Decide whether the model needs training or better evidence
Compare fine-tuning with retrieval through two different metaphors: changing how someone answers and giving them the right reference material. The choice depends on behavior, knowledge freshness, cost, and control.
THE QUESTION THIS PAGE ANSWERS
ANSWER FIRSTWhat is the key idea behind “Decide whether the model needs training or better evidence”?
Compare fine-tuning with retrieval through two different metaphors: changing how someone answers and giving them the right reference material. The choice depends on behavior, knowledge freshness, cost, and control.
Ask what should change: behavior or knowledge. Use retrieval when facts need to stay current and inspectable. Consider fine-tuning when a repeatable behavior or format needs to become the model’s default.
Classify your request as behavior, knowledge, style, or a mix before choosing a method.
Fine-tuning a changing knowledge base, or retrieving a behavior problem.
Fine-tuning sends the AI back to class and drills knowledge into its body. Feeding it docs (also called a knowledge base, or RAG) lets it take an open-book exam — look it up as needed. What most companies actually need is the latter.
Same goal: get the AI to answer questions about your company. The left path is fine-tuning; the right path is feeding it docs. Hit the button below and see what each path actually involves.
One key difference: fine-tuning changes the model itself, like drilling knowledge into muscle memory. Feeding it docs doesn't touch the model at all — it just lets the model open a book during the exam. So when the docs update, the open-book side just swaps the book; the class side has to sit through the course again.
The idea is clear — how do you pick on a real task? These four are the most common company scenarios. Click one to see the recommended plan and why.
| 🏫 Fine-tuning (class) | 📚 Knowledge base (open-book) | |
|---|---|---|
| Time to work | Days to weeks | Ready the same day |
| Cost scale | Starts at tens of thousands of dollars, up to hundreds of thousands — plus dedicated staff to maintain it | Often one to two orders of magnitude cheaper; a regular team can stand it up |
| Updating knowledge | When the docs change, you retrain from scratch | Swap a file and it's live |
| Best for | Changing style, voice, or picking up a specialist "feel" | Knowledge that changes: policies, prices, document Q&A |
Here's a money-saving order: try a prompt first, then attach a knowledge base, and only then consider fine-tuning. A lot of "we must fine-tune" needs are solved by writing a careful prompt. If that's not enough, hang a knowledge base on it and let it take the exam open-book. If you've tried both and it's still a bit short, that's when you spend real money sending it back to class. Side note: when someone at dinner says "we trained a model," most of the time they only did the first two steps. This page helps you hear what they actually did.
Why “Watch a demo · Class vs open-book, side by side” can find relevant content
“Fine-tuning sends the AI back to class and drills knowledge into its body.” moves retrieval beyond storing material: the real question is how to find what is relevant. That decision shapes the input quality of RAG, recommendation, and image-search systems.
Similarity is not the answer
In the flow described by “Same goal: get the AI to answer questions about your company.”, embeddings place items in a comparable semantic space and a neighbor index narrows the search. The final answer still depends on whether the retrieved chunks cover the question, whether the distance metric fits, and whether the evidence is current.
- Fine-tuning = class : knowledge is drilled into the model's body; timeline in weeks, cost in the tens of thousands
- Knowledge base = open-book exam : docs sit on a shelf, look them up as needed; ready the same day, update by swapping a file
- If knowledge updates often, go open-book ; changing personality, style, or learning a specialist feel is what needs class
Separate findable from relevant
Turn “Here's a money-saving order: try a prompt first, then attach a knowledge base, and only then consider fine-tuning .” into a small test: prepare queries with known answers, record relevance, misses, and distractors, then decide whether chunking, the index, or reranking needs to change.
From “Watch a demo · Class vs open-book, side by side” to “Find your match · Which one fits your scenario”
“Watch a demo · Class vs open-book, side by side” grounds the problem in “Same goal: get the AI to answer questions about your company. The left path is fine-tuning; the right path is feeding it docs. Hit the button below and see what each path actually involves”. “Find your match · Which one fits your scenario” then moves it toward “The idea is clear — how do you pick on a real task? These four are the most common company scenarios. Click one to see the recommended plan and why”. Together, they show that the lesson is not just a conclusion to remember, but a claim with conditions.
Carry the judgment into the next situation
The same logic applies to retrieval: define what counts as relevant, check whether recall covers the question, and then inspect whether ranking, chunking, or freshness pushed useful evidence out.
- “Watch a demo · Class vs open-book, side by side”: Same goal: get the AI to answer questions about your company. The left path is fine-tuning; the right path is feeding it docs. Hit the button below and see what each path actually involves
- “Find your match · Which one fits your scenario”: The idea is clear — how do you pick on a real task? These four are the most common company scenarios. Click one to see the recommended plan and why
- “The closing point”: Money-saving order : try a prompt first, then attach a knowledge base, and only then consider fine-tuning
The final “The closing point” brings the discussion to “Money-saving order : try a prompt first, then attach a knowledge base, and only then consider fine-tuning”. The useful thing to carry forward is knowing which judgments must be revisited when input, scale, or risk changes.
✅ What this page wants to share with you
- Fine-tuning = class: knowledge is drilled into the model's body; timeline in weeks, cost in the tens of thousands
- Knowledge base = open-book exam: docs sit on a shelf, look them up as needed; ready the same day, update by swapping a file
- If knowledge updates often, go open-book; changing personality, style, or learning a specialist feel is what needs class
- Money-saving order: try a prompt first, then attach a knowledge base, and only then consider fine-tuning
INTERACTIVE PRACTICE
Turn a vague request into a useful prompt
Clarify the goal, context, and constraints, then carry the finished prompt into the AI tool you use.
I now start by asking how often the knowledge changes. Frequently changing material points to RAG; fine-tuning is for stable changes in style or behavior. That is more useful than asking only how much data we have.
With only a few hundred high-quality examples, could fine-tuning simply make the model resemble the training set? Besides a held-out test set, what simple checks show that it learned a behavior rather than memorized examples?
A three-column comparison would help: answer the same question with a base model, RAG, and a fine-tuned model, then record knowledge freshness, format stability, and maintenance cost. That would make the trade-off concrete.
No discussion on this article yet.