問題文
A bank must make a general chat model answer questions about its own internal product catalog, which is updated every week. The team has no labeled training data and wants to ship in days. Which customization approach should they try first?
選択肢
- Raising the temperature so that the model explores more answers and reaches the current catalog wording on its own.
- Importing a compatible open model onto a dedicated AI cluster, where a self-hosted model can read the catalog directly from the internal file share.
- Retrieval-augmented generation, because it retrieves the current catalog content at request time and grounds the answer in it without training a model.
- Fine-tuning a base model on the catalog, which is the only way to make answers reflect content the base model has not seen.