Fine-Tuning vs Prompting: When You Actually Need to Train

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-models-fine-tuning-vs-prompting-r8a8-teen

1. What does fine-tuning do to a language model?

   1. It changes the model's hardware requirements
   2. It adjusts how the model responds without changing its core design
   3. It trains the model from scratch on entirely new data
   4. It teaches the model new patterns by continuing its training on specific data

2. A company wants their AI to always format financial reports in a very specific way for thousands of different queries. What approach would likely work best?

   1. Switch to a different AI company
   2. Fine-tune the model on examples of correct formatting
   3. Add a system prompt explaining the format
   4. Use zero-shot prompting with no examples

3. What is 'few-shot prompting'?

   1. Using the smallest available AI model for a task
   2. Training a model with only a small number of examples
   3. Including several examples directly in the prompt to show the AI what you want
   4. Testing prompts multiple times to find the best one

4. A teenager builds a chatbot for fun that should respond in the style of a medieval knight. What is the most practical way to achieve this?

   1. Train the model on historical knight documents
   2. Use a system prompt describing the knight persona
   3. Build a completely new AI from scratch
   4. Fine-tune a model on knight dialogue from movies

5. Why is fine-tuning described as 'expensive and slow' in the context of AI development?

   1. It can only be done by expensive professional AI companies
   2. It requires buying new computer hardware every time
   3. It takes longer than writing prompts
   4. It needs significant computing power, time, and often costs money for training resources

6. What does the lesson say most people actually need instead of fine-tuning?

   1. Human reviewers for every response
   2. New computer programs
   3. More expensive AI models
   4. Better examples in their prompts

7. A startup includes 5 example customer service replies in every prompt to get consistent responses. According to the principles in this topic, what should they do?

   1. Keep using prompting since it works well for this case
   2. Switch to a cheaper AI model
   3. Fine-tune their own model immediately
   4. Hire more human agents

8. What does the term 'retrieval' most likely mean in the phrase 'better prompts plus retrieval'?

   1. Searching through old fine-tuning data
   2. Looking up relevant information to include in the prompt
   3. Recovering deleted AI conversations
   4. Finding the right AI model to use

9. A medical research company wants to use an AI to analyze their private clinical notes. They need the AI to understand medical terms and formats specific to their organization. What approach makes sense?

   1. Fine-tune a model on their clinical notes
   2. Ask patients to re-write their notes
   3. Use a general-purpose AI without any customization
   4. Write a detailed system prompt

10. What is the main reason the lesson recommends trying prompting 'hard' before fine-tuning?

    1. Prompting is free but fine-tuning costs money
    2. Fine-tuning makes the AI too smart
    3. Fine-tuning breaks the AI sometimes
    4. Most tasks can be solved with good prompts without the time and expense of training

11. What does it mean to 'scale' a task in the context of deciding between prompting and fine-tuning?

    1. Enlarging the computer screen
    2. Expanding the AI's vocabulary
    3. Running the same task thousands or millions of times
    4. Making the text font larger

12. A student wants an AI to write haikus about space. They add three haiku examples to their prompt and ask the AI to write more like those. What technique are they using?

    1. Model training
    2. Zero-shot prompting
    3. Few-shot prompting
    4. Fine-tuning

13. According to the concepts covered, what is a system prompt?

    1. A prompt that runs automatically
    2. A backup of your original prompt
    3. Instructions embedded in the AI's configuration that set its overall behavior
    4. A prompt that only works on computer systems

14. What makes fine-tuning different from regular prompting?

    1. Fine-tuning actually changes how the model behaves permanently
    2. Fine-tuning only works on weekends
    3. Fine-tuning only works with images, not text
    4. Fine-tuning requires no examples

15. A legal tech company needs their AI to include exact citations in a specific format for millions of different legal queries. Why would fine-tuning be appropriate here?

    1. Legal work is always done by fine-tuned models
    2. The exact, consistent format is needed at massive scale and prompting might not guarantee consistency
    3. They can afford the expensive GPU costs
    4. Fine-tuning makes the AI faster