GPT (Generative Pre-trained Transformer)
GPT stands for Generative Pre-trained Transformer, a type of large language model built on the transformer neural network architecture. It is pre-trained on a broad range of text to predict the next word, which lets it generate coherent, contextually relevant responses in natural language. GPT is the model family behind ChatGPT.
Each part of the name describes the model. Generative means it produces new text. Pre-trained means it first learns general language patterns from a large dataset before being adapted to specific tasks. Transformer is the neural network architecture, which uses attention to relate every word in the input to every other word.
After pre-training, GPT models are typically fine-tuned and aligned with human feedback so they follow instructions. A business might use a GPT-based assistant to summarize documents or answer questions, often grounded in its own data with retrieval-augmented generation.
GPT is a specific family of models, while large language model is the general category. Other LLMs use similar transformer designs under different names. ChatGPT is a chat application built on GPT models, not the model itself.