Lesson 10New

Program of Thought

Code instead of text

The Problem: AI often makes arithmetic errors even with Chain of Thought. When calculations get complex, how can we ensure accuracy?

The Solution: Use a Calculator, Not Mental Math

Program of Thought (PoT) has the AI write code to solve computational problems instead of trying to calculate in its head. It's like using a calculator instead of doing mental math — the computer handles the numbers precisely. Where Chain-of-Thought reasons in text, PoT reasons in code for exact results.

Think of it like using a calculator vs mental math:

1. Read the problem: "Calculate compound interest over 5 years..."
2. Write code: Express the logic in Python/JavaScript
3. Execute: Run the code and get exact results
4. Explain: Describe what the code does and the answer

Where Is This Used?

Financial Calculations: Compound interest, loan payments, ROI
Data Analysis: Statistical calculations, aggregations
Scientific Computing: Physics, chemistry calculations
Date/Time Math: Days between dates, timezone conversions

Fun Fact: Program of Thought achieves near-perfect accuracy on math word problems where Chain of Thought only gets ~70%. The key insight: let AI do what it's good at (logic) and computers do what they're good at (calculation)!

Try It Yourself!

Use the interactive example below to see how generating code instead of calculating directly leads to more accurate results.

Program of Thought — Code Instead of Text

💻 Program of Thought — instead of text reasoning, the model generates and executes code. This eliminates calculation errors and gives precise results.

Choose example:

Task:

If you invest $1000 at 5% annual interest for 5 years with yearly compounding, how much will you have?

Text Reasoning

1.Initial amount: $1000

2.Interest: 5% = 0.05 per year

3.Formula: P × (1 + r)^n

4.1000 × 1.05^5 ≈ 1000 × 1.276

5.Answer: approximately $1276

Result:

~$1276

⚠️ Imprecise result due to rounding

Program of Thought

👆 Click "Run" to see how code solves the problem

💡 When to use:

🔢

Math

📅

Date/Time

📊

Statistics

Program of Thought prompt example:

If you invest $1000 at 5% annual interest for 5 years with yearly compounding, how much will you have?

Write Python code to solve this problem. Print the result.

Key Insight

LLMs are great at generating code but bad at mental math. Program of Thought uses the model's strength (code) to compensate for its weakness (arithmetic). Code is executed by an interpreter that doesn't make mistakes.

Try it yourself

Interactive demo of this technique

Technique Comparison

Demo Mode

Pre-recorded responses

TaskIntermediateMath

Deposit $10,000 at 5% annual interest compounded monthly for 3 years. What's the final amount?

Without technique

Prompt

Deposit $10,000 at 5% annual interest compounded monthly for 3 years. What's the final amount?

Response

Using compound interest formula: A = P(1 + r/n)^(nt) A = 10000(1 + 0.05/12)^(12×3) A = 10000 × 1.004167^36 A ≈ $11,614.72

Tokens:38/68

Time:420ms

Quality:

With Program of Thought

With technique

Prompt

Solve by writing Python code for exact calculation. Problem: Deposit $10,000 at 5% annual interest compounded monthly for 3 years. What's the final amount? Write code, execute it mentally, give exact answer.

Response

# Data
principal = 10000  # initial amount
annual_rate = 0.05  # annual rate
compounds_per_year = 12  # monthly compounding
years = 3

# Compound interest formula
monthly_rate = annual_rate / compounds_per_year
total_periods = compounds_per_year * years

final_amount = principal * (1 + monthly_rate) ** total_periods

print(f"Final amount: ${final_amount:.2f}")
print(f"Interest earned: ${final_amount - principal:.2f}")

🖥️ EXECUTION: monthly_rate = 0.05 / 12 = 0.00416666... total_periods = 12 × 3 = 36 final_amount = 10000 × (1.00416666...)^36 final_amount = 10000 × 1.16161678... final_amount = 11616.17

✅ EXACT ANSWER: Final amount: $11,616.17 Interest earned:$ 1,616.17

💡 Code advantage: penny-accurate, no rounding errors from manual calculation.

👁️Define variables: principal, rate, periods

🔢monthly_rate = 0.05/12 = 0.004166...

🔢total_periods = 12 × 3 = 36

✅final = 10000 × 1.004166^36 = $11,616.17

Tokens:72/285

Time:920ms

Quality:

Why this works

Program of Thought gives exact answer ($11,616.17) vs approximate ($11,614.72). Code eliminates rounding errors.

1 / 4

Practice Challenges

Create a free account to solve challenges

3 AI-verified challenges for this lesson

Related lessons:Chain Of Thought Least To Most

This lesson is part of a structured LLM course.

My Learning Path

The Solution: Use a Calculator, Not Mental Math

Think of it like using a calculator vs mental math:

1. Read the problem: "Calculate compound interest over 5 years..."
2. Write code: Express the logic in Python/JavaScript
3. Execute: Run the code and get exact results
4. Explain: Describe what the code does and the answer

Where Is This Used?

Financial Calculations: Compound interest, loan payments, ROI

Data Analysis: Statistical calculations, aggregations

Scientific Computing: Physics, chemistry calculations

Date/Time Math: Days between dates, timezone conversions

# Data principal = 10000 # initial amount annual_rate = 0.05 # annual rate compounds_per_year = 12 # monthly compounding years = 3 # Compound interest formula monthly_rate = annual_rate / compounds_per_year total_periods = compounds_per_year * years final_amount = principal * (1 + monthly_rate) ** total_periods print(f"Final amount: ${final_amount:.2f}") print(f"Interest earned: ${final_amount - principal:.2f}")