Create AI-enhanced input tables (Beta)

🚩

This feature is currently in open beta and subject to quick, iterative changes. As a result, the latest product version may differ from the contents of this document.

Input tables support AI-generated columns that allow you to augment your data and derive valuable insights for improved data-driven analysis and decision-making.

Sigma helps you accurately convey your data requirements through structured AI prompts that create the following types of AI-generated columns:

  • Classification: Assigns existing column data to distinct categories or groups.
  • Sentiment analysis (opinion mining): Determines emotion and tone expressed by textual data.
  • Column fill (data synthesis): Generates data based on existing and provided context.

This document introduces the concepts of AI-generated columns and explains how to apply them to your input tables.

System and user requirements

The ability to add AI-generated columns to input tables requires the following:

About AI-generated columns

Sigma leverages OpenAI language models to enhance input tables with AI-generated columns. The AI model analyzes existing data and contextual information, then predicts new data based on identified patterns, trends, and associations.

Input tables feature AI prompt templates that employ AI data classification, sentiment analysis, and column fill. The following sections describe the AI capabilities and various ways you can use them. For information about applying AI-generated columns to input tables, see Add an AI-generated column in this document.

Classification

AI classification can assign existing column data to distinct categories or groups. This information can improve data organization and extraction for more streamlined or targeted analysis.

There are many ways to classify data with AI, including (but not limited to) the following methods:

  • User-defined categories: Provide a set of category labels that prompt the AI model to assign data to one or more labels in the set.
  • AI-generated categories: Enter a generalized term or phrase (for example, “category,” “type,” or “group”) that prompts the AI model to assign data to auto-generated labels.
  • Binary classification: Enter a single category label that prompts the AI model to assign “yes” or “no” based on how the data fits the provided label.

Sentiment analysis

AI sentiment analysis (or opinion mining) can determine emotion and tone expressed by textual data. This insight can help you understand the personal interpretations and subjective attitudes towards a specific topic, such as a product or service.

There are many ways to analyze sentiment with AI, including (but not limited to) the following methods:

  • User-defined sentiments: Provide a set of sentiment labels that prompt the AI model to assign data to one of the labels in the set.
  • Standard sentiments: Enter a generalized term or phrase (for example, “sentiment,” “tone,” or “opinion”) that prompts the AI model to classify data as positive, negative, or neutral.

Column fill

AI column fill (or data synthesis) generates data based on the context of existing data and the freeform AI prompt. Sigma enables you to provide unrestricted input that can accommodate a wide range of queries and tasks. You can also use column fill to create more detailed classification or sentiment analysis instructions when your request requires more dimensions and flexibility.

Add an AI-generated column

Add an AI-generated column to a new or existing input table to create or augment a dataset. You can edit, analyze, and visualize AI-generated data just as you would with queried or manually entered data.

  1. Open a workbook in Edit mode.

  2. Create a new empty or linked input table, or locate an existing one to supplement with AI-generated columns.

    💡

    If the existing input table isn’t editable in Edit mode (indicated by the Data Entry tag), set the data entry permission to Only in Draft version, then add the AI-generated columns.

  3. In the input table, click the caret () in an existing column header to open the column menu.

  4. Hover over Add column via AI, then select an option to open the applicable AI prompt.

    • Classify column: Assign an existing column’s values to specific categories.
    • Sentiment analysis: Analyze the emotion or tone expressed by an existing column’s values.
    • Fill column: Generate column values based on the existing dataset and user-provided context.

  5. Complete the applicable prompt (click to expand below):

    Classify data via AI
    • Column to classify: Select an existing column for data classification.

    • Column description: Describe the contents or purpose of the selected column. Provide clear, specific information to help the AI model correctly analyze the data and understand your intent.

    • Categories: Add one or more labels or terms for the data classification. Press the Enter key after each entry.

    Sentiment analysis via AI
    • Column to analyze: Select an existing column for sentiment analysis.

    • Column description: Describe the contents or purpose of the selected column. Provide clear, specific information to help the AI model correctly analyze the data and understand your intent.

    • Sentiment: Add one or more labels or terms for the sentiment analysis. Press the Enter key after each entry.

    Fill column via AI
    • Data context: Describe the data you want to generate. Provide clear, specific information to help the AI model correctly analyze the data and understand your intent. You can extract data, reference existing columns, specify the desired output format, and more. You can also prompt the AI model to provide new data unrelated to existing columns in the input table.

    💡

    For more information about optimizing your prompts for better AI-generated results, see OpenAI’s GPT best practices.

  6. In the prompt modal, click Generate sample to preview up to 10 rows of AI-generated data.

  7. If the sample data meets your requirements, click Apply to column. Otherwise, repeat the previous steps to modify the prompt and regenerate the sample.

    When you click Apply to column, the modal displays the AI model’s progress and closes when data generation is complete. The new column is added to the input table with up to 500 rows of AI-generated data.

  8. Review the AI-generated column and edit the data as needed.

    💡

    If you changed the input table’s data entry permission in step 2, revert the permission to Only on Published version when you’ve finished adding AI-generated columns.

Frequently asked questions

What large language models (LLMs) are utilized by AI-enhanced input tables?

Sigma integrates with OpenAI to leverage OpenAI language models, including GPT-3.5 and GPT-4 models that power ChatGPT.

Can I add AI-generated columns to table elements that represent data from my warehouse?

Sigma supports AI-generated columns in input table elements only.

Sigma’s table elements represent database tables from your cloud data warehouse (CDW) or database management system (DBMS). Due to their potentially extensive size, LLMs cannot optimally handle interactions with these tables.

You can, however, augment CDW or DBMS data with AI-generated columns by creating linked input tables that use existing table elements as data sources. The linked input table will support up to 500 rows of AI-generated data.

How can I avoid false information and ensure accurate AI-generated data?

AI hallucinations (false information that deviates from facts, logic, or reality) are inherent to large language models, including all OpenAI models leveraged by Sigma. While you may not be able to eliminate hallucinations, there are steps you can take to prevent them and reduce their influence on input tables.

  1. Use simple, direct language in prompts to help the AI model correctly interpret your intent.
  2. If the sample data doesn’t meet your expectations or contains inaccurate information, refine and test your prompt until the AI model generates the desired data. To refine your prompt, try changing your phrasing, replacing vague language with specific details, adding more context, or providing examples that demonstrate the type of output and format you prefer.
  3. Review AI-generated columns to verify and validate all output before applying the data to analyses and reports. Correct false information and modify input table values as needed to ensure an accurate dataset.

For information about AI security and data handling, refer to Frequently asked questions in Manage OpenAI integration.