What is GPT?
The Harvard Chan AI Glossary
In our classes, we often pause to explain the ideas behind AI and the language used to describe it. These posts grew out of those conversations. They’re for anyone who would like to understand AI better, with no technical background assumed.
GPT stands for generative pre-trained transformer. “Generative” means producing content; “pre-trained” refers to learning from data before being adapted or used for a particular task; a transformer is a design for combining information through attention calculations. OpenAI uses GPT in the names of its model families, including generations such as GPT-6. A model is the trained component that interprets information and produces responses. GPT is not a synonym for ChatGPT: ChatGPT is a product that combines models with a conversational interface and other available capabilities.
A model-family name and a particular model identifier are different levels of description. A family can contain several versions or variants intended for different combinations of capability, speed and cost. For example, GPT-6 names a generation, while GPT-6 Astra identifies a model within that generation. These examples explain the naming; they are not a permanent statement about the newest model.
A language model works with patterns learned from examples and the information supplied for the current task. That current information is context. It may contain instructions, text and results from tools. A tool is a software function such as searching documents or running a calculation. Whether the model can use such a function depends on the application and its permissions.
Developers may access GPT models through an API, an agreed method for software to send requests and receive responses. A person using ChatGPT interacts with a more complete application. Codex names an agent product for coding and related project work. The model, assistant application and coding environment should not be treated as the same thing.
For an illustrative task, asking a model to explain a program is different from allowing a coding agent to edit and run it. The first produces text; the second requires file access and execution tools. A strong model can contribute to both, but the surrounding software determines what actually happens.
The useful questions are which model is being used, what information it receives and what actions the application can carry out. A generation number alone does not answer whether a particular task is handled reliably. Version-specific documentation and task-specific evaluation are more informative than assuming that a newer or more expensive label settles the choice.
Sources: OpenAI — Key Concepts; OpenAI — Models and GPT-6 Model Guide.
If you would like to find out more about AI and improve your leadership skills, consider the AI in Health Care Certificate of Specialization from the Advanced Learning Academy at Harvard T.H. Chan School of Public Health. You can take an individual course or work toward the full certificate.
AI is a fast-moving field. This explanation reflects our class discussions and related material at the time of writing. Terminology, capabilities and tools may change.
For suggested additions or corrections, please email me at tpanch@hsph.harvard.edu.
Dr. Trishan Panch
Program Director, AI in Health Care
Harvard T.H. Chan School of Public Health — Advanced Learning Academy