GPT

GPT (Generative Pre-trained Transformer) is a family (and by now a whole class) of large language models built on the “decoder” half of the transformer architecture. The name spells out the recipe: generative (it produces text), pre-trained (it first learns broadly from a huge pile of text before any task-specific tuning), and transformer (the attention-based network underneath). At its core it’s a next-word predictor: given the text so far, it guesses the most likely next chunk of text (a token), adds it, and repeats. At enough scale, that simple loop yields fluent writing, reasoning, coding, and more. OpenAI’s GPT-2/3/4 and successors popularized the term, but “GPT” now gets used loosely for decoder-only LLMs in general. One unrelated name collision worth knowing: GPT is also GUID Partition Table, a disk-partitioning scheme. Context tells them apart.