The company, founded a year ago by Sun and chief scientist Henry Liu out of the University of Michigan, grew from academic work on challenging autonomous driving AI in simulation. Their 2023 paper on the subject made the cover of Nature, and interest from autonomous vehicle companies followed. The name nods to Laplace's demon, an 1814 thought experiment about predicting the future with total precision; Sun says the startup is "essentially using simulation to predict the real world's future".

At its core is a high-fidelity simulator built on a post-trained version of Nvidia's Cosmos, reconstructed from real-world map data, Street View imagery and live traffic information from providers such as TomTom. Developers can adjust scenarios with language prompts, such as adding afternoon traffic. In a demo, when a test vehicle was directed to swerve into oncoming traffic, the oncoming car moved to avoid the collision, which Sun presented as evidence of a genuinely interactive world model rather than a static one.

The differentiator, the company argues, is AI agents that automatically design corner cases, the tricky rare scenarios that can trip up an AI system, drawing on sources like news and police collision reports. Autonomous vehicle developers today mostly rely on recorded footage turned into simulations or on humans inventing scenarios, and some AV firms such as Waabi already use AI to generate edge cases. Sun contends the need will sharpen with generalized robots, where there is less data and no one knows what the edge cases will be.

Laplace is one of 20 startups competing in the Startup Battlefield 200 pitch competition at TechCrunch Disrupt in San Francisco, presenting on the Disrupt Stage on Tuesday 13 October.