This Startup Thinks Robotics Is About to Have Its ChatGPT Moment

The rise of large language models has completely transformed how software is built, and now one startup believes robotics is standing on the edge of the same revolution. Just as ChatGPT and similar tools replaced the need to build custom AI systems from scratch for every single task, General Intuition is betting that general‑purpose foundation models will soon become the standard for intelligent machines, drastically reducing the time, cost, and effort required to make robots work reliably in the real world.

The Shift That Changed AI

Before OpenAI introduced GPT‑3 and the era of foundation models, artificial intelligence was built on narrow, task‑specific systems. If a company wanted software to understand text, translate languages, or summarize documents, it had to gather massive amounts of specialized data, design a custom model, and train it from the ground up, a slow, expensive, and inflexible process. Today, almost every developer starts with a powerful, general‑purpose model like GPT, Claude, or Llama, and then fine‑tunes or guides it to fit their specific needs.

Pim de Witte, CEO and co‑founder of General Intuition, argues that the field of robotics is stuck in that same earlier phase and that change is coming fast.

“Right now, most companies are doing lots of specialized work focused on individual robots, specific environments, and isolated tasks,” de Witte explained during a recent appearance on TechCrunch’s Equity podcast. “They’re collecting millions of hours of footage and sensor data just to make one type of machine work in one setting. But much of that effort will soon become redundant.”

A New Approach: General Intuition

The core idea behind General Intuition is simple but transformative: instead of building models that only work for one machine or one job, create a foundation model that understands the fundamental rules of how the physical world works, concepts like space, movement, cause and effect, and how objects interact. Once a system has that basic intuition, it can be adapted to new tasks and environments with only a tiny amount of additional training.

“The generalization of the model itself is the product,” de Witte said. “The fact that it already has a base level of reasoning about space and time means companies won’t need to collect hundreds of thousands or millions of hours of real‑world data anymore. The reality is, you’ll only need a few minutes.”

What makes this possible, according to the company, is an unconventional source of training data: millions of hours of video game footage. General Intuition’s model was trained not just on what was happening on screen, but also on the exact inputs players used, which buttons were pressed, when, and in what sequence. That combination of visual information and corresponding human action teaches the system how to navigate, react, and make decisions in simulated 3D environments.

Lead investor Vinod Khosla agrees that this action‑based data is the missing piece in robotics development. “To build machines that move and think as we do, you first need to teach them how actions connect to outcomes,” Khosla said. “Video games provide a safe, controlled, and unlimited environment to learn those lessons without the risks or costs of real‑world testing.”

Proof of Concept and Funding

The startup’s approach has already attracted major backing. Last month, General Intuition raised $320 million in new funding**, pushing its valuation to **$2.3 billion. That investment is a vote of confidence in its thesis, and it comes after the company demonstrated results that would have seemed impossible just a few years ago.

In one demonstration, the same model that had spent thousands of hours learning inside video games was transferred to control a physical quadrupedal robot. Instead of needing months or years of real‑world training, the system was fine‑tuned using just eight minutes of actual robotics data. Even more remarkably, it operated using only the feed from a single front‑facing camera, with no additional sensors or pre‑programmed maps.

“The fact that it could navigate an office space, avoid obstacles, adjust to people walking past, and react to moving objects all after so little real‑world input was a huge surprise even to us,” de Witte said. “It’s exactly the kind of transferability we’ve been working toward, and it’s a clear sign of what’s coming next.”

The End Goal: Powering the Industry

General Intuition is not looking to build its own line of robots, self‑driving vehicles, or industrial machines. Instead, it wants to become the underlying platform that powers the entire field of embodied AI, the equivalent of what GPT models are to software developers today.

“We’re not trying to become a self‑driving car company,” de Witte said plainly. “We’re trying to make it ten times easier for the next entrepreneur or team to build that company. By providing a ready‑made foundation of physical intuition, we remove the biggest barrier to entry: the time and cost of teaching machines how to understand the world.”

If the vision holds, robotics could soon follow the same trajectory as generative AI: a shift from custom‑built, limited‑use tools to flexible, adaptable systems that can be quickly deployed across industries from logistics and manufacturing to healthcare, agriculture, and home assistance.