What is ChatGPT?
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.
ChatGPT is OpenAI’s assistant application. It combines AI models with a user-facing environment for conversation and other available forms of work. A model is the trained component that interprets supplied information and produces responses. ChatGPT is therefore not one fixed GPT model, even though the names are closely related.
The application can organize messages, files and connected capabilities around the model. A tool is a software function that obtains information or performs an operation, such as searching a source or working on a file. Which tools and models are available depends on the account, product configuration and current offering.
Imagine asking for a summary of an uploaded report. The visible experience is a conversation, but several questions matter underneath it. Did the application obtain the report’s contents? Which parts reached the model? Were any external sources consulted? A fluent summary does not answer these questions on its own.
ChatGPT can also support work involving actions rather than only text. The distinction between preparing a proposed change and carrying it out remains important. A draft email is not a sent email, and a claim that a document was updated should be supported by the actual document. The application’s tools and permissions govern those actions.
Memory and model training are separate concepts. Memory in an application means stored information that may be supplied in later interactions. Training changes the model’s learned internal numerical settings. An assistant appearing to remember a preference does not imply that the underlying model has been retrained on that conversation.
Using ChatGPT for discussion or project work is different from calling a model directly through an API, an interface for software requests and responses. The application supplies additional capabilities and working arrangements around the model.
For a healthcare research discussion, a useful exercise is to supply a public study report and ask for its population, intervention, comparison, outcomes and limitations, with supporting passages. Then compare the answer with the report rather than treating a fluent response as a completed appraisal. This uses the application to work with specified evidence, not as an authority on an unseen paper. A public paper or fictional clinical record is also a clearer teaching starting point than assuming a familiar account is appropriate for patient information.
Source: OpenAI — ChatGPT Overview.
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