Nando de Freitas, research director at DeepMind, talks with Pieter Abbeel on The Robot Brains Podcast about making AI useful and accessible worldwide, generalist agents, robotics, and future research.
De Freitas sees AI as a tool that can extend what people can do, from creating knowledge to tackling energy and environmental problems. He describes volunteering with Deep Learning Indaba and Khipu as a way to help students build confidence, grow local research communities, and lead AI efforts in their regions. He argues that inclusive communities and attention to local needs matter alongside access to large models.
De Freitas says important AI work can happen outside large-scale model training, including fundamental research and making existing models cheaper to use and adapt. He gives the example of a health advice service accessed by phone in South Africa, where simple guidance could help address dehydration when doctors are scarce. He suggests language models could extend people’s ability to provide basic help, while emphasizing that useful applications depend on understanding local problems.
De Freitas describes AlphaCode as a transformer system that generates many possible solutions to programming problems and narrows them down; he says it performed remarkably well but still produced unnecessary code. He explains that Gato aimed to handle varied inputs and tasks, from dialogue and images to controlling a robot arm, by representing them as sequences of tokens for one model. He also discusses learning from online videos, including Atari gameplay, as a way to transfer skills without relying only on expensive reinforcement learning or robot demonstrations.
De Freitas reflects on learning algorithms that can themselves be learned, connecting earlier meta-learning research with few-shot learning in large language models. He describes experiments where agents develop discrete signals to coordinate, while noting that compositional communication remains difficult. Looking ahead, he hopes AI can support scientific reasoning and tools such as AlphaFold already show promise, while robotics could make tedious or dangerous work easier and help people focus on more meaningful tasks.
Briefs summarise talks and interviews; timestamps open the video at that moment. Claims in a talk are the speaker's own.
Geoffrey Hinton, AI researcher and former Google employee, talks with host Pieter Abbeel on The Robot Brains Podcast about AI’s growing capabilities, risks, potential benefits, and his decision to leave Google.
Weights & Biases founder and CEO Lukas Biewald demos the company’s machine learning tools for Robot Brains Podcast host Pieter Abbeel.
Pieter Abbeel, director of Berkeley’s Robot Learning Lab and co-director of BAIR, presents on deep learning for robotics at a CMU Robotics Institute seminar hosted by Dave.
