A look at startups like General Intuition working on large action models, aka world models, which are trained on videogames and simulations, to pilot robots (Christopher Mims/Wall Street Journal)

General Intuition and similar startups are developing large action (world) models trained on video games and simulations to enable robots to perform complex tasks, mirroring the impact of ChatGPT on language work. Nvidia’s Groq 3 LPX inference accelerator has entered full production, delivering 3,400 tokens per second on a Gemma-4 31B benchmark with a 100 k-token input, and its first customer Nebius has begun deployment.