General Intuition, a startup that leverages video game data to train AI agents, has secured $320 million in funding at a $2.3 billion valuation. The company's approach involves using human action data from gameplay to develop agentic models that can generalize from simulation to real-world environments. By utilizing hundreds of millions of hours of uploaded gameplay, General Intuition's model can learn spatial-temporal reasoning and understand how to move through space and time. The key ingredient in this process is the action labels embedded in gameplay clips, which provide a more accurate understanding of causality than inferring actions from video alone. General Intuition's technology has impressed investors, including Khosla Ventures, General Catalyst, and notable individuals like Jeff Bezos and Eric Schmidt. The funding will be used to scale compute capacity and pre-train the next version of the model, with a focus on making its API more broadly available. The company's vision is to create a generational company that enables others to build on top of its technology, with a focus on ethics and responsible AI development. General Intuition's CEO, Pim de Witte, has drawn a clear line on the use of its technology, stating that it will not be employed to harm humans. The company is also exploring ways to create a data flywheel, where it can collect and utilize real-world data from customers to improve its models and drive research forward.

Source: TechCrunch