An employee wants to improve the answer quality of an internal FAQ bot. The plan is to improve output quality by refining how instructions and example sentences are written, without retraining the model or adding an external database. Which concept represents this approach?
Select the approach that improves output by refining instructions without retraining.
Fine-tuning
Fine-tuning is an approach that additionally trains the model's weights with labeled data to adapt it to a specific task.
This question is about an approach that improves output by refining instructions without retraining, so it is incorrect.
Pre-training
Pre-training is the process of training a model from scratch on a large amount of data to acquire general-purpose capabilities.
It is not an approach that improves with instructions without retraining, so it is incorrect.
Prompt engineering
This is correct. Prompt engineering is an approach that improves output quality by refining how instructions, examples, and output formats are written, without changing the model's weights. It is the first means to try, as it is the easiest.
Retrieval-augmented generation (RAG)
RAG is an approach that searches an external knowledge source before generating a response and uses it as the basis, and shares the point of requiring no retraining.
However, this question explicitly states the plan to 'not add an external database, and refine how instructions and example sentences are written,' so RAG, which references external knowledge, does not fit the requirement, making it incorrect.
Distinguish the two improvement approaches that do not retrain.
- Prompt engineering: refine 'how the input is written' such as instructions, examples, and format (no external data needed).
- RAG: search external knowledge and use it as the basis (adding a data source required).
In this question, 'without adding an external database' is the deciding factor for prompt engineering. Fine-tuning and pre-training are approaches that change the weights, contrary to the no-retraining plan.