Prompt
Definition
A prompt is the input that a language model or other generative AI system processes to generate an output. It can consist of an instruction, context, examples and format requirements. Prompt engineering, by contrast, is the method of deliberately designing and improving prompts.
A prompt is the instruction to an AI system. Well-written prompts include context, role, desired format and examples. Recurring tasks can be documented in a prompt library.
How it works
- The model processes the prompt as text and generates a continuation or answer based on it.
- Examples in the prompt can steer the model towards a task without new training; Brown et al. describe this as few-shot learning.
- Besides the user input, there is often a system prompt with overarching instructions.
Why it matters in practice
Clear prompts make results more traceable and comparable.
Even a good prompt does not guarantee a correct answer; results still need to be checked.
Example from business
Fictional example: A team stores a prompt for meeting minute summaries with goal, audience, desired length and a sample paragraph in its prompt library.
Limits and pitfalls
- Carelessly entering confidential data into prompts.
- Mixing several tasks in one unclear prompt.
- Expecting a prompt to work the same with every model.
Frequently asked questions
How does it differ from prompt engineering?
The prompt is the individual input; prompt engineering is the method of systematically designing and testing prompts.
Does a prompt have to be long?
No. What matters is that the task, context and desired result are clear.
Sources
- Language Models are Few-Shot Learners (2020) · Brown, T. et al.
- Prompt engineering overview · Anthropic
Editorially reviewed on: