What is a prompt?

A prompt is the input provided to a Generative AI model, that is used to guide its output. This input serves as the interface through which users communicate their desires to the model.

While prompts are often textual, they can also take other forms such as images, audio, videos, or a combination thereof.

In the context of Large Language Models (LLMs), prompts can vary from simple queries to complex problem statements. Prompts generally consist of instructions, questions, input data, and examples. To obtain a result from an AI model, a prompt must contain either instructions or questions, while other elements are optional.

The idea of using natural language prefixes, or prompts, originated before the GPT-3 and ChatGPT era. Earlier definitions sometimes considered only the input data (e.g., "llama") as the prompt, separating it from the "task description" (e.g., "Translate English to French:"), but more recent papers, including the sources provided, refer to the entire string passed to the LLM as the prompt.

The action of sending a chunk of text or uploading an image constitutes prompting. Prompts are often constructed via a prompt template, which is a function containing variables that are replaced by media (usually text) to create a prompt instance.