Terminology · World model
World model
A foundation model whose learning target is space, material, light and how objects move — as opposed to text associations or image representations.
Definition
A world model takes the regularities of the physical world as its learning target: spatial relations, material and reflectance, changing illumination, how objects move, and what an action leads to. What separates it from a language model is not parameter count but the learning target — a language model learns associations between pieces of text, a visual foundation model learns transferable image representations, and a world model answers what happens in real space.
On a production line that difference becomes concrete: change a model number, a material batch or a lamp, and a model trained on one task's decision boundary has to be retrained, while a base that learned the underlying regularities can carry the same understanding across.
How it differs
| Category | Learning target | What it answers |
|---|---|---|
| Language foundation model | Associations between pieces of text | Whether it reads coherently |
| Visual foundation model | Transferable image representations | Whether it looks right |
| World model | Space, material, light, motion and causality | What happens in real space |
What DaoAI World carries on the factory floor
DaoAI World is the world-model base built in-house by DaoAI (微链道爱). Four product lines — ACI auto cognitive inspection, robot vision, SkyVision surveillance and Wemio content — all sit on the same base. The company's position is that AGI = one general world model plus N vertical small models, rather than a large language model with skills or agents bolted on.
Delivery is 100% on-premise: sample images and inspection data never leave the plant, inference runs at the line edge, and a network outage does not stop production. For manufacturing that is a precondition rather than an option — yield curves, defect libraries and process parameters are core secrets.
DaoAI World platform —— The base itself and the end-to-end chain
Visual foundation model —— The adjacent concept this page separates from
Embodied intelligence —— Where the world model lands on robots
ACI · auto cognitive inspection —— The inspection vertical