Startup: Robotics Nears Its ChatGPT Breakthrough

The landscape of artificial intelligence has dramatically shifted since OpenAI’s GPT-3 introduced the concept of foundation models. Previously, companies meticulously crafted specialized natural language processing (NLP) models from the ground up, training each on extensive, task-specific datasets. Today, the prevalent approach involves leveraging a general-purpose model, such as OpenAI’s GPT series, Claude, or Llama, and then adapting it through fine-tuning or prompting to address particular requirements.

Pim de Witte, CEO of General Intuition, anticipates a similar evolutionary trajectory for embodied AI. Instead of industries investing in the arduous collection of massive real-world datasets to develop highly specialized robot models, de Witte advocates for a strategic pivot towards generating higher-quality datasets. These datasets, he contends, should be capable of producing foundational models that can impart intuitive understanding of movement and interaction across a diverse array of environments.

“A lot of companies right now are doing lots of specialized work focused on individual embodiments, individual environments, and individual robots,” de Witte observed during a recent appearance on TechCrunch’s Equity podcast.

He further posits that much of this specialized effort will soon become obsolete with the advent of generalized models, akin to the innovative system General Intuition is currently developing and deploying.

“The generalization of the model itself is the product,” de Witte stated. He elaborated, “The fact that it has a base level of reasoning about space and time is going to be the reason why people stop collecting hundreds of thousands or millions of hours of real-world data. Because the reality is, you only need a few minutes.”

General Intuition successfully constructed its own foundation model by training it on millions of hours of video game data. This comprehensive dataset included crucial details such as human controller inputs and their precise timing. Both de Witte and General Intuition’s lead investor, Vinod Khosla, firmly believe that this "action data" is paramount for cultivating a human-like intuition for spatial-temporal reasoning.

This compelling thesis recently propelled the startup to secure $320 million in funding last month, valuing the company at $2.3 billion. General Intuition has impressively showcased its model's dual capabilities: not only can it engage in video gameplay for extended periods, but it can also effectively power a quadrupedal robot—a feat achieved after merely eight minutes of fine-tuning with real-world robotics data.

“The fact that [the robot] was actually able to zero-shot on just the front camera, with no other sensors, in the office with dynamic objects being introduced and people walking by was a very big surprise to us,” de Witte remarked, adding, “I think it’s a sign of what’s to come.”

General Intuition's ultimate objective is not to manufacture robots directly, but rather to establish itself as the foundational model for physical AI. This means providing a core intelligence that other robotics companies can build upon for their specific machines. As de Witte succinctly put it: “We’re not gonna build a self-driving car company. We’re gonna make it 10 times easier for the next person to build a self-driving car company.”