New AI Models Could Become the Intelligence Layer Behind Future Robots
Artificial intelligence transformed software with large language models such as ChatGPT, Claude, and Llama. Now, a growing number of researchers and startups believe the next major AI revolution will happen in the physical world.
The idea is simple:
Instead of building thousands of specialized robot systems trained for individual tasks, the robotics industry could develop general-purpose AI models capable of understanding movement, space, and interaction.
This approach could bring robotics its own version of the ChatGPT moment.
Startup General Intuition believes embodied AI will follow a similar path to modern language AI — where foundation models become the core intelligence layer powering many different applications.
From ChatGPT to Physical AI
Before large language models became popular, companies typically created specialized AI systems designed for specific tasks.
For example:
- One model for translation
- One model for speech recognition
- One model for search
Each system required large amounts of task-specific training data.
The arrival of foundation models changed this approach.
Instead of building everything from zero, developers can start with a powerful general AI model and adapt it for different purposes.
General Intuition believes robotics could experience the same transformation.
The Current Problem With Robot Training
Today, many robotics companies train robots individually.
A robot designed for:
- Warehouse picking
- Household tasks
- Manufacturing
- Autonomous navigation
often requires its own specialized datasets and training process.
This creates a major challenge.
The physical world contains almost unlimited variations:
- Different environments
- Different objects
- Different movements
- Different human behaviors
Collecting enough real-world robot data is expensive and time-consuming.
A Foundation Model for Robots
General Intuition’s approach is to create a general-purpose AI model that understands physical interactions.
Instead of teaching robots every possible task separately, the company aims to build an intelligence system that already understands concepts such as:
- Space
- Movement
- Objects
- Cause and effect
- Physical interaction
This could allow robots to learn new skills with much smaller amounts of additional training data.
The company believes that better-quality datasets could eventually replace the need for millions of hours of specialized robot demonstrations.
Training AI Through Virtual Worlds
One of the interesting aspects of General Intuition’s approach is using video game data to train its models.
Video games provide enormous amounts of structured information:
- Visual environments
- Player actions
- Movement decisions
- Object interactions
Unlike ordinary videos, game data includes information about what actions caused specific results.
For AI systems learning physical reasoning, this connection between action and outcome is extremely valuable.
Robots Learning Faster With Less Data
According to the company, its AI model has demonstrated the ability to control robotic systems after being fine-tuned with only a small amount of real-world robotics data.
This suggests a possible future where robots do not need years of individual training.
Instead, they could start with a shared intelligence model and quickly adapt to new environments.
This approach could dramatically accelerate robotics development.
Why Foundation Models Could Change Robotics
If successful, robotics foundation models could impact many industries.
Manufacturing
Robots could quickly adapt to changing production environments.
Logistics
Warehouse robots could learn new tasks without complete retraining.
Healthcare
Assistive robots could better understand human needs.
Homes
Domestic robots could perform a wider range of everyday activities.
Autonomous Vehicles
Physical AI models could improve how machines understand complex environments.
The Difference Between AI Software and Physical AI
However, robotics is much harder than software AI.
A chatbot can make mistakes in a conversation.
A physical robot must deal with:
- Gravity
- Objects
- Human safety
- Unexpected environments
- Mechanical limitations
Understanding the physical world requires more than language intelligence.
It requires a combination of:
- AI reasoning
- Computer vision
- Robotics hardware
- Sensor understanding
- Real-world experience
The Future: One Brain, Many Robots
General Intuition’s long-term goal is not to manufacture robots itself.
Instead, the company wants to become a foundation technology provider for the robotics industry.
The vision is similar to how operating systems or AI platforms support many applications.
A future scenario could look like this:
- One AI model provides intelligence
- Different companies build different robot bodies
- Robots adapt the same intelligence to different environments
This could accelerate the development of humanoid robots, industrial machines, and autonomous systems.
The Beginning of a New Robotics Era
The robotics industry may be approaching a turning point.
Just as foundation models transformed software AI, physical AI foundation models could become the foundation for the next generation of intelligent machines.
The biggest breakthrough may not come from creating a single amazing robot.
It may come from creating the intelligence that allows thousands of robots to learn, adapt, and improve.
The next ChatGPT moment may not happen on a screen.
It may happen in the physical world.
