Blog Details

Overview

In this lesson, you will learn:

• How to control AI outputs using temperature, maximum tokens, and system messages.
• Where and how to test these parameters in real-world tools.
• Beginner-friendly, hands-on strategies to fine-tune AI responses.

By the end, you will be able to experiment with AI parameters in practical environments to achieve desired outputs.

Key Concepts

• Temperature: Adjusts randomness and creativity (low → predictable, high → creative).
• Maximum Tokens: Controls the length of AI outputs.
• System Messages: Sets AI role, tone, and behavior.
• Output Control: Helps you make responses consistent, relevant, and in your desired style.

Practical Approach

1. Where to Test These Parameters

• ChatGPT (OpenAI Playground / ChatGPT Plus):
  • Temperature slider: 0–1
  • Max tokens field: limit response length
  • System messages: Use the “Custom Instructions” or GPT system role
• Google Bard / Gemini:
  • Options for creative vs. factual responses
  • Input boxes can handle step-by-step instructions and role prompts
• Notion AI / Dify / Copy.ai:
  • Set tone and output length
  • Use templates to test different temperatures or creativity levels
• Coding Environment (Python + OpenAI API):

from openai import OpenAI
client = OpenAI()

response = client.chat.completions.create(
    model="gpt-4",
    messages=[{"role":"system","content":"You are a friendly teacher."},
              {"role":"user","content":"Explain photosynthesis."}],
    temperature=0.7,
    max_tokens=150
)
print(response.choices[0].message.content)

2. How to Test Parameters Practically

Step 1: Pick a Task

• Examples: Explain AI, write a poem, summarize text, or solve a math problem.

Step 2: Adjust Temperature

• Low temperature (0–0.3) → precise, factual, deterministic output.
• High temperature (0.7–1.0) → creative, varied, playful output.

Step 3: Limit Output Length

• Use max tokens to control concise summaries or long-form responses.

Step 4: Use System Messages

• Define AI’s role for tone and style (teacher, assistant, coach, creative writer).

Step 5: Experiment Iteratively

• Run multiple versions with different settings.
• Compare outputs and observe the effect of each parameter.

Practical Examples

Example 1: Controlling Creativity (Temperature)

• Tool: ChatGPT Playground
• Prompt: “Write a one-line description of AI.”
• Temperature 0.2 → “AI is the simulation of human intelligence by machines.”
• Temperature 0.8 → “AI is like teaching machines to dream, reason, and imagine!”

Example 2: Controlling Length (Max Tokens)

• Tool: ChatGPT Playground
• Prompt: “Summarize photosynthesis.”
• Max tokens 50 → Short summary
• Max tokens 150 → Detailed explanation

Example 3: System Message / Role

• Tool: OpenAI API or Playground
• System Message: “You are a friendly teacher for 10-year-olds.”
• Prompt: “Explain the water cycle.”
• Output: Simple, engaging, age-appropriate explanation

Hands-On Beginner Activity

1. Open ChatGPT Playground or your preferred AI tool.
2. Select a topic (e.g., “AI in daily life”).
3. Set temperature = 0.2, run prompt → observe output.
4. Change temperature = 0.8, run prompt → observe differences in creativity.
5. Limit max tokens = 50, then 150, compare outputs.
6. Add a system message (role = teacher/coach) → note changes in tone/style.
7. Record results in a small table:

| Parameter | Value | Output Example | Observations |

Exercises

1. Summarize a paragraph in 50 tokens and then in 150 tokens. Compare.
2. Write a short poem about AI at temperature 0.2 and 0.8.
3. Ask AI to explain a concept using a system message as a teacher.
4. Experiment with all three parameters together: temperature, max tokens, system role.

Summary & Key Takeaways

• Temperature controls randomness and creativity.
• Max tokens controls output length.
• System messages define role, tone, and behavior.
• Use real tools (ChatGPT Playground, APIs, Notion AI) to experiment practically.
• Iterative testing builds intuition on how parameters affect outputs, preparing learners for advanced prompt optimization and automation.