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:
RTX/RT2:
Interactive Agents:
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:
Apple Podcasts:
IHeartRadio:
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
Music composition: Eleni Shaw
Camera Director and Video Editor: Tommy Bruce
Audio Engineer: Perry Rogantin
Video Studio Production: Nicholas Duke
Video Editor: Bilal Merhi
Video Production Design: James Barton
Visual Identity and Design: Eleanor Tomlinson
Commissioned by Google DeepMind
___
Subscribe to our channel
Find us on X
Follow us on Instagram
Add us on Linkedin
source