Definition:
ChatGPT is an artificial intelligence assistant developed by OpenAI that allows users to converse in natural language to find information, create content and carry out tasks. It combines AI models with tools for working with text, files, images and other sources of information.
It is an application, not the name of a single model. Its capabilities depend on the models, tools and features available in the service being used.
Origins of ChatGPT
OpenAI introduced ChatGPT in 2022 as a conversational system that allowed people to interact with its models through written questions and instructions.
The application later expanded its capabilities with tools for working with files, searching for information and creating content in different formats. Describing ChatGPT solely as a chatbot based on GPT-3 therefore no longer reflects how it works.
How ChatGPT works
ChatGPT generates responses based on the instructions it receives and the available context. This may include the conversation, files provided by the user and information obtained through tools.
Instructions, known as prompts, explain the task and guide the result. For example, a user can request a brief explanation for beginners or a technical document with a specific structure.
Models are prepared through training and refinement processes that may incorporate human evaluations and feedback. This preparation is distinct from the conversation: responding to a message does not mean that the model is being retrained on it.
When a task requires searching for information, analysing a file or running code, ChatGPT can use the available tools and produce a response based on their results.
What ChatGPT is used for
ChatGPT can support tasks involving the creation, understanding and organisation of information. Its applications include the following:
- Writing and editing: preparing drafts, correcting text, summarising or adapting tone.
- Working with documents: locating information, comparing content and extracting data.
- Learning: explaining concepts, developing examples and preparing exercises.
- Programming: explaining code, proposing solutions and helping investigate errors.
- Working with images: interpreting visual materials and generating or editing images when the feature is available.
- Research: searching the web for information and comparing sources using enabled tools.
ChatGPT’s features also allow users to tackle multi-step tasks and prepare files they can review. Not all accounts and environments offer the same options.
Context and personalisation
Instructions and supplied materials allow the work to be adapted to a specific need. For example, a user can specify the audience for a text, provide a style guide or bring together files related to a project.
When memory or personalisation features are available, they may retain useful information for other conversations. Saving a preference is not the same as changing the model’s parameters.
It is also important to distinguish information provided in the chat from information accessible through connected services. These connections depend on the permissions granted and the tools enabled.
Limitations to be aware of
ChatGPT can produce incorrect responses, misinterpret a document or cite sources that do not support its claims. A fluent response does not demonstrate that its content is accurate.
Web search is not necessarily used in every response either. For tasks that require up-to-date information, sources and their dates need to be checked.
Before providing confidential data, users should review the terms and controls of the service being used. How information is retained and handled may vary according to the account and its settings.
Difference between ChatGPT, GPT and the OpenAI API
ChatGPT is the application with which the user interacts. GPT identifies a family of OpenAI models; the initials stand for Generative Pre-trained Transformer. ChatGPT can use different models, so it does not correspond to a fixed version of GPT.
The OpenAI API allows developers to incorporate models and other capabilities into their own applications. In this case, the software determines what information it sends and how it uses the response.
Using ChatGPT and developing an application through the API are different ways of accessing AI capabilities, with their own access and usage terms.