What does GPT stand for?

What does GPT stand for?

GPT stands for Generative Pre-trained Transformer. Each part of the name describes a different aspect of the model: what it does, how it is initially trained, and the architecture used to process information. The acronym alone does not specify a model’s size, accuracy, available tools, or application features.

Generative means producing a continuation

A generative language model can produce text conditioned on an input sequence. For example, a prompt beginning “Write a short invitation to a study group” can lead to newly generated invitation text. The model typically chooses successive tokens rather than selecting a complete answer from a fixed list.

Generation can produce useful explanations, summaries, and code, but it can also produce incorrect content. Fluency is an output characteristic, not evidence that every statement has been verified.

Pre-trained means learning before the final task

Pre-training gives a model broad learned patterns from an initial training objective. For a GPT-style language model, predicting subsequent tokens is a central example. Later training can adapt behavior toward following instructions or performing specific tasks.

The term does not mean that a model knows every fact or automatically learns permanently from each conversation. Ordinary inference uses the parameters available for that request.

Transformer names the architecture

GPT-style models use transformer processing with causal attention, allowing each position to use preceding context without accessing future positions. Learned transformations build representations used for subsequent-token predictions.

An assistant application may add conversation formatting, retrieval, tools, and other capabilities around a model. Those application features should not be inferred merely from the letters GPT. Understanding the acronym provides a starting point for examining the actual model and system.