DeepMind on Gemini 3.5 Flash, Omni Video, Agent Harness & AI Strategy
In this in-person episode recorded at Google headquarters, DeepMind's Logan Kilpatrick and Tulsee Doshi discuss the upcoming launch of Google's Gemini 3.5 Flash model and related AI product integrations announced at Google IO 2024. They emphasize the strategic focus on cost-effective, fast models like 3.5 Flash that balance performance and latency to serve billions of users across Google's diverse product ecosystem. The conversation highlights the integration of models with a robust agent harness infrastructure, enabling standardized, agentic AI experiences across Google products such as the Gemini app, AI Studio, and search. They also introduce Gemini Omni, a new video generation and editing model aiming to create a 'nano banana moment' for video by tightly integrating language, reasoning, and pixel-level understanding. The guests discuss Google's approach to recursive self-improvement, model psychology, and welfare, emphasizing a collaborative human-AI research process rather than fully autonomous AI-driven development. They address why context windows have plateaued, the rationale behind model size choices (no ultra model yet), and the ongoing research into diffusion models and multimodal AI capabilities.
Original sources
Podcast The Cognitive Revolution
YouTube The Cognitive Revolution
AI Radar wrote this summary from the publisher’s notes with AI, not from a transcript. How it works.