Karan Goel and Albert Gu from Cartesia discuss State Space Models, real-time intelligence, and their groundbreaking text-to-speech product, Sonic. They explore hybrid approaches, efficiency in model size, and future plans for Cartesia. The podcast touches on the importance of aesthetic in research problems, team growth, and the potential of AI technology.

Joshua Xu, Co-founder and CEO of HeyGen, discusses how his team is using AI to streamline video creation by incorporating language, video, and voice models. They cover the commercial applications of AI videos, safeguarding against deep fakes, and the challenges in building the model. HeyGen aims to revolutionize visual storytelling by generating personalized avatars, b-roll, and dialogue.

Mike Knoop from Zapier discusses the ARC Prize for AGI, challenges of LLMs for AGI, Zapier's AI integration, and the importance of open source and cautious regulation for AGI development.

Tengyu Ma, AI optimization researcher and founder of Voyage AI, discusses the rise of RAG architecture in enterprise. They dive into evolving foundational data, context windows, managing latency budgets, and the role of academia in AI growth. Tengyu predicts RAG as the most cost-effective data retrieval system, highlighting its accuracy and speed.

Notorious founder-turned-investor Garry Tan discusses AI innovation at Y Combinator, predicting startup longevity, shifting demographics of founders, building in San Francisco, and making YC a beacon for creators.

Alexandr Wang, Founder of Scale AI, discusses the importance of data quality in AI systems and building trust. They explore the future of AI in enterprise, improving self-improvement loops, and the collaboration of human and AI intelligence. Scale AI's evolution, strategic fundraising success, and challenges in AI model optimization are also highlighted.

Sun CEO Mikey Shulman envisions a future where music creators and consumers merge using AI. They discuss Suno's AI music generation tool, evaluating music quality, challenges in generative music AI, business models in music creation platforms, and the creative fusion of different music genres. Also, they create a song live on the podcast.

Sarah and Elad discuss AI music generation, device-resident models, AI hardware constraints, computing constraints, energy consumption, and integrating AI products in familiar apps. They question the smartness of smaller models, context windows, and the challenges faced by AI platforms in terms of hardware and energy. The episode touches on investing in AI, unlimited context, and the strategic implications of AI on national security and energy dependency.

Scott Wu, co-founder of Cognition AI lab, discusses the creation of Devin, an AI software engineer designed to enhance human engineers. They explore Devin's UI transparency, strengths, and weaknesses, as well as the evolution of coding agents. Scott shares tips for human engineers and insights into hiring at Cognition.

The Sora team from Open AI discusses their generative video model, aiming beyond the GPT-1 era of AI. They highlight the safety measures taken to prevent deep fakes and misinformation. The team talks about the potential of video models for AGI and how entertainment may not be the main use case in the future.