問題文
A pipeline returns the same generic answer for many different questions. What is happening?
選択肢
- The retrieval step returns the same low-relevance passages regardless of the question, so the model is given nothing specific at all to work with
- The max output tokens is set too high, so the model fills the available room with general background instead of answering the specific question
- The temperature is set too high, so the model drifts away from the specifics of each question and falls back on the safest general statement it can make
- The content filter is annotating the responses, and the application is falling back to a safe default message whenever an annotation is present