Chain-of-Thought (CoT)

Making LLMs Think Step by Step

CoT Reasoning Flow

Chain-of-Thought prompting improves LLM performance on reasoning tasks by asking the model to show its work. Instead of jumping to an answer, the model breaks down the problem into intermediate steps.

Why It Works

LLMs generate one token at a time. When forced to generate reasoning steps, each step provides additional context for the next token prediction, leading to more accurate final answers.

Variants

• Manual CoT: You provide step-by-step examples
• Zero-shot CoT: Simply add "Let's think step by step" to your prompt
• Self-consistency: Generate multiple CoT paths and vote on the most common answer
• Tree of Thought: Explore multiple reasoning branches, backtrack when needed

1from openai import OpenAI
2client = OpenAI()
3
4# Without CoT (often wrong for complex math)
5response_no_cot = client.chat.completions.create(
6    model="gpt-4o",
7    messages=[{"role": "user", "content": 
8        "If a store has 5 apples and gets 3 shipments of 12 apples each, "
9        "then sells 60% of all apples, how many are left?"
10    }]
11)
12
13# With CoT (much more reliable)
14response_cot = client.chat.completions.create(
15    model="gpt-4o",
16    messages=[{"role": "user", "content": 
17        "If a store has 5 apples and gets 3 shipments of 12 apples each, "
18        "then sells 60% of all apples, how many are left?\n\n"
19        "Let's solve this step by step."
20    }]
21)