Training, then use
During training, a model learns from examples, such as past transactions marked as fraudulent or legitimate. It adjusts its internal settings to get better at spotting fraud. This can take a lot of data and computing power.
During inference, the trained model handles new information. It might assess a new transaction or generate a reply to a question. Most interactions with an AI product are inference.
Training usually happens on powerful servers. Inference may happen on a company’s servers or on your device. Where it runs affects speed, cost, privacy, and whether it can work offline.
Changing the model or guiding a response
Fine-tuning continues training an existing model on selected examples. A prompt tells the model what to do during inference. Retrieval looks up relevant information for the model to use in its reply. Prompting and retrieval do not retrain the model.