Comparison questionMedium~10 minAI Fundamentals / 7 of 20

Interview question

Compare inference, training, fine-tuning, prompting, and RAG.

Separates runtime model use from model training, fine-tuning, prompt design, and retrieval-augmented generation as different engineering levers.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

TL;DR

Inference is running a trained model to produce an output. Training creates model parameters from data, while fine-tuning further adapts an existing model with examples. Prompting changes the instructions and runtime context without changing model weights. RAG retrieves external information at request time and places selected evidence in the context. Prompting is usually the first lever for task framing, RAG helps with current or private knowledge, and fine-tuning can improve repeatable behavior or style when enough evidence justifies it.

Full answer below

Practice focus

The ability to choose the least complex adaptation method that addresses behavior, knowledge, freshness, or control.

Answer first

Answer path

Core answer first, then checkpoints toward a stronger answer.

Start
Core answer
Start here. This is the direct spoken answer to practice first.
Overview
These terms describe different points in the system lifecycle and solve different problems; treating them as interchangeable leads to expensive designs.
Inference is running a trained model to produce an output. Training creates model parameters from data, while fine-tuning further adapts an existing model with examples. Prompting changes the instructions and runtime context without changing model weights. RAG retrieves external information at request time and places selected evidence in the context. Prompting is usually the first lever for task framing, RAG helps with current or private knowledge, and fine-tuning can improve repeatable behavior or style when enough evidence justifies it.
Checkpoint
Stronger answer
Optional upgrade with examples, trade-offs, and sharper phrasing.
I choose from the failure. If the model lacks current product policy, fine-tuning is usually the wrong first answer because the fact will change; retrieval gives freshness and provenance. If the model repeatedly produces the wrong response shape despite clear instructions and constrained output, fine-tuning may be worth evaluating. Training a foundation model is a different scale of investment. These approaches can combine, but each adds data, evaluation, deployment, and maintenance work.
Finish
Advanced answer
Optional deeper angle for production judgment and nuance.
I establish a baseline with the simplest viable method, use a fixed evaluation set, and preserve an ablation path so we know what produced the gain. RAG introduces retrieval and authorization failures; fine-tuning introduces dataset quality, privacy, versioning, and rollback concerns; long prompts add latency and cost. A model can also memorize undesirable data during adaptation. The decision therefore includes lifecycle ownership, not just one benchmark score, and every change is evaluated against the same product success criteria.

Review signals

Use these to catch weak phrasing, shallow reasoning, and risky claims.

Common mistakesCandidateCommon mistakesMistakes that weaken the answer or signal shallow understanding.Open details

Details

Using fine-tuning to keep frequently changing facts current.
Calling any prompt example fine-tuning.
Assuming RAG changes model weights.
Combining every technique before establishing a baseline.

Red flagsCandidateRed flagsSignals that an interviewer may treat as risky.Open details

Details

The team cannot name the failure each added component solves.
Private training data has no consent, retention, or deletion boundary.
A quality gain cannot be attributed because several levers changed at once.

Follow-up prompts

Quick interviewer prompts inside this drill.

Follow-up questionsCandidateFollow-up questionsLikely next questions after the main answer.Open details

Details

Treat these as the next interviewer turn. Answer one out loud, then save the gap in your practice notes.

When is RAG preferable to fine-tuning?
What is a good baseline before adaptation?
What new failure modes does RAG add?
When might fine-tuning improve behavior?

Continue the section

Next in AI Fundamentals: Hallucination

Step 8 of 20

What does hallucination mean in an AI application?

Practice route

Read core answer Review checklist Follow-up prompts Track progress

Finished this drill?

Practice freely. Sign in only when you want to save completion, confidence, notes, or bookmarks.

New to Aporeon? Create account

Drill details

Human editorial direction

Aporeon is shaped by Aleksandar Tomovski, a software developer with experience on both sides of technical interviews. Content is reviewed for accuracy, natural spoken delivery, useful depth, and honest trade-offs.

About the creator Editorial standards

Was this drill useful?

Help improve the interview library.

Found this answer useful?

Say thanks with a standalone, one-time $5 contribution. No account required.

No subscription or perks.Choose another amount

Opening drill

Compare inference, training, fine-tuning, prompting, and RAG.

Separates runtime model use from model training, fine-tuning, prompt design, and retrieval-augmented generation as different engineering levers.

1Say itAnswer out loud first

2CompareCompare with Core answer

3RefineRefine with Stronger and Advanced

Compare inference, training, fine-tuning, prompting, and RAG.

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section

Compare inference, training, fine-tuning, prompting, and RAG.

Practice focus

Answer path

Overview

Review signals

Follow-up prompts

Continue the section