Skip to main content

Training Techniques

Updated Oct 24, 2025 ·

Techniques​

These are different ways to guide a model in understanding and answering tasks.

  • Zero-shot Learning

    • No examples are given before the task
    • The model relies on prior training to answer new questions
    • Example: “Write a poem about the tranquility of mountains.”

  • One-shot Learning

    • One example is given before the question
    • Helps the model learn from a single pattern
    • Example: “Mexico City is the capital of Mexico. What is the capital of Vatican City?”

  • Few-shot Learning

    • Several examples are shown before the task
    • Helps the model understand context through multiple samples
    • Example: Asking for the capital of Malaysia while formatting with the country’s flag

Pattern Matching and Recognition​

Few-shot learning turns ChatGPT into a pattern-matching and pattern-generation engine:

  • Writing style for emails, formatting preferences for reports, etc.
  • Analyzes examples, mirrors patterns, generates new content.
  • Extends ChatGPT's capability beyond simple responses to complex tasks.

Chain of Thought (COT) Prompting​

Chain of Thought Prompting (COT) provides a roadmap for answering:

  • Zero-shot COT: Provide a scenario (e.g., traveling to space and encountering aliens) and prompt "think step by step."
    • Result: Thoughtful breakdown, revealing the model's reasoning.
  • One-shot COT: Provide one example to teach the model the approach.
    • Example: Acknowledging the number of astronauts interacted with to prevent errors.

These techniques enhance how you interact with ChatGPT, making it a more effective tool.