Deep dive · module 71 of 100
Understanding fine-tuning
Understand when fine-tuning is useful and when a good prompt suffices.
After this lesson
You can explain what fine-tuning does and when it is worth the effort.
In view · How it connects
Fine-tuning builds on examples
First determine if a good prompt already sufficiently supports the stable task.
Fine-tuning uses specific examples to further adjust model behaviour.
Test on examples that you did not use for training.
Part 1 of 3 · Repeatable task
Your preference for less movement is on. Choose each part yourself.
What is fine-tuning?
Further training the model on your specific examples so that it fits your style or task better.
When useful?
For a recurring task with a clear style or format where prompts alone are not enough.
When not?
If you have few examples or the task is not yet stable. Start with prompts.
New practice example
From explanation to a usable result.
Fine-tuning teaches a model extra patterns from examples; it is not an obvious way to keep up with current facts.
Fictional practice situationA fictitious team wants the same short format for service messages every time. The opening hours change regularly and are contained in a separate approved document.
View a possible solution
This is how it can be done
First, test a clear prompt with several good formatting examples. Use the correct source for current opening hours. Only consider extra training if the recurring behaviour needs to be demonstrably improved with representative examples.
Why this works
A style problem and a knowledge problem do not automatically require the same solution. Training with old opening hours will not make them current.
Make it slightly more difficult
Create two test sets: known style examples and new situations. Which results must you compare before speaking of improvement?
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
A style problem and a knowledge problem do not automatically require the same solution. Training with old opening hours will not make them current.
A step further
Create two test sets: known style examples and new situations. Which results must you compare before speaking of improvement?
A prompt to get you started
Explain in a maximum of 100 words when fine-tuning is useful and when a good prompt is sufficient.
Apply it
You want to summarise one public report. A colleague suggests fine-tuning a model immediately.
Which approach best fits this situation?
- Train a model immediately for this one-off task.
- Use fine-tuning as a guarantee for up-to-date facts.
- Start with a defined prompt and appropriate source input; evaluate the necessity first.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
When is fine-tuning usually not the first choice?
- If you have few examples or the task is not yet stable.
- Always.
- Only for large companies.
Extra practice
Name one task where fine-tuning could be useful and one where a prompt suffices.
Useful: fixed writing style for client emails. Prompt suffices: one-off summary of a report.
