Definition: AI training vs. inference

In contrast to traditional programming, after the AI architecture (neural network, GPT, etc.) is designed and set up, the next phase is the training (learning) phase which analyzes the patterns in the data samples. It can take days, weeks and months. For example, GPT-4 took several months and was trained on trillions of words.

The AI that does the work (generates content, makes predictions, etc.) takes considerable computer time but nowhere near as much as the learning phase. This processing phase is called "inference," which dates back to the early expert systems in artificial intelligence. See AI datacenter.

   Non-AI
   Development     AI Development

   1. Design       1. Design
   2. Programming  2. Programming
                   3. Training and adjusting

   Non-AI
   Processing      AI Processing

   Processing      Inference