GPT

/dʒiː piː tiː/ · Abbreviation · AI & Machine Learning · Origin: 2018

Definitions

Generative Pre-trained Transformer — OpenAI's series of large language models. GPT-3 (2020) demonstrated emergent abilities; GPT-3.5 powered the original ChatGPT; GPT-4 (2023) showed multimodal capabilities. The series that made 'AI' a dinner-table conversation topic.
In plain English: The AI model series by OpenAI that powers ChatGPT — trained on vast amounts of text, it can write, code, analyze, and reason about almost any topic.
The 'T' in GPT — Transformer — is the architectural innovation. The 'P' — Pre-trained — reflects the two-phase approach: expensive pre-training on internet-scale data, followed by cheaper fine-tuning for specific tasks. This approach made LLMs economically viable.
Example: 'GPT is just a brand name. The real innovation is the transformer architecture — Google invented it, OpenAI scaled it, and now everyone uses it.'
Source: acronym breakdown / architecture

2018: OpenAI publishes 'Improving Language Understanding by Generative Pre-Training,' introducing the GPT architecture with 117M parameters
2020: GPT-3's 175B parameters demonstrate emergent abilities — few-shot learning, code generation — that surprise even its creators
2022: ChatGPT (based on GPT-3.5) becomes the fastest-growing consumer application in history, making GPT a household acronym