The training phase of an AI collects from thousands to trillions of data examples and requires considerably more computer processing time than running the AI to do work (make predictions, generate content, etc.)
In contrast to traditional programming, after the AI architecture (neural network, GPT, etc.) is designed and set up, the next phase is the learning, or training, phase, which is very time consuming. The actual AI processing, known as "inference" or "inferencing," takes less computer power than the training phase. See
AI datacenter.
Traditional AI
Development Development
1. Design 1. Design
2. Programming 2. Programming
3. Training and adjusting
Traditional AI
Processing Processing
1. Processing 1. Inference