General Intuition Secures $220 Million to Enhance AI Training Technology
General Intuition Inc. has announced the successful closure of a $220 million funding round, valuing the company at $6.2 billion. This substantial capital injection comes from notable investors including Valor Equity Partners, Atreides, Seven Seven Six, Point72, Khosla Ventures, and General Catalyst.
This investment follows a strategic shift for General Intuition, which spun off from the video sharing startup Medal B.V. just a year ago. Medal is known for providing a free tool that enables users to capture and record video game footage, which General Intuition leverages to train its advanced artificial intelligence models.
In order to train AI models effectively, developers must provide ample examples of tasks for which the models will be responsible. For instance, an AI algorithm designed to control a robotic arm might require extensive footage of welding operations in automotive manufacturing. However, manually recording such footage is often time-consuming and resource-intensive.
General Intuition aims to alleviate this challenge by developing world models capable of generating synthetic footage for AI training applications. In June, the company unveiled its latest algorithm, MIRA, which surpasses previous models in several critical aspects.
Unlike traditional video generators that produce clips limited to a few seconds or minutes, MIRA can simulate workflows that run indefinitely without diverging. This capability makes it particularly valuable for developers engaged in robotics and automation, allowing for realistic simulations of complex scenarios where multiple objects are in motion, such as training robots to navigate around obstacles.
Efficient in its operation, MIRA can render footage at a rate of 20 frames per second at a resolution of 720 by 576 pixels when utilizing a single B200 graphics card. This efficiency is largely attributed to its application of a processing technique known as latent diffusion, which calculates on a data structure referred to as latent space. Latent space is a compressed representation of video frames that consumes less memory, thus expediting processing times.
Further enhancing its efficiency, MIRA consists of only 5.6 billion parameters, significantly fewer than those found in leading-edge AI models. Although currently a research demonstration rather than a commercial training data generator, MIRA can generate synthetic video footage for just one specific video game. Nonetheless, General Intuition describes MIRA’s rendering quality and efficiency as a promising “stepping stone to physical AI.”
The company is currently trialing a commercial version of its technology with select clients focusing on applications in robotics, simulation, and entertainment. With this funding announcement, General Intuition has also opened a waitlist for its commercial offering, indicating plans to utilize the capital for expanding its team of AI researchers.