General Intuition in talks to raise $300M at around $2B valuation
Summary
General Intuition, a New York-based startup, is in talks to secure approximately \$300 million in funding, valuing the company at over \$2 billion. This raise follows an earlier \$134 million seed round eight months prior, after spinning out from Medal. The company, backed by investors including Jeff Bezos, Eric Schmidt, Khosla Ventures, and General Catalyst, develops a foundation model that trains AI agents in spatial-temporal reasoning. It leverages Medal's unique dataset of 2 billion videos annually from 10 million monthly active users, which provides interactive, first-person gameplay data ideal for teaching machines to perceive, anticipate, and interact in real-time simulations. This dataset has reportedly drawn interest from major AI labs like OpenAI. While the world model space is competitive with players like Runway and Google's Genie 3, General Intuition focuses on training agents as its core product. The new capital will be used to expand compute capacity for a product launch by late summer or early fall.
Key takeaway
For Directors of AI/ML evaluating embodied AI agent development, General Intuition's strategy highlights the value of unique, interactive first-person datasets. Your teams should explore specialized data sources, like gaming footage, to build robust spatial-temporal reasoning capabilities in agents. This approach suggests focusing on agent performance as the core product, rather than selling the underlying world models, potentially offering a distinct market advantage. Consider how proprietary data can differentiate your AI initiatives.
Key insights
General Intuition leverages unique interactive video game data to train embodied AI agents for spatial-temporal reasoning.
Principles
- Interactive first-person data enhances spatial-temporal AI.
- World models can serve as agent training platforms.
- Proprietary datasets offer competitive advantage.
Method
The startup trains embodied AI and world models using a dataset of 2 billion interactive, first-person video game clips annually to teach deep spatial-temporal reasoning.
In practice
- Develop AI agents for real-time simulation.
- Utilize gaming data for embodied AI training.
- Focus on agent training, not model sales.
Topics
- AI Agents
- World Models
- Spatial-Temporal Reasoning
- Embodied AI
- Video Game Data
- Startup Funding
Best for: Investor, Entrepreneur, Director of AI/ML
Related on AIssential
Editorial summary, takeaway, and curation by AIssential. Original article published by AI News & Artificial Intelligence | TechCrunch.