
In this episode of The Information Bottleneck, Ravid Shwartz-Ziv interviews Yann LeCun, a Turing Award winner and AI pioneer, about his new startup, Advanced Machine Intelligence (AMI), and its focus on world models. LeCun discusses the shift in AI research from open, collaborative environments to more secretive, industry-driven labs, and AMI's commitment to open-source research. He elaborates on the limitations of current LLMs, particularly in handling high-dimensional, continuous, and noisy data, and advocates for the use of world models and techniques like JEPA for building more robust and reliable AI systems. The conversation explores the history and evolution of unsupervised learning, the importance of abstract representation spaces, and the challenges of preventing collapse in joint embedding architectures. LeCun shares his perspectives on the role of synthetic data, the grounding of AI in reality, and the necessity of incorporating constraints and guardrails to ensure the safety and ethical use of AI technologies. He also touches upon the competitive landscape between the US, China, and Europe in AI research and development, and his personal motivations for continuing to push the boundaries of AI despite his numerous accolades.
Sign in to continue reading, translating and more.
Continue