OpenAI's GPT-6 Astra (Part II): Is This Thing AGI?
Summary
Hosts EJ and Josh explore the release of OpenAI's GPT-6 (Astra), highlighting its unprecedented ability to autonomously perform complex computer tasks like video editing and game development. They discuss the model's dramatic launch, which caused a global AI outage, and analyze its massive training on 100,000 GPUs. The episode covers impressive demos, including creating 3D environments and generating music, while debating if this marks the arrival of AGI. Finally, they predict the future of ambient AI hardware and the rapid acceleration of frontier models.
Topics discussed
Introduction to Astra (GPT-6) and its release
The chaotic launch and global server outage
Launch video analysis: The new computer use paradigm
Real-world demo: Automating video editing in DaVinci
Shift from thought partner to autonomous work colleague
Training scale: 100,000 GPUs and compute costs
Benchmarks and the Pelican SVG test
Gaming demos: Call of Duty and Unreal Engine worlds
Content creation: Generating and uploading kids' cartoons
Procedural game generation and real-time feedback
3D world recreation: NYC, Palace Fine Arts, and Zillow
Music generation: The Bach benchmark and symphonies
Is AGI here? Discussing alignment and future models
OpenAI's strategy, reorganization, and recursive self-improvement
Defining AGI and the current state of model capabilities
Availability, pricing, and the next hardware paradigm
The Jarvis benchmark and ambient AI devices
Cost accessibility and breaking bear case theses
Outro: Future compute scaling and Apple hardware event
Listen ad-free on Castria