NCA-GENL Question 3
Single answerAn AI team is developing a generative language model for summarizing documents. They decide to use supervised learning to fine-tune a pre-trained model. Which of the following is a key aspect of supervised learning that applies in this scenario?
- A
Using unlabeled data to train the model
- B
Providing labeled input-output pairs for fine-tuning
- C
Allowing the model to learn without human intervention
- D
Employing reinforcement signals to guide training
Show answer and explanation
Correct answer: B
Explanation
Supervised learning is characterized by the use of labeled input-output pairs. In this case, fine-tuning a pre-trained generative language model for summarization requires labeled training data, where each document (input) is paired with its corresponding summary (output). This distinguishes it from other techniques like unsupervised learning (which uses unlabeled data) or reinforcement learning (which uses reward signals).
- A. Incorrect.
Supervised learning requires labeled data, so using unlabeled data directly does not apply in this context.
- B. Correct.
Supervised learning involves providing labeled input-output pairs, such as documents and their corresponding summaries, which is key to fine-tuning the generative model.
- C. Incorrect.
While unsupervised learning involves learning without human intervention, supervised learning explicitly relies on labeled data, making this incorrect for the given scenario.
- D. Incorrect.
Reinforcement learning involves reward signals to guide training, which is not part of supervised learning or the described fine-tuning process.