noun · AI practice
Prompt engineering
also called prompt design
Getting better answers by asking better. It means writing clear instructions, adding examples, and testing what works.
The practice of designing and testing prompts, including instructions, examples and structure, to get reliable output from a model.
The practice of designing the instructions, examples, format and surrounding context of a prompt, then testing variations against a set of cases. Because the model, not the prompt, produces the output, results are measured rather than assumed.
Real-life example
You write a prompt that works on your three test messages and ship it. A week later, strange inputs break it. A better approach would have been to collect forty real messages, run each version of the prompt over all of them, and keep the version that scored best.
Where the analogy breaks
-
There are no magic phrases. Clear instructions, the right information and good examples matter more than clever wording.
-
A prompt that works once is not proven. It has to be tested on many inputs, because small changes can swing results.
-
Prompts can break when the model changes. A new model may read the same words differently, so keep your tests and run them again.