@AnimocaVentures
Training is how you create intelligence.Inference is how the physical world uses it.That’s why the bottleneck flipped.A training run ends. A robot does not. Every Optimus, every FSD car, every factory policy is batch-1 decode against a power, latency, and thermal wall. GPUs were built for the first problem. Etched is built for the second: tokens per watt, not FLOPs on a slide.The scarce resource in physical AI isn’t another pretrain. It’s sustained inference that doesn’t miss a control loop. Disagree?View quoted post on X
time.com TIME100 AI 2026: Yann LeCun Find out why Yann LeCun made TIME’s list of the most influential people in artificial intelligence Games have been an important testbed for our AI research for over 15 years. 🎮
From mastering Atari to reaching Grandmaster in StarCraft II, they have driven some of our biggest AI breakthroughs.
Our work with SIMA taught agents how to understand 3D worlds, but learning to navigate real human dynamics takes a living, persistent universe.
Through our research partnership with @FenrisCreations, we’re exploring how to tackle open challenges in AI:
🔵 Continual learning to acquire new skills without forgetting past knowledge.
🔵 Deep memory systems that store and retrieve information far beyond today’s context windows.
🔵 Long-horizon planning over weeks, months, or years.
🔵 Multi-agent dynamics spanning cooperation, negotiation, economics, and emergent behaviors.
Our long-term goal is to use AI to discover entirely new gameplay experiences in partnership with game developers – making games more accessible and personalized – while applying what we've learned to problems in the real world and scientific discovery.
Find out more → goo.gle/4xcwzOHView quoted post on X
#100toWatch company Botinkit in China is digitizing the food service industry with its OMNI robot chef.View quoted post on X
Click here for the full Forbes Asia 100 to Watch list: forbes.com/100toWatch
Sponsored by @FedEx
