Don't Worry About the Vase Podcast AI #147: Flash Forward
27 snips
Dec 18, 2025 The discussion dives into the utility of language models, with contrasting views on their practical benefits. Exciting upgrades in AI technology are highlighted, including improvements to coding capabilities and advanced media generation tools. The ethical implications of deepfakes and AI negotiations spark debate, while a Disney lawsuit against Google raises copyright questions. Predictions for AI's future emphasize efficiency boosts and the importance of sensible regulations amidst a potentially bubble-like tech landscape.
AI Snips
Chapters
Transcript
Episode notes
Update Your Priors With New Evidence
- Fixed Bayesian priors without updating contradict evidence; one must update on new data.
- Holding a 2009 belief unchanged despite new evidence misunderstands Bayes' rule.
Design Practical Slowdown Tools
- Prefer building a pause-button or slowdown capacity rather than blanket moratoria.
- Distinguish supersized clusters from consumer AI hardware when crafting policy.
Executive Order Offers Little Substance
- The Trump EO aims to preempt state AI rules via funding threats while promising a 'minimally burdensome' federal framework.
- Zvi expects most aspects to fail legally or accomplish little in practice.