Where randomness meets reason

The Prompt Lab — Chain-of-Thought Prompting

Learn the chain-of-thought prompting prompting technique with concrete before/after examples.

prompt-engineeringtechniquestutorial
One technique, one before/after. Get better at talking to models.

The Prompt Lab #1

Chain-of-Thought Prompting

The Technique

Chain-of-thought prompting involves explicitly asking the model to show its reasoning process step-by-step before arriving at a final answer. This technique leverages the model’s ability to break down complex problems into logical sequences, dramatically improving accuracy on multi-step reasoning tasks.

The Naive Prompt

You're helping a startup calculate their Series A dilution. The company has 10 million shares outstanding. Founders own 7 million shares, employees own 2 million shares through an option pool, and angels own 1 million shares. The Series A will raise $5 million at a $15 million pre-money valuation. What percentage will founders own after the Series A?

Why It Falls Short

The model jumps straight to a percentage without showing calculations, making it impossible to verify the logic or catch errors. Complex financial calculations involving multiple variables are particularly prone to mistakes when the reasoning steps aren’t explicit, and GPT-5.4 can still make arithmetic errors in multi-step problems.

The Improved Prompt

You're helping a startup calculate their Series A dilution. The company has 10 million shares outstanding. Founders own 7 million shares, employees own 2 million shares through an option pool, and angels own 1 million shares. The Series A will raise $5 million at a $15 million pre-money valuation. 

Please work through this step-by-step:
1. First, calculate the post-money valuation
2. Determine how many new shares will be issued
3. Calculate the total shares outstanding after the Series A
4. Finally, determine what percentage founders will own

Show your work for each step.

Why It Works

By explicitly requesting each calculation step, the model is forced to work methodically through the problem, reducing computational errors and making the logic transparent. The structured breakdown also makes it easy to spot mistakes and ensures the model doesn’t skip crucial steps like calculating post-money valuation before determining share issuance.

When to Use This

  • Multi-step calculations involving finance, engineering, or scientific problems where accuracy is critical
  • Complex reasoning tasks where you need to verify the logic, not just the final answer
  • Educational contexts where showing the process is as important as getting the right result