Advanced Techniques

Recall: Common Mistakes

In Lesson 6, you learned to diagnose and fix common prompting mistakes. Now let’s level up with techniques that unlock AI’s deeper capabilities.

By the end of this lesson, you’ll be able to:

Use chain-of-thought prompting for better reasoning
Stack personas for multi-faceted tasks
Apply structured thinking techniques
Know when to use (and not use) advanced methods

Technique 1: Chain-of-Thought (CoT)

What it is: Asking AI to show its reasoning step-by-step before giving an answer.

Why it works: When AI “thinks aloud,” it’s less likely to make errors. Each step constrains the next step, leading to more coherent conclusions.

Basic CoT Prompt

“Think through this step-by-step before giving your final answer:
[Your question or task here]”

Advanced CoT with Structure

“I need to make a decision about [topic]. Help me think through this systematically:
First, identify the key factors to consider
Then, analyze each factor
Consider potential counterarguments
Finally, give me your recommendation with confidence level
Show your reasoning for each step.”

When to Use CoT

Complex decisions with multiple factors
Math or logic problems
Analyzing tradeoffs
Tasks where you need to verify AI’s reasoning

When NOT to Use CoT

Simple factual questions (“What’s the capital of France?”)
Creative writing (can make it feel stiff)
When you need quick, short responses

Technique 2: Persona Stacking

What it is: Combining multiple expert roles in one prompt.

Why it works: Some tasks benefit from multiple perspectives. Instead of asking twice (once as developer, once as security expert), combine them.

Example: Single Persona

“You are a senior software developer. Review this code.”

Example: Stacked Personas

“You are a senior software developer AND a security specialist. Review this code for both functionality and security vulnerabilities. Consider it from both perspectives.”

Powerful Combinations

Task	Persona Stack
Business decision	MBA strategist + industry veteran + devil’s advocate
Content creation	Writer + SEO expert + brand strategist
Code review	Senior developer + security specialist + performance engineer
Product design	UX designer + accessibility expert + business analyst

Template

“You are a [expertise 1] AND a [expertise 2]. Approach this task from both perspectives:
[Your task here]
For each major point, indicate which lens you’re using.”

Technique 3: Structured Thinking Frameworks

AI can apply classic thinking frameworks when you ask it to. Here are the most useful:

The Premortem

“Imagine this project has failed spectacularly six months from now. What went wrong? Work backwards from the failure to identify the most likely causes and how to prevent them.”

Steelmanning

“I’m considering [position X]. Before I commit, steelman the opposing argument—give me the strongest possible case against my position. Don’t hold back.”

First Principles

“Break this down to first principles. What are the fundamental truths here? Build up the analysis from those foundations, questioning each assumption.”

The Five Whys

“For this problem [describe problem], apply the Five Whys technique—keep asking why until we reach the root cause.”

Quick check: Before moving on, can you recall the key concept we just covered? Try to explain it in your own words before continuing.

Technique 4: Calibrated Confidence

What it is: Asking AI to rate its own confidence and acknowledge uncertainty.

Why it works: AI often sounds equally confident whether it’s certain or guessing. Asking for calibration makes it more honest about limitations.

Prompt Addition

“After your response, rate your confidence in this answer:
HIGH (very confident, well-documented area)
MEDIUM (reasonable inference but some uncertainty)
LOW (educated guess, would recommend verification)
Also note any specific areas where you’re uncertain or where I should verify.”

Example Usage

“What are the key differences between Kubernetes and Docker Swarm for a 50-person team’s production environment?
Rate your confidence and note areas I should verify with current documentation.”

Technique 5: Output Constraints for Quality

Sometimes limiting AI’s options improves quality:

Forced Brevity

“Explain [concept] in exactly 3 sentences. No more, no less.”

This forces AI to prioritize the most important information.

Structured Constraints

“Give me exactly 5 reasons, no more, no less. Rank them from most to least important.”

Devil’s Advocate Constraint

“You must argue against [popular position]. Even if you disagree with the counterargument, present it as compellingly as possible.”

Instead of trying to get everything in one shot, design for iteration:

First Pass: Generate

“Give me a rough draft of [X]. Don’t worry about polish—focus on getting the key ideas down.”

Second Pass: Critique

“Now review that draft critically. What’s weak? What’s missing? What would a skeptical reader challenge?”

Third Pass: Refine

“Incorporate that feedback and produce a stronger version.”

This three-pass approach often produces better results than trying to get perfection immediately.

Advanced Technique Combinations

The real power comes from combining techniques:

Example: Complex Decision Making

“[Persona Stack] You are a CFO with 20 years experience AND a startup founder who’s seen companies fail.
[Chain of Thought] Think through this step-by-step:
[Steelman] First, give me the strongest argument FOR expanding to the new market.
[Premortem] Then, imagine we expanded and it failed—what went wrong?
[Calibrated Confidence] Finally, give me your recommendation with confidence level, noting any key assumptions I should test.
[Constraint] Keep total response under 500 words.”

When Advanced Techniques Backfire

Not every prompt needs advanced techniques. They can backfire when:

The task is simple: Chain-of-thought on “What’s 2+2?” is overkill
Speed matters more than depth: Multi-step prompts take longer
Creative freedom is needed: Too much structure kills creativity
You’re exploring: Sometimes you want AI’s unconstrained perspective first

Rule of thumb: Start simple. Add complexity only when simple prompts don’t work.

Key Takeaways

Chain-of-thought improves reasoning by forcing step-by-step thinking
Persona stacking brings multiple expert perspectives to one task
Structured frameworks (premortem, steelman, first principles) apply proven thinking methods
Calibrated confidence helps AI be honest about uncertainty
Combine techniques for complex tasks, but keep it simple when you can

Up Next

Lesson 8 is your capstone—a practical project where you’ll apply everything you’ve learned. You’ll work through a realistic scenario using all the techniques from this course.