Self-Consistency 🟡
Need: You need higher confidence in the answer to a complex problem.
Explanation: Generate multiple answers (via CoT) and pick the most common/best one — “voting.”
Prompt format:
“Solve the problem in 3 ways, compare the results, and choose the best one.”
QA usage example
“Propose 3 test coverage strategies for the payments module. Compare them and choose the most effective one.”
Meta Prompting 🟡
Need: You need AI to write a better prompt than you would write yourself.
Explanation: The model generates or optimizes a prompt — a prompt that creates a prompt.
Prompt format:
“Write an optimal prompt for [task] that will produce the best results with [model].”
QA usage example
“Write an optimal prompt for generating API test scenarios that accounts for edge cases, Gherkin format, and priorities.”
Least-to-Most 🟡
Need: The problem is too complex for a single prompt — it needs to be broken into smaller pieces.
Explanation: You split the problem into subproblems, solve from the simplest, and build on the results.
Prompt format:
“Break the problem into smaller parts. Solve the simplest one. Use the result to solve the next one.”
QA usage example
“How to test the entire payment flow? 1) First: form validation. 2) Then: integration with the payment gateway. 3) Finally: error handling and retry.”
PAL 🔴
Need: You need precise calculations or logic — the model assists itself with code.
Explanation: Program-Aided Language — the model generates code (Python, JS) as a reasoning tool instead of computing on its own.
Prompt format:
“Solve problem [X] using Python code. Show the code and the result.”
QA usage example
“Calculate test coverage: 47 scenarios across 12 modules, but 3 modules have 2 configuration variants each. Write Python that computes the minimum number of tests.”
In the next post: RTF, CARE, CREATE, and Other Starters — simple frameworks for a quick start.