5.2.1. Prompt Engineering Principles
💡 First Principle: The principles are the invariants behind every good prompt: supply clear intent and relevant context, reduce ambiguity, decompose complexity, and iterate. Any specific tactic is just one of these principles applied.
Where best practices are concrete moves, principles are the reasons they work:
| Principle | Why it works |
|---|---|
| Clarity of intent | The model completes what's most likely; clear intent makes the right completion most likely |
| Relevant context | The model only sees the constructed prompt; relevant context is the raw material it reasons over |
| Specificity | Specifics collapse the space of "plausible" answers toward the correct one |
| Decomposition | Smaller tasks fit the context window and are easier to get right and to validate |
| Iteration | First output is a draft; refining the prompt is faster than fighting a bad result |
Understanding the principles is what lets you handle an unfamiliar exam scenario: if a prompt is failing, you can diagnose which principle is being violated (ambiguous? missing context? too big?) and prescribe the fix, rather than reaching for a memorized trick.
💡 Key Point: Tactics tell you what to do; principles tell you why, which is what you need when the scenario is new.
⚠️ Exam Trap: Principles aren't interchangeable with tactics. A question asking why a technique improves results wants the principle (e.g., "it reduces ambiguity"), not just a restatement of the technique.
Reflection Question: Pick a prompt best practice and trace it back to the underlying principle it expresses — why is naming the principle more useful for an unfamiliar problem?
