Prepare your dataset
Task Builder accepts datasets as CSV or JSON files, and you can also paste JSON directly. This article shows you how to format a CSV file. Each row becomes one task, and each column becomes a field.
Basic structure
Start with a header row at the top of your file to name your columns. Task Builder uses these names as field names, so clear headers make your data easier to organize.
Important guidelines
Create your file with any number of rows, as there’s no limit
Add all the columns you need to organize your data
Use clear, descriptive headers, which become field names in Task Builder
Keep your file clean by removing any instructions or notes
Make sure your file stays under 5GB
You don’t need to add instructions to your file. After you upload your dataset, you add a task name, introduction, and questions in Task Builder.
Example CSV:
Example in Task Builder:
Example datasets by use case
These examples show how to structure your file for common types of work. Use them as a starting point and change the columns to suit your task.
Reinforcement learning from human feedback (RLHF)
Participants compare AI responses to show which ones they prefer. Add a prompt column and one column for each response you want them to compare.
CSV example:
Example in Task Builder:
Sentiment analysis
Participants read a piece of text and label the feeling it expresses. Add one column containing the text you want them to label.
CSV example:
Example in Task Builder:
Evaluation
Participants judge the quality, accuracy, or appropriateness of AI outputs. Add a prompt column and a column containing the output you want them to judge.
CSV example:
Example in Task Builder:








