フリー問題

問題文

Which treatment is most appropriate for the risk that a generative assistant produces confident but incorrect statements to customers?

選択肢

1. Add a disclaimer to the interface stating that responses are generated automatically and may contain errors and require customers to acknowledge that disclaimer before they begin using the assistant.
2. Restrict the assistant to answering from an approved source and show the source with each answer.
3. Increase the size of the underlying model so that its factual knowledge is broader and the frequency of incorrect statements falls as a result of the improvement.
4. Log every response so that incorrect statements can be identified after customers report them and the pattern can be analyzed by the quality team each month.