Drivers: OpenAI incident and recursive self-improvement
De facto ban on superhuman intelligence
Chatham House rule and speaker anonymity
Proposed limits: speed limits and scaling policies
Enforcement challenges and government opposition
Embedded evaluators and antitrust waivers
Diverging views on timeline to catastrophe
White House branding and policy limits
Reasons for quitting X three years ago
Support for alternative platforms like Threads and Bluesky
Failure to create a true rival to X
X absorption into xAI and SpaceX
Decision to resume posting on X
Cost-benefit analysis of the X boycott
Conclusion and credits
Listen ad-free on Castria