Glossary
Training (AI models)
The process of building a model by adjusting its parameters on large amounts of data.
Training is how an AI model learns. The model is shown large amounts of data, makes predictions, and has its parameters adjusted to reduce its errors, over and over. For large models this can take weeks or months on thousands of specialized chips.
Developers often separate pretraining, where a model learns general patterns from a broad dataset, from later stages (sometimes called post-training) such as fine-tuning and reinforcement learning from human feedback, which shape how the model behaves and responds.
Training is a one-time, large up-front cost for each model version. Running the finished model for users, called inference, is a separate and ongoing cost. Stories about AI's energy use or chip demand often mix the two.