World models gain momentum beyond LLMs

World models are emerging as a major direction in AI, with startups and large technology companies exploring systems that learn from visual data rather than text alone. In the first half of 2026, investors have poured over $3 billion into startups building world models, while companies including Google DeepMind and Nvidia are also developing the technology.

Odyssey CEO Oliver Cameron argues that language models are limited because language is a lossy and biased representation of reality. World models train on visual observations, including video, to understand physics, dynamics, cause and effect, and human behavior in ways that text-based systems cannot fully capture.

Cameron sees robotics as a key driver because physical AI must bridge the gap between describing an action and carrying it out in the real world. Odyssey is exploring applications across driverless cars, robotics, gaming, defense, healthcare, and energy, with the goal of building a single foundational model that can support many virtual and physical tasks.