Glossary of AI Prompt Terms

It's becoming clearer that one of the skills we will all need master as AI advances is the knowledge and use of terms used in machine learning. A knowledge of such terms will help you to understand and thereby guide your AI generation. So, in the spirit of learning here are some common AI prompt related terminology

1. Prompt Engineering

   Definition: The practice of designing and structuring prompts to elicit desired responses from large language models (LLMs).

   Example: If you want a model to generate a poem, instead of just saying "Write a poem," you could engineer the prompt as follows:"Write a 4-line poem about the beauty of the sunset, using metaphors."

   
   

2. Chain-of-Thought Prompting

   Definition: A technique that encourages LLMs to explain their reasoning step by step, improving the accuracy of responses.

   Example: If you're asking the AI how to calculate the area of a triangle, you might prompt it with: "First, explain how to calculate the area of a triangle. Then, apply the formula to a triangle with a base of 5 units and a height of 3 units."

   
   

3. Generative AI

   Definition: AI systems capable of generating content such as text, images, or music based on prompts.

   Example: "Generate a short story about a dragon who befriends a knight in the middle of a battle."This would lead the AI to generate a complete narrative based on the input.

   
   

4. Large Language Models (LLMs)

   Definition: Advanced AI models trained on extensive text data, enabling them to understand and generate human-like text.

   Example: "What is the impact of AI on healthcare?"An LLM would generate a well-informed, detailed response drawing on its vast training data.

   
   

5. One-Shot Prompting

   Definition: Providing a single example within the prompt to guide the AI’s response.

   Example: "Translate the following English sentence into French: 'I am going to the market.' "This prompt provides one example sentence to help guide the translation.

   
   

6. Zero-Shot Prompting

   Definition: Designing prompts that enable the AI to perform tasks without prior examples, relying on its pre-existing knowledge to generate responses

   Example: "Translate the sentence 'I will be there soon' into German."This prompt asks for a task without providing an example translation, relying on the AI's internal knowledge.

   
   

7. Prompt Injection

   Definition: Incorporating additional instructions or contextual details into a prompt to influence the AI's behavior.

   Example:"Incorporate a surprising plot twist into the following short story about a detective solving a crime."This prompt includes an extra instruction to shape the story's direction.

   
   

8. Adversarial Testing

   Definition: Evaluating AI models by providing inputs designed to challenge their robustness and identify vulnerabilities.

   Example: Asking the model to generate a list of 100 valid domain names when it can only generate a few would challenge its capabilities and reveal weaknesses.

   
   

9. Hallucinations

   Definition: Instances where AI models generate responses that sound plausible but are incorrect or nonsensical.

   Example: Asking the model:"Who won the 1999 World Series?"And the model responds: "The Detroit Tigers," which is incorrect (the correct answer is the New York Yankees).

   
   

10. Bias in AI

    Definition: Systematic favoritism or prejudice in AI outputs, which can arise from biased training data or flawed model design.

    Example: If an AI model trained mostly on text data from Western sources provides responses that consistently favor Western viewpoints, it reflects bias in the model's training data.

    
    

11. Mean Average Precision at k (MAP@k)

    Definition: A metric used to evaluate the quality of a model's output across multiple prompts, assessing the precision of the top-k responses.

    Example: If a search engine provides 5 results for a query, MAP@k would measure the precision of the top 5 answers provided, ranking them based on relevance.

    
    

12. Seed

    Definition: A specific value used in the AI model’s random number generation,

    ensuring consistency in the AI’s output when using the same prompt.

    Example: If you use a seed like "12345" with an AI image generator, it will generate the same image when the prompt is run again with that seed.

    
    

13. Iteration

    Definition: The process of refining or adjusting the prompt and its details in response to the AI’s outputs.

    Example: After the first response from the AI is not what you expected, you adjust the prompt to be more specific:"Can you generate a detailed scene of a beach at sunset with vibrant colors and a calm sea?"

    
    

14. Aspect Ratio

    Definition: The proportional relationship between the width and height of an image. Example:"Generate an image of a sunset with a 16:9 aspect ratio."This instructs the AI to create an image that fits within a widescreen format.

    
    

15. Feedback Loop

    Definition: The process of adjusting and improving prompts based on the output received, allowing users to learn what works and refine their prompting strategies.

    Example :You might input a prompt for an AI to generate an image of a "dragon flying over a city" and then refine it after feedback:"Add more details to the dragon's wings and make the city look futuristic."