General Intuition Inc. today announced that it has raised $220 million in funding at a $6.2 billion valuation.
The capital was provided by Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures and General Catalyst. The investment comes a year after General Intuition spun out of video sharing startup Medal B.V.
Medal provides a free tool that consumers can use to capture and record video game footage. General Intuition uses that footage to train its artificial intelligence models.
Developers train AI models by providing them with examples of the tasks they will be expected to perform. For example, an algorithm tasked with powering a robotic arm might be given footage of robotic arms welding car parts. Manually recording such footage can be a highly time-consuming process.
General Intuition is working to ease the task. It develops world models that can generate synthetic footage for AI training projects. The company released its newest algorithm, MIRA, in June. It says the AI is more capable than earlier models in several respects.
Traditional video generators can only render clips that are a few seconds or minutes long. As a result, they can’t generate synthetic videos of long-running factory automation workflows. That limits their usefulness for robot developers.
MIRA, in contrast, can “run infinitely without diverging.” Furthermore, it’s capable of rendering complex scenes in which multiple fast-moving objects interact with one another. The ability to simulate movement is useful for tasks such as training robots to avoid collisions.
MIRA generates footage in a highly efficient manner. According to General Intuition, it can render 20 frames per second with a resolution of 720 by 576 pixels using a single B200 graphics card.
The model’s efficiency partly stems from its use of a processing approach called latent diffusion. Whereas traditional media generators process video frames, MIRA runs calculations on a data structure called a latent space. It’s a compressed representation of video frames that requires less memory. Shrinking the memory footprint of a file reduces the amount of time required to process it.
MIRA’s size also factors into its efficiency. The algorithm comprises only 5.6 billion parameters, a fraction of the number included in frontier models.
MIRA is a research demonstration rather than a commercial training data generator. In its current iteration, the model can only generate synthetic footage of a single video game. Nevertheless, General Intuition says, MIRA’s efficiency and rendering quality make it a “stepping stone to physical AI.”
The company is testing a commercial version of its technology with a limited number of customers who are working on robotics, simulation and entertainment use cases. General Institution opened a waitlist for the offering in conjunction with today’s funding announcement. The company said it will use the capital to hire more AI researchers.