In an AI team's design review, there is debate over whether to use RAG or fine-tuning. To share decision criteria, which option BEST explains how to choose between the two?
A question that asks for the correct explanation of how to choose between RAG and fine-tuning.
Use fine-tuning for latest and changing information, and RAG to fix the writing style.
The roles of RAG and fine-tuning are reversed.
Referencing the latest information is RAG, and fixing the writing style is fine-tuning, so it is incorrect.
Both are methods that search external data at every inference and use it in the answer.
RAG searches external data at every inference and reflects it in the answer, but fine-tuning is a method that incorporates data into the model weights at training time and does not perform external search at inference time.
Summarizing both as "methods that search each time" confuses the fundamental difference between RAG, which references external data, and fine-tuning, which builds knowledge into the weights, so it is incorrect.
RAG is more expensive in every situation.
In general, cost is RAG < fine-tuning, and it is not true that RAG is always more expensive.
The cost ordering is reversed, so it is incorrect.
Use RAG for latest and changing information, and fine-tuning to fix the writing style and behavior.
Correct. RAG is suited to referencing external, latest, frequently changing information and reflecting it, while fine-tuning is suited to building a fixed writing style, output format, and behavior into the model.
Remember the core of choosing between RAG and fine-tuning.
・RAG (retrieval-augmented generation): References external, latest, frequently changing, or company-specific information and reflects it (no retraining needed). Example: search the latest product manuals or internal policies and use them in the answer.
・Fine-tuning: Builds repetitive behavior into the model itself, such as a fixed writing style, output format, classification, and technical terminology (updates the weights). Example: answer in the tone and standard format of the company's own support desk.
Options that reverse the roles or reverse the cost ordering are the classic traps. The two can also be combined.