Asimov: Building An Omniscient RL Oracle with ReflectionAI’s Misha Laskin
No Priors: Artificial Intelligence | Technology | Startups
In this episode of No Priors, Sarah interviews Misha Laskin, co-founder of ReflectionAI and former leader at Google DeepMind, about building superintelligent autonomous systems. Misha discusses ReflectionAI's mission to co-design products and research, focusing on solving real-world problems with their code comprehension agent, Asimov. He highlights the importance of understanding and context in coding, contrasting it with the current industry's overemphasis on code generation. Misha shares insights on evaluating AI systems, the challenges of scaling reinforcement learning, and the future of AI deployment in various fields, emphasizing the need for verticalization and a focus on customer needs. He also touches on the Windsurf non-acquisition and the potential for AI to drive significant productivity increases across industries.
Part 1: Introduction and Background
Part 2: Asimov: Design, Functionality, and Evaluation
Part 3: Technical Deep Dive and Challenges
Part 4: Future and Conclusion
Sign in to continue reading, translating and more.
Open full episode in Podwise