Loading the player…

Before OpenAI’s GPT-3 introduced the era of foundation models, businesses used to create specialized natural language processing models from scratch, training each one on significant volumes of task-specific data. Nowadays, most organizations start with a versatile model like OpenAI’s GPT series, Claude, or Llama, and then refine or guide it to meet their unique requirements. Pim de Witte, CEO of General Intuition, believes that embodied AI will follow a similar trajectory. Rather than gathering massive real-world datasets to construct specialized robot models, he suggests the industry should focus on higher-quality datasets that can produce foundation models capable of transferring an intuitive grasp of movement and interaction across diverse environments. He contends that much of this work will soon become unnecessary with the emergence of general models like the one General Intuition has been developing and deploying. “The generalization of the model itself is the product,” he stated. “The fact that it has a basic understanding of space and time will be the reason people stop collecting hundreds of thousands or millions of hours of real-world data. Because in reality, you only need a few minutes.”

General Intuition built its own foundation model after training on millions of hours of video game data, which included details like which buttons on a controller a human pressed and when. Both de Witte and General Intuition’s lead investor, Vinod Khosla, argue that this action data is essential for developing a human-like intuition for spatial-temporal reasoning. The startup recently raised $320 million at a $2.3 billion valuation, driven by this thesis. The company has shown that its current model can both play a video game for hours and control a quadrupedal robot—the latter after fine-tuning it on just eight minutes of real-world robotics data. “The fact that [the robot] could operate zero-shot using only the front camera, with no other sensors, in an office environment with dynamic objects and people walking by was a huge surprise to us,” de Witte said. “I think it’s a sign of what’s to come.”

The ultimate goal for General Intuition isn’t to build robots itself, but to become the foundation model for physical AI—a base model that other robotics companies can build upon for their own machines. As de Witte put it: “We’re not going to start a self-driving car company. We’re going to make it ten times easier for the next person to start one.”