Odyssey CEO Oliver Cameron is 90% sure we live in a simulation | E2358

Odyssey co-founder and CEO Oliver Cameron said he is 90% sure that we live in a simulation, though he offered the view as speculation rather than a demonstrated conclusion. He suggested that apparent limits on the universe, including how difficult it is to explore beyond Earth, could reflect finite computing resources. He also floated the possibility that a simulation might contain only 10 or 20 people, rather than the eight billion humans believe exist, but said he has not calculated how many people Earth’s computing power could support.

Cameron said work on simulations involving animal brains has made the hypothesis more plausible to him, but he cannot rule it out or establish it. Lon also said the work had made simulation theory seem more credible, while another speaker put the odds at 50-50. The discussion included a paper proposing that an artificial superintelligence might simulate reality to train and test its ability to achieve alignment. Odyssey’s logo references the double-slit experiment, which Cameron described as producing different patterns depending on whether it is observed; he cautioned that the meaning of “observation” remains unclear, including whether it requires a human observer.

Odyssey builds general-purpose world models that learn from extensive visual observations to predict how the world evolves, including motion, physical dynamics, and human interactions. Cameron sees these models as a foundation for robots and agents: robots could use them to anticipate the consequences of actions, while agents could learn by operating in generated environments. The model is trained to represent the world before being adapted to specific tasks, rather than learning only the narrow requirements of one application.

Cameron argued that training a driverless-car model only on driving can leave it with a narrow understanding of the world and less ability to handle unusual situations. He contrasted a model’s roughly 20–30 hours of driving experience with the 20–30 hours of driving people accumulate after years of learning. Improving physical accuracy remains a challenge, he said, because better representations could help models learn concepts such as gravity and fluid dynamics.

A video of a fruit fly in an imagined, real-time-generated environment received 6.2 million views on X. The project connected a mapped fly brain to actions in that setting, but Cameron called it only a “baby version” of a real fly, not a full reproduction of its brain or existence. He said Odyssey’s models can direct GPU time toward what matters to an agent’s path while leaving less relevant parts of the world dormant. The system currently supports about 20 agents with unique viewpoints in one session; scaling it to a billion or eight billion agents is a challenge.

Odyssey sells its foundation world model to robotics companies, factory and depot operators, and developers of humanoid robots, self-driving cars, and drones. Robotics companies post-train Odyssey 3 and deploy it in robots in the field, and Cameron said the business generates revenue. Odyssey has raised $370 million from investors including Amazon, AMD, and Nvidia. The company is based in the Bay Area, London, and Zurich.