#175 - GPT-4o Mini, OpenAI's Strawberry, Mixture of A Million Experts
Summary
This episode covers OpenAI’s launch of the cheaper GPT-4o Mini and its secret "Strawberry" reasoning project. The hosts discuss Hyper’s new video model, Anthropic’s Android app, and Google’s AI presentation tools. They also analyze XAI’s controversial Memphis supercomputer build, data scraping controversies involving YouTube transcripts, and Andrej Karpathy’s new education startup.
Topics discussed
Sponsors: OutShift and Box
Intro, host banter, and listener comments
Anthropic Claude 3.5 Sonnet release and pricing
Runway Gen-3 Alpha and video generation news
Google Gemini 1.5 Pro and AI music generation
OpenAI's reasoning models and 'Strawberry' project
xAI's Memphis supercomputer and energy needs
OpenAI training on YouTube data and Tesla AI
Menlo Ventures fund for AI startups
Mamba, Mistral, and Cosmopedia datasets
Stability AI licensing changes and business model
Flash Attention 3 and inference optimization
Research: MoE, lifelong learning, and benchmarks
Scalable oversight and prover-verifier systems
Trump's AI executive order and policy shifts
US export controls on AI chips to China
OpenAI whistleblower lawsuit and outro
Listen ad-free on Castria