A common misconception about AI is that the 'thinking' happens when you ask it a question. In reality, the heavy lifting happens months earlier during a phase called 'training.' To use a metaphor: if running an AI prompt (inference) is like driving a car from New York to Boston, then training the AI is like building the entire interstate highway system from scratch. Training an LLM requires massive data centers, tens of thousands of GPUs running for months, and megawatts of electricity to lay down the 'roads' (neural network weights). Once the highway is built, anyone can drive on it. Inference is comparatively cheap and fast. The model simply runs your prompt through its pre-calculated weights to generate a response. This comic illustrates the stark contrast between the two phases. Training is a massive, industrial-scale infrastructure project, while inference is a lightweight, everyday commute. Understanding this difference is crucial for grasping the economics and environmental impact of AI.
