Skip to content
The Chat Archive
gpt-3.5-turbo-0613· OpenAI Arena Human Preference 55K

You are good at image classification. There is a sentence: A person doing Apply Eye…

  1. User

    You are good at image classification. There is a sentence: A person doing Apply Eye Makeup might look like they are holding a makeup brush and applying makeup to their eyelids or under-eye area, with a focused expression. Please reconstruct the above sentence into 4 parts and output a JSON object.
    1. All entities,
    2. All attributes,
    3. Relationships between entity and entity,
    4. Relationships between attributes and entities.
    The target JSON object contains the following keys:
    Entities,
    Attributes,
    Entity-to-Entity Relationships,
    Entity-to-Attribute Relationships.
    For the key Entity-to-Entity Relationships, the value is a list of JSON object that contains the following keys: entity1, relationship, entity2.
    For the key Entity-to-Attribute Relationships, the value is a list of JSON object that contains the following keys: entity, relationship, attribute.
    Do not output anything other than the JSON object.

  2. Assistant

    { "Entities": ["person", "Apply Eye Makeup", "makeup brush", "eyelids", "under-eye area", "expression"], "Attributes": ["holding", "applying", "focused"], "Entity-to-Entity Relationships": [ {"entity1": "person", "relationship": "doing", "entity2": "Apply Eye Makeup"}, {"entity1": "person", "relationship": "holding", "entity2": "makeup brush"}, {"entity1": "Apply Eye Makeup", "relationship": "applying", "entity2": "makeup to eyelids"}, {"entity1": "Apply Eye Makeup", "relationship": "applying", "entity2": "makeup to under-eye area"}, {"entity1": "person", "relationship": "have", "entity2": "focused expression"} ], "Entity-to-Attribute Relationships": [ {"entity": "person", "relationship": "look like", "attribute": "holding makeup brush"}, {"entity": "person", "relationship": "look like", "attribute": "applying makeup to eyelids"}, {"entity": "person", "relationship": "look like", "attribute": "applying makeup to under-eye area"}, {"entity": "person", "relationship": "look like", "attribute": "having focused expression"} ] }

Responses were produced in different, uncontrolled contexts (system prompts, settings, prior turns). This is not a benchmark.