Prompting
Clear instructions, delimiters, few-shot examples, step-by-step reasoning, chaining and structured output.
6 notes · 18 min total
- 1Clear instructionsWhy most bad outputs are bad prompts, and how to state the task, audience, format and constraints.3 min
- 2DelimitersHow to wrap outside content in tags so the model can tell your instructions from the data, and what that does and doesn't protect.3 min
- 3Few-Shot ExamplesHow to show the model two or three input-output pairs so it copies your format, and how to avoid the ways examples mislead it.3 min
- 4Step by StepWhy asking a model to reason before it answers improves hard tasks, how to ask, and what it costs.3 min
- 5ChainingHow to split a complex task into several prompts, check each result, and know when one prompt is enough.3 min
- 6Structured OutputThree levels of getting machine-readable output from a model, and why you still validate every response in code.3 min