31. Which of the following is a disadvantage of GPT models?
A) They require a large amount of training data
B) They are not suitable for large-scale language tasks
C) They have a limited vocabulary
D) They are not capable of generating coherent text
32. What is the purpose of the “beam search” decoding technique used in GPT models?
A) To control the length of generated text
B) To limit the vocabulary used in generated text
C) To prioritize the most probable words for generation
D) To increase the diversity of generated text
33. What is the purpose of the “attention mechanism” in GPT models?
A) To generate text based on input
B) To learn features from input
C) To compute loss during training
D) To allow the model to focus on relevant parts of input
34. What is the “zero-shot” performance of GPT-3?
A) The ability to generate text in a language the model has not been trained on
B) The ability to generate text for a specific task without fine-tuning the model
C) The ability to generate text with a limited vocabulary
D) The ability to perform a task accurately with very few examples
35. What is the purpose of the “pre-training” phase in GPT models?
A) To generate text based on input
B) To learn general language representations from large amounts of data
C) To fine-tune the model on a specific task
D) To evaluate the performance of the model on a test set
