See the Difference
Same task, same number of examples. One uses simple examples, the other uses complex ones with more reasoning steps.
"John has 5 apples. He eats 2. How many left?"
Answer: 5 - 2 = 3
AI: 30 - 12 = 18. Answer: 18 muffins.
Only did one step. Skipped the return and the second sale entirely.
"John has 5 apples. He buys 3 more. Gives 2 to Mary. Mary returns 1. He eats 2. How many left?"
Start: 5. Buy: 5+3=8. Give: 8-2=6. Return: 6+1=7. Eat: 7-2=5. Answer: 5
AI: Start: 30 muffins. Morning sales: 30 - 12 = 18. Return: 18 + 3 = 21. Afternoon sales: 21 - 8 = 13. Answer: 13 muffins.
Tracked every step because the example showed what thorough reasoning looks like.
Why This Works
AI mirrors the depth of the examples you show it. A two-step example teaches AI that two-step answers are fine. A five-step example teaches AI that tracking every detail matters.
The complexity of your examples sets a ceiling on the model's reasoning. Simple examples work for simple problems but fail on hard ones. Complex examples work on both — they teach thoroughness that generalizes, not shortcuts that break.
How to Apply This
- When choosing examples to include in your prompt, pick the ones with the most steps and detail
- A single thorough example outperforms several shallow ones
- Show every intermediate step in your example — don't skip any
- If the model is taking shortcuts, your examples are probably too simple
The Technique
When providing examples for AI to learn from, choose the most detailed, multi-step examples available. AI matches the depth of what it's shown. Rich examples produce thorough reasoning.
When to Use This
- • When doing math or logic problems with multiple steps
- • When AI keeps skipping steps or jumping to conclusions
- • When you have examples of varying complexity to choose from
- • When thoroughness matters more than speed
- • When you're already using examples (few-shot prompting) and want better results
Why Simple Examples Fail
- • They teach shortcuts — a 2-step example says "2 steps is enough"
- • They don't generalize up — simple patterns fail on hard problems
- • They hide the process — skipping intermediate steps means AI skips them too
Think of it like training a new employee. If you only show them easy tasks, they won't know how to handle complex ones. But if you show them complex tasks done well, they can figure out the easy ones on their own.
How It Relates
This technique extends Show by Example. While "Show by Example" says to provide examples, Complexity-Based Prompting tells you which examples to pick — the most detailed ones. It also pairs naturally with Think Step by Step, since your complex examples will naturally include step-by-step reasoning.