問題文
A fine-tuning corpus assembled from support transcripts contains customer names, addresses, and account numbers. Which handling is appropriate before the corpus is used?
選択肢
- Keep the values but restrict access to the trained model to internal staff, where the risk arises only if the model is exposed externally and an internal deployment does not require the corpus to be modified.
- Replace the identifying values with consistent placeholders so the structure is preserved while the values are not learned.
- Delete every row that contains any identifying value.
- Encrypt the values in place so they appear as ciphertext in the corpus.