DeepMind's SIMA project introduces large-scale cross-embodied language-conditioned agents in various video game domains, showcasing positive transfer where generalist agents outperform specialists. The project aims to build societies of AI agents with emergent intelligence and reasoning, revolutionizing embodied agents in virtual worlds.
It’s wonderful to see this work coming to fruition. Congratulations SIMA team. I want to use this opportunity to highlight @scott_e_reed who after finishing Gato (a generalist agent) worked tirelessly with our colleagues to build video agent models and code infrastructure that… https://t.co/su8yFKKYK2
SIMA paper 'Scaling Instructable Agents Across Many Simulated Worlds' out today from @GoogleDeepMind ! This is revolutionary for embodied agents in virtual worlds - so excited for non goal oriented simulations like Sim Francisco. https://t.co/O3qLhkdcse https://t.co/If9GhucR7z
Large scale cross-embodied language-conditioned agents in a variety of video game domains! What’s particularly exciting is seeing positive transfer: generalist agents outperform specialist agents. https://t.co/YZzV5dNKCE
Yayyy! SAGA (Skill to Action Generation for Agents) covered by the great @80Level We’re building societies of AI agents with emergent intelligence & reasoning: Westworld with the good parts and not the bad parts… Devs &Hackers sign up to SAGA below! https://t.co/jFnP7JLDdK
Really cool to see an agent generalizable across 3D environments since they’re such a rich data environment. So many possibilities but still super early - without human language inputs performance degrades significantly https://t.co/H8SyfUyOFL
Excited to share a project I have been working on at DeepMind. Congrats to the entire, wonderful SIMA team! I look forward to hearing what the community thinks of this advance. https://t.co/x3Ru40piCS