When Fine-Tuning Actually Beats Just Writing a Better Prompt

End-of-lesson check

15 questions · take it digitally for instant feedback at tendril.neural-forge.io/learn/quiz/end-builders-modelfamilies-ai-fine-tuning-when-r9a8-teen

1. A company wants their AI customer service chatbot to sound exactly like their brand's voice in thousands of responses every day. What approach would work best?

   1. Write a detailed prompt describing the brand voice
   2. Use RAG to pull from a brand guidelines document
   3. Add a system message with three example responses
   4. Fine-tune the model on past customer service replies

2. A user asks an AI to answer questions about their own uploaded PDF documents. Which approach would give the most accurate results?

   1. Train a new model from scratch on the documents
   2. Add the full PDFs to the system prompt
   3. Fine-tune the AI on the PDF documents
   4. Use RAG to retrieve relevant passages before answering

3. According to the concepts in this lesson, what happens when you fine-tune an AI model on factual information?

   1. The facts get baked in and become hard to update
   2. The model learns to cite sources automatically
   3. The model becomes better at looking up new facts
   4. The facts remain flexible and can be changed via prompts

4. A student needs help writing a single essay in a specific format their teacher requires. What should they try first before considering fine-tuning?

   1. Train a custom model from scratch
   2. Fine-tune a model on five essays from classmates
   3. Use a prompt with three example essays in that format
   4. Build a RAG system with essay examples

5. Why might fine-tuning cost more than simply writing a better prompt?

   1. Fine-tuning requires collecting training data, compute resources, and testing, while prompts cost nothing
   2. Prompts require expensive API calls for each use
   3. Fine-tuning is always free
   4. Fine-tuning uses less electricity than prompting

6. What advantage does LoRA offer when fine-tuning large language models?

   1. It enables efficient fine-tuning by updating only small parts of the model
   2. It allows fine-tuning without any computational cost
   3. It replaces the need for any training data
   4. It makes fine-tuned models smaller than the original

7. A healthcare company needs an AI that always outputs diagnosis information in the exact same structured format for their records system. What's the best approach?

   1. Write a prompt explaining the format each time
   2. Use RAG to pull format templates from a database
   3. Fine-tune the model on examples of properly formatted outputs
   4. Add format rules to the user instructions

8. What did the lesson suggest is often true when comparing a prompt with a few examples to a fine-tuned model for single tasks?

   1. The prompt needs hundreds of examples to compete
   2. Fine-tuning is always faster to set up
   3. The fine-tuned model always wins
   4. The prompt with examples usually performs as well or better

9. A startup wants their AI to generate marketing copy in their unique brand voice for every social media post. Why is fine-tuning appropriate here?

   1. Because prompts cannot handle creative tasks
   2. Because they need consistent style across high-volume output
   3. Because fine-tuning is the cheapest option
   4. Because the model needs to learn new facts about their product

10. What problem occurs when you try to use fine-tuning to teach an AI about rapidly changing company policies?

    1. The model cannot easily update its knowledge and becomes outdated
    2. The policies get stored in RAG instead
    3. Fine-tuning makes the model refuse to answer policy questions
    4. The model becomes too creative with the policies

11. A developer is building a feature where users ask questions about their uploaded company documents. The developer chooses RAG instead of fine-tuning. Why was this likely the right choice?

    1. Because fine-tuning doesn't work with text data
    2. Because RAG creates better marketing copy
    3. Because users need accurate knowledge from specific documents
    4. Because fine-tuning is faster to implement

12. A teacher wants an AI to grade essays using a specific rubric they created. They need the same criteria applied to every essay. What's the best approach?

    1. Train a completely new model
    2. Write a prompt describing the rubric for each essay
    3. Fine-tune on examples of graded essays using that rubric
    4. Use RAG to look up the rubric for each essay

13. A team needs their AI to output quarterly reports in a complex table format without having to repeat formatting instructions. What does the lesson recommend?

    1. Fine-tune to learn the exact table format
    2. Switch to a different AI model family
    3. Use RAG to retrieve format templates
    4. Write a longer prompt with the format each time

14. What does the lesson say about the relationship between prompt complexity and fine-tuning costs?

    1. They cost exactly the same
    2. Prompts are more expensive than fine-tuning
    3. Fine-tuning reduces the need for any prompts
    4. A better prompt is free while fine-tuning costs resources

15. An AI developer chooses fine-tuning for a project. Based on the lesson, which scenario would make this choice most justified?

    1. Generating a single poem in a specific style
    2. Looking up product prices from a database
    3. Teaching a model to write in a specific company's voice for thousands of daily emails
    4. Answering questions about a large library of policy documents