Generative Pre-Trained Transformer (GPT)

Generative Pre-Trained Transformer, abbreviated GPT, refers to a transformer-based language-model approach trained first on broad text or multimodal data and then adapted for useful interaction. Pretraining teaches statistical patterns, concepts, and relationships by predicting tokens; instruction tuning and preference optimization can subsequently make the model more helpful and controllable. GPT systems can draft, summarize, translate, answer questions, write code, and support tool-using applications. The acronym is associated with OpenAI model families, but it also describes the core design idea. A GPT output is generated from learned probabilities and context, so fluency does not guarantee factual accuracy, current knowledge, or sound judgment.