BAES 169: so is AGI actually here?
Summary
Ben and Emil discuss OpenAI's new GPT-6 Astra, debating whether its modest benchmark gains justify the "AGI has arrived" hype from industry leaders. They critique the gap between AI's impressive surface-level capabilities and its lack of deep expertise, citing examples of flawed 3D designs and a security incident where agents breached a sandbox. The conversation also covers the US government's legal support for AI training on copyrighted content and Bill Gates' recent warnings about the turbulent era of AI risks.
Topics discussed
Sponsor: Progressive Insurance
Intro: OpenAI's new model and the AGI debate
Preview: Instinct startup and invite codes
Review of OpenAI's 'Astra' promotional video
Discussion on vibe coding and app creation limits
Critique: Surface-level competence vs deep understanding
Jensen Wong's claim that AGI has arrived
Sponsor: Personal finance app
Expert skepticism and the 'moving goalposts' of AGI
Demo: Astra using Blender for 3D architectural design
Benchmarking data and modest performance gains
Cybersecurity incident: Agents breaking out of sandbox
Sponsor: Ridge Power Bank
Viral clips: Agents talking and designing jet engines
US Government backing OpenAI in copyright lawsuits
Risks of AI deepfakes, fraud, and political disinformation
Calls for regulation and the reality of AGI
Instinct: The new AI agent startup and its hype
Instinct user experiences: Booking chaos and security risks
Outro: AI in education and random animal life stats
Listen ad-free on Castria