Before OpenAI’s GPT-3 ushered in the era of foundational models, companies built specialized natural language processing models from scratch and trained each one on large amounts of task-specific data. Today, most organizations start with a generic model such as OpenAI’s GPT series, Claude, or Llama, and then tweak or prompt it to solve their specific needs.
Pim de Witte, CEO of General Intuition, believes that embodied AI will follow a similar pattern. He argues that rather than collecting huge real-world datasets to build specialized robot models, the industry should focus on higher-quality datasets that can generate foundational models that can convey intuition about movement and interaction across many environments.
“A lot of companies are doing a lot of specialized work right now that focuses on individual embodiments, individual environments, individual robots,” de Witte told TechCrunch on a recent episode of Equity.
Much of that work will soon become unnecessary, he argues, with the advent of universal models like the one General Intuition is developing and deploying.
“The product is a generalization of the model itself,” he said. “The fact that we have a basic 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 in reality, all you need is a few minutes.”
General Intuition built its own such foundational model after training it on millions of hours of video game data, including information such as when humans pressed which buttons on a controller. Both De Witte and General Intuition’s lead investor Vinod Khosla argue that behavioral data is the key to developing human-like intuition in space-time reasoning.
The startup raised $320 million last month based on its thesis at a $2.3 billion valuation. The company has demonstrated that its current model can play video games for hours and can power a quadrupedal robot. The latter after fine-tuning based on just 8 minutes of real-world robot data.
“In an office where dynamic objects are introduced and people walk by, the fact that[the robot]was actually able to perform a zero shot with just the front camera without any other sensors was a huge surprise to us,” de Witte says. “I think this is a sign of things to come.”
General Intuition’s ultimate goal is not to build the robot itself, but to become a foundational model for physical AI, a base model for other robotics companies to build for their machines. Or, as De Witte puts it, “We’re not going to build a self-driving car company. We’re going to make it 10 times easier for the next person to build a self-driving car company.”
If you buy through links in our articles, we may earn a small commission. This does not affect editorial independence.
