Also known as: model training
In plain English
Training is the learning phase. The model looks at huge numbers of examples, makes guesses, finds out how wrong it was, and adjusts itself a tiny bit each time. After millions of rounds, it gets good at the task.
In practice
Training a large model from scratch costs millions of dollars and is done by a handful of labs. Most organisations instead take a pre-trained model and adapt it through prompting, retrieval (RAG) or fine-tuning, which is far cheaper.
Under the hood
Training optimises model parameters to minimise a loss function, typically with stochastic gradient descent variants such as Adam, over many passes through the data. Large language models go through pre-training on broad text, then post-training such as instruction tuning and reinforcement learning from human feedback.
Example
"The vendor says customer data is never used for training their models."