Zero Shot

This method consists of a two part prompt, which follows the template:

<instruction text>
<example>

where the instruction text defines the task to perform and the example is input for the task.

Example 1:

Prompt:

Classify into negative or positive sentiment the input text:
Blue sky makes us feel so high

Response:

The input text "Blue sky makes us feel so high" can be classified as having a positive sentiment. The use of words like "blue sky" and "feel so high" conveys a sense of joy and euphoria, which typically indicates a positive emotion.

But this is not what we would have expected. An analysis has been provided without being warranted. Instead a simple positive/negative answer would have been much closer to our expectation. Framing our instruction appropriately can have a better effect and give the desired ouput.

Examples 2:

Prompt:

Respond with only the word negative or positive for the sentiment in the input text:
Blue sky makes us feel so high

Response:

Positive

Impersonation

Instead of further specifyin the instruction it could have been possible to achieve the same result by instructing the model to act with a differntly defined personality. These types of zero-shot prompt follow the template:

<persona>
<instruction text>
<example>

A persona is a instruction that affects the output in general, whereas the instruction remains specific to the text. It is true that in free-form the persona and instruction functionalities may overlap.

As an AI-system, you respond only with positive and negative based on the classification of the text: Blue sky makes us feel so high

Positive