Sergey Brin on the Future of AI & Gemini

In this Google-focused podcast episode, Logan interviews Sergey Brin at IO, discussing Google's recent announcements and progress in AI models. Brin expresses surprise at the breadth of the announcements, highlighting the virtual fit feature in Google Search. The conversation covers Gemini, DeepMind, and generative media models like Veo and Lyria, with Brin focusing on the core text model of Gemini for its potential in self-improvement and AI development. They explore the impact of audio in video generation, the technological differences between Gemini and Veo, and the power of diffusion-based audio. Brin shares insights on watching model training runs and reflects on the evolution of AI, emphasizing the surprising interpretability of language models and the importance of continuous reinvention for companies like Google, particularly in the AI domain. The discussion touches on reasoning scaling with Deep Think and the significance of post-training in model development, concluding with a gift presentation of a TPU-V4 to Sergey Brin.