問題文
An analyst asks a model to classify support tickets into three categories. The first attempt gives only the instruction and the ticket. The second attempt prepends three solved tickets with their categories. What is the second attempt, and what does it change?
選択肢
- It is supervised fine-tuning on a very small dataset, and the three examples are folded into the parameters so that later requests need not include them.
- It is few-shot prompting, and it shows the model the expected input and output shape at inference time without altering the stored parameters.
- It is retrieval-augmented generation, and the three examples are added to the input as retrieved passages before the model runs.
- It is prompt tuning, and the three examples are converted into virtual tokens that an optimizer adjusts by gradient descent before the request is served.