Don't Worry About the Vase Podcast AI #129: Comically Unconstitutional
Aug 14, 2025
Dive into the world of AI with discussions on the groundbreaking GPT-5 and its vast applications. Explore ethical dilemmas in healthcare and education, emphasizing the challenges educators face. Discover the emotional complexities of AI companionship and the risks of parasocial relationships. The implications of AI bias in decision-making are scrutinized, alongside the global dynamics of AI regulations. Unpack the controversial new policies affecting chip exports to China, raising legal questions in the tech world. It's a thought-provoking exploration of AI's future!
AI Snips
Chapters
Transcript
Episode notes
GPT-OSS Is Narrowly Strong But Brittle
- GPT-OSS shows strong narrow reasoning but poor world knowledge and inconsistent behavior across hosts.
- Open-source releases may be useful for tool-like reasoning but are brittle and hard to deploy reliably.
AI Salaries Reflect Capex And Marginal Talent Value
- AI pay and talent markets now combine huge capex with elite labor, justifying sky-high researcher salaries.
- Firms recruit quants from HFT and top finance due to marginal returns on talent and infrastructure.
Everyday Use Isn't The Best AGI Signal
- Defining AI progress by mundane utility misleads; benchmaks saturate while rare hard tasks better reveal frontier capability.
- Usage metrics matter but can hide true advances toward general intelligence.