Jia Lissa
The quest for Artificial General Intelligence (AGI)—a machine capable of understanding, learning, and applying intelligence across any task a human can perform—has long been dominated by the logic of Large Language Models (LLMs). While models like ChatGPT, Claude, and Gemini have demonstrated a masterful grasp of syntax, sentiment, and semantic retrieval, they harbor a fundamental blind spot: they live in a world of tokens, not reality. They excel at predicting the next word in a sentence but struggle to grasp the physical laws of cause and effect in the three-dimensional world.
Enter General Intuition, a New York-based startup that is betting $2.3 billion that the path to true AGI isn’t buried in the vast archives of the internet’s text, but in the virtual physics of video games. With a massive $320 million funding round recently closed, backed by titans including Coatue, Eric Schmidt, and elite researchers from Google DeepMind and MIT, the company is positioning itself at the vanguard of "Physical AI."
The Core Thesis: Gaming as the Ultimate Simulator
The central limitation of current generative AI is its lack of "world models." An LLM can describe a falling cup with poetic accuracy, but it does not fundamentally understand the physics of gravity, momentum, or collision.
General Intuition argues that video games are the perfect training ground for artificial agents. Unlike static text, games provide dynamic, interactive environments that obey consistent—if sometimes stylized—laws of physics. By training AI agents within these virtual worlds, General Intuition aims to teach models how to navigate space, time, and physical interactions. The hope is that this "gaming intuition" will translate seamlessly into the real world, allowing robots and software agents to perform complex, unscripted tasks in physical environments.
Chronology: From Medal TV to a Multi-Billion Dollar Bet
The genesis of General Intuition is rooted in the gaming ecosystem. CEO Pim de Witte, a veteran of the gaming industry, previously founded Medal TV, a platform that gained massive popularity for allowing gamers to record and share their gameplay highlights.
- The Inception: During his time at Medal TV, de Witte observed the sheer volume of high-fidelity, interactive data generated by gamers. He recognized that this data wasn’t just entertainment; it was a map of human decision-making in complex, simulated environments.
- The Pivot: Recognizing that the next leap in AI would require move-and-act capabilities, de Witte spun out the core technology and research team into General Intuition.
- The Validation: The startup quickly caught the eye of Silicon Valley’s elite. The recent $320 million funding round served as a major validation point, signaling that the venture capital community is looking beyond the "text-only" era of AI.
- The Current State: With a $2.3 billion valuation, General Intuition has transitioned from a niche experimental lab into one of the most well-capitalized startups in the field of physical AI.
Supporting Data: Why "Spatial Intelligence" Matters
The shift toward video game data is supported by a growing body of academic research. Traditional machine learning relies on "Big Data," but the quality of that data is increasingly being questioned. LLMs have already consumed much of the high-quality text on the internet.
The industry is now pivoting toward "Synthetic Data"—information generated by simulations. Gaming engines like Unreal Engine and Unity provide:
- High-Frequency Feedback Loops: Unlike text, where a model waits for a prompt, games provide millions of frames of interaction per second, allowing agents to learn through "trial and error" at an accelerated pace.
- Multimodal Integration: Gaming data combines visual input (pixels), auditory input (sound cues), and logical input (game state), forcing the AI to synthesize multiple streams of information simultaneously.
- Generalization: An AI that can win in a first-person shooter or a complex strategy game is demonstrating a form of "strategic thinking" that is far more flexible than the pattern-matching seen in current LLMs.
Official Responses and Strategic Vision
In a recent appearance on TechCrunch’s Equity podcast, CEO Pim de Witte articulated the company’s vision. He argued that the current reliance on text is a "local maximum"—a peak that AI research has reached but cannot climb higher from without a fundamental shift in architecture.
"We are building world models that actually understand how the world moves," de Witte explained. When pressed on the technical challenges, he emphasized that the transition from game physics to real-world physics is the primary engineering hurdle. "The goal is to bridge the ‘sim-to-real’ gap," he noted, referring to the process of training models in simulations so they can operate robots or autonomous systems in factories, warehouses, or homes.
The Ethical Red Lines
With great power comes the inevitable scrutiny of how such technology is deployed. Because General Intuition’s models are designed to understand physical movement and tactical decision-making, they are inherently dual-use. They could power a warehouse robot—or, if integrated into hardware, a defense system.
During the Equity interview, the conversation turned toward the "red lines" of development. De Witte acknowledged that as the company grows, it must establish clear boundaries regarding defense applications. The startup is under pressure to ensure its "physical intelligence" is used for productivity and societal advancement rather than autonomous weaponry. This ethical framework will be a defining factor in whether the company maintains its current level of institutional support, particularly from figures like Eric Schmidt, who has been vocal about the ethics of AI in national security.
Implications for the Future of AGI
The emergence of General Intuition suggests a massive correction in the trajectory of the AI industry. If they succeed, the next generation of AI agents will not just be "chatbots," but "do-bots."
The Impact on Robotics
The "brain" of a robot has historically been its biggest bottleneck. Hardware has advanced significantly, but software often lacks the intuitive sense of space required to navigate a messy human environment. By leveraging gaming data, General Intuition hopes to provide the "motor cortex" for the next generation of humanoid robots.
The Impact on the Labor Market
If AI gains the ability to interact with the physical world, the scope of "automated tasks" expands from white-collar desk work to blue-collar physical labor. This transition will likely accelerate debates around the economic impact of AI, shifting the focus from "Will AI replace my office job?" to "Will AI replace my physical labor?"
The New Benchmark for Intelligence
We are moving away from the Turing Test—which focused on the ability to deceive a human through conversation—toward a "Physical Performance Test." If an AI can enter a kitchen it has never seen, identify the objects, and successfully prepare a meal without breaking a glass, it has achieved a level of intelligence that no current LLM can claim.
Conclusion: A High-Stakes Gamble
General Intuition’s $2.3 billion valuation is more than just a bet on a company; it is a bet on a paradigm shift. The founders and investors are banking on the idea that the internet’s text is a dead end for true intelligence, and that the future of AGI lies in the simulated, high-stakes environments of video games.
As the company continues to scale, it faces immense pressure to prove that its "gaming-trained" agents can survive the transition from the screen to the real world. If it works, we may be looking at the birth of the first truly intelligent, physically capable machines. If it fails, it will serve as a stark reminder that while simulations can teach us a great deal, they remain a simplified map of a much more complex, chaotic reality.
For now, the world watches as a startup with roots in gaming highlights the limitations of the current AI boom, daring the industry to look up from its keyboards and finally learn how to move.
Siti Muinah
Asep Darmawan
