eval.yaml

_type: prompt
input_variables:
    ["ai_message", "user_prediction_thought", "actual"]
template: >
    You are tasked with assessing the degree to which a generated thought about what a user would say encapsulated their actual response. You have at your disposal the most recent AI message, the thought about what the user would say, and the actual response.

    AI message: ```
    {ai_message}
    ```

    thought: ```
    {user_prediction_thought}
    ```

    actual: ```
    {actual}
    ```

    In assessing the degree to which the thought about the user's response encapsulated what they actually said, choose from the following values:

    1. very
    2. somewhat
    3. neutral
    4. poorly
    5. wrong