Scenario split:
I will provide you with a sentence and I will also provide you with four categories. You should categorize this sentence
into one of the four categories based on the content of this sentence. Note that you only need to output the result of
the categorization, i.e. the name of the category to which this sentence belongs. There is no need to generate
additional content
categories:
sentence:
Please output the category that best fits this sentence based on the sentences and categories above. Note that your
result can only be one of these categories without generating additional content.

Agent Decision:
Suppose you are a data creator and you are now required to accomplish the following task: I will
provide you with a question and a corresponding answer, and you need to create three new incorrect
answers about this question based on these two parts. Note that the answers you generate should
be similar in format to the correct answer and have some similarity and confusion in content, but
you must ensure that the generated answers are the wrong answers to the question I provided
you with.
Here are the question and the answer:
Question:
Answer:
Note that you should only generate the answers using the question and corresponding answer above.
And the answers you generate should conform to the following format: [answer1, answer2, answer3]
No additional content should be generated

Agent Plan:
Suppose you are a data creator, now you are asked to complete the following tasks: first you need to generate a general task goal,
and second you need to decompose this goal into a sequence of sequential sub-tasks that match the logical relationships.
For each sub-task, in addition to generating its text, you also need to generate the caption of the image of the sub-task,
and I'll provide you with some samples for your reference.
Note: The overall task goal needs to be similar in content to the example I have given, and the subtasks need to fulfill
this requirement as well. The generated format should match the following:
Goal:
Caption1:
Task1:
Caption2:
Task2:
......
Where Goal is the overall task goal. task1 is the first subtask of the decomposition, and so on. caption1 is the caption
of the image of task1, and so on.
In addition to this, you need to be aware of the diversity of the overall task objective, which should cover as much
scope as possible. The sequence of subtasks that the overall task goal is broken down into should also be logical and
make sense. The captions for the images of the subtasks should closely match the content of the subtasks and be detailed
and logical.
Here is an example:
Goal:
Task1:
Task2:
...
Taskn:
Note that you should only generate the content conforming the formate requirement. No additional content should be generated


Original data conversion:
Given a verb and a noun, you should generate a sentence from the two words. In the sentence, the subject should be in
the third person, and the other part should follow the examples that I will provide in the following part.
Here is some examples:
Noun:
Verb:
Sentence:
You should learn from the examples above and generate a sentence from the following noun and verb
Noun:{}
Verb:{}
Note that you should only generate a sentence using the noun and verb above. No additonal
content should be generated

AVQA conversion:
Given the task question and answer, you should generate a sentence from the question and class labels, which can correctly
express what is in the question and answer I will also provide you with 8 examples with the following format:
question:\n\
answer:\n\
sentence:\n'
Here is some examples:
Example:
question: {}
answer: {}
sentence: {}
you should learn from the above examples and generate a sentence corresponds to the following question and answer:
question: {}
answer: {}
Note that you should only give a sentence correspond to the question and answer, no additional content should be given