Prompt Engineering

The practice of composing natural language instructions to elicit knowledge from large language models in a structured way.

Unlike fine-tuning or retraining, prompt engineering operates solely on the embedded knowledge within the model’s parameters. It requires no parameter modification and no task-specific training data, only carefully constructed text that guides the model toward desired outputs.

This distinction matters: prompt engineering is accessible to anyone who can write clear instructions, while fine-tuning requires computational resources, labeled data, and ML expertise.

Related: 05-atom—prompting-vs-finetuning, 05-molecule—chain-of-thought-prompting, 05-atom—task-specificity-of-prompting