SIMA & general intelligence | Frederic Besse

Games are a very good training ground for agents. Think about it. Perfectly packaged, neatly constrained environments where agents can run wild, work out the rules for themselves, and learn how to handle autonomy. In this episode, Research Engineering Team Lead, Frederic Besse, joins Hannah as they discuss important research like SIMA (Scalable Instructable Multiworld Agent) and what we can expect from future agents that can understand and safely carry out a wide range of tasks - online and in the real world. Timecodes 00:00 Intro 02:11 Definition of an agent 03:46 Autonomy 06:59 Virtual and embodied agents 08:11 Differences between agents and large language models 11:04 Games as a proving ground 12:09 NPCs 13:42 AlphaStar, AlphaGo, AlphaZeo 17:37 Deep dive into SIMA 18:59 Sandbox games 22:30 Goat simulator 3 and demos 29:24 Generalizing to unseen environments 30:30 Results of SIMA 35:02 Progressing the research towards AGI 39:49 Examples of future agents 42:29 Reflections from Hannah Further reading SIMA: https://deepmind.google/discover/blog/sima-generalist-ai-agent-for-3d-virtual-environments/ RTX/RT2: https://deepmind.google/discover/blog/scaling-up-learning-across-many-different-robot-types/ Interactive Agents: https://deepmind.google/discover/blog/building-interactive-agents-in-video-game-worlds/ Want to share feedback? Have a suggestion for a guest that we should have on next? Leave us a comment and stay tuned for future episodes. ___ Search for Google DeepMind: The Podcast on: Spotify: https://open.spotify.com/show/39fjU5Q5L5UecTCRMeqjwb Apple Podcasts: https://podcasts.apple.com/gb/podcast/google-deepmind-the-podcast/id1476316441 IHeartRadio: https://www.iheart.com/podcast/269-deepmind-the-podcast-48983807/ Thanks to everyone who made this possible, including but not limited to: Presenter: Professor Hannah Fry Series Producer: Dan Hardoon Editor: Rami Tzabar, TellTale Studios Commissioner & Producer: Emma Yousif Production support: Mo Dawoud M