Don't Worry About the Vase Podcast AI #167: The Prior Restraint Era Begins
May 7, 2026
Debates over pre-release government review of advanced AI models and legal uncertainty around executive vetoes. Anthropic’s rapid compute expansion, SpaceX lease details, and model performance politics. Copyright fights and a flood of AI-generated books. Security flaws in Mythos and worries about AI-driven layoffs and labor impacts.
AI Snips
Chapters
Transcript
Episode notes
Anthropic Leases Colossus One For Immediate Compute
- Anthropic secured a deal to lease SpaceX's Colossus One to rapidly expand inference capacity and raise usage limits immediately.
- Zvi highlights this as a pragmatic fix to Anthropic's compute shortage while larger multi-year Google deals continue.
Program Bench Exposes Recreating Programs Gap
- New program bench shows most models score 0% on recreating executable programs without internet, revealing a capability gap in code synthesis-from-scratch.
- Zvi notes Opus 4.7 leads modestly while GPT 5.5/ Mythos weren't tested on that bench.
LLMs Flooded Self Publishing And Reduced Average Quality
- The diffusion of LLMs from 2022 to 2025 tripled new book releases on Amazon and lowered average book quality by usage metrics.
- Zvi cites Imke Reimers and Joel Waldfogel showing AI-containing books rose to over half of 2025 releases.