AIToday

General Intuition raised $320 million(約510億円) at a $2.3 billion(約3700億円) valuation to train AI agents using video game footage, betting that gameplay is a scalable way to teach machines how to act in the physical world.

TechCrunch AI10h ago6 min read
General Intuition raised $320 million(約510億円) at a $2.3 billion(約3700億円) valuation to train AI agents using video game footage, betting that gameplay is a scalable way to teach machines how to act in the physical world.

Key takeaway

General Intuition, a startup using video game footage to train AI agents for the real world, secured $320 million(約510億円) in funding at a $2.3 billion(約3700億円) valuation. The company's key advantage is access to human action data embedded in gameplay clips from its parent company Medal, which it argues gives its agents a better understanding of causality and physical intuition than competitors. If successful, this approach could reduce the expensive real-world data collection that currently slows down robotics and embodied AI development.

Summaries like this, in your inbox every morning.

Sign up free →

3 Key Points

  • What happened

    General Intuition, a startup spun out of Medal (a video game clip platform), announced a $320 million(約510億円) funding round led by Khosla Ventures. The company's total disclosed funding now stands at $454 million(約730億円), after a $134 million(約210億円) round at launch last October. The startup is using hundreds of millions of hours of gameplay—paired with action labels showing exactly what buttons players pressed—to train AI agents (autonomous decision-making systems) that can understand space, time, and cause-and-effect.

  • Why it matters

    Most competitors try to infer actions from video alone, but General Intuition embeds the player's actual inputs into its training data, which the CEO argues gives the model a richer understanding of causality and the distinction between "self" and "environment." The same AI model that plays Fortnite can also control a physical quadrupedal robot with minimal real-world fine-tuning (eight minutes of data in this demo). If gameplay proves to be a scalable shortcut to building agents that work in the physical world, it could reduce the need for slow and expensive real-world data collection—a major bottleneck in robotics and embodied AI.

  • What to watch

    The vast majority of the funding will go toward scaling compute capacity through a deal with CoreWeave and pre-training the next version of the model. The company plans to make its API more broadly available by the end of summer. Currently, General Intuition has a handful of customers in gaming, simulation, and robotics, but Khosla described this as a "generational company" bet rather than an acquisition target, citing the proprietary action data from Medal as a key competitive moat.

FAQ

How is General Intuition's approach different from other AI agent companies?
Most competitors try to infer actions from video alone, but General Intuition uses action labels embedded in gameplay clips—records of exactly what buttons players pressed and when. The CEO argues this gives the model a richer understanding of causality and helps it distinguish between "self" and "environment" in a way that produces more intuitive behavior.
Where is the funding going?
The vast majority will go toward scaling compute capacity through a deal with CoreWeave and pre-training the next version of the model. A slice has been earmarked for making the API more broadly available by the end of summer.
What does the company plan to sell?
General Intuition ultimately wants to sell the agentic model itself. The world model it demonstrated (a simulated environment generated frame-by-frame) is used internally as a training environment, referred to as "the gym."

Discussion

No comments yet. Be the first to share your thoughts!

Log in to join the discussion

Related Articles

Stay ahead with AI news

Get curated AI news from 200+ sources delivered daily to your inbox. Free to use.

Get Started Free

Free · takes 30 seconds · unsubscribe anytime

5 minutes a day. The AI essentials.

200+ sources · Email / LINE / Slack

Get it free →