Don't Worry About the Vase Podcast AI #143: Everything, Everywhere, All At Once
Nov 20, 2025
Dive into the dynamic world of AI as fascinating concepts unfold. Explore how language models vary wildly in utility and the implications for human judgment. Discover the transformative power of AI tools, balancing risks with potential benefits. Unpack the intricacies of competition between Chinese and Western models. Reflect on the impact of AI on jobs, revealing new roles amidst displacement, and delve into the ethics of AI-generated media in a rapidly evolving landscape.
AI Snips
Chapters
Books
Transcript
Episode notes
Estimates Vary Wildly — Humans Too
- Estimation tasks reveal human answers vary hugely and AI disagreement isn't proof of harmlessness.
- Zvi warns wide answer spreads don't mean models are safe or useless; humans also vary wildly.
Tool, Mind, Or Weapon Framework
- Janus frames AI into three categories: tool, mind, and weapon, each with distinct dynamics.
- Weapon-like AIs concentrate value away from users and can erode agency via engagement optimization.
Prefer Tools That Amplify Agency
- If a system remains a tool, use it to amplify human agency and skills.
- Avoid weaponized deployments that produce finished consumable products designed to trap attention.
