If you ask an AI model, "I need to get my car cleaned, and the car wash is 100 metres away, should I walk or drive?", several popular LLMs tell you to walk. They give you a well-written breakdown on saving petrol, squeezing in a quick bit of exercise, and avoiding the hassle of starting an engine for a short trip.
It sounds sensible, until you remember the physical reality: walking 100 metres leaves your dirty car sitting in the driveway.
The model isn't reasoning. It catches keywords like "100 metres", "walk", and "drive", and immediately routes down a path of health and environmental advice. It completely misses the actual goal because it lacks a basic mental model of how the physical world fits together.
It sounds sensible, until you remember the physical reality: walking 100 metres leaves your dirty car sitting in the driveway.
The model isn't reasoning. It catches keywords like "100 metres", "walk", and "drive", and immediately routes down a path of health and environmental advice. It completely misses the actual goal because it lacks a basic mental model of how the physical world fits together.
