Поділитися через


FAQ for Copilot

The answers to these frequently asked questions (FAQ) provide some insight into how the Copilot feature uses AI in Copilot Studio.

What is Copilot?

The system is designed to generate a single copilot topic from a human-written description.

What are Copilot's capabilities?

It combines the natural language understanding models already in Copilot Studio with Azure OpenAI to:

  • Understand what the copilot maker wants to achieve by parsing their request
  • Apply knowledge of how nodes within a topic work together, and how a topic should be constructed for the best effect
  • Generate a series of connected nodes that together form a full topic
  • Use plain language in any node that contains user-facing text that corresponds with the copilot maker's request

What is Copilot's intended use?

The Create with Copilot option in the Copilot Studio canvas lets copilot makers simply describe what they want to achieve, and then produces a topic path that achieves that goal.

How was Copilot evaluated? What metrics are used to measure performance?

The capability was evaluated on a collection of manually curated prompt-and-topic datasets, covering common, edge-case, offensive content, and synthetic generation.

During evaluation, topics generated from the capability were manually reviewed and scored for relevance to the input prompt, usefulness, and offensiveness.

What are the limitations of Copilot? How can users minimize the impact of limitations when using Copilot?

The system only supports English. Inaccurate responses might be returned when users converse with the system in languages other than English.

  • Your copilot must be created in the US region. Other regions, and languages other than English, aren't currently supported.

  • This capability might be subject to usage limits or capacity throttling.

  • Topics generated by the capability aren't always perfect, and might not accurately reflect the logic you wanted to implement.

    • We have implemented mitigations to filter out irrelevant and offensive language from appearing in the configured topic, and the system is designed not to respond when offensive language is detected.

    • We also monitor output and the feedback that copilot users provide to continually improve our content filters. These filters and mitigations aren't foolproof.

    Note

    You should always test and review your copilots before publishing them.

What operational factors and settings allow for effective and responsible use of Copilot?

To get the most out of Copilot, include granular instructions in your description and limit the scope of the description to a single topic. If the generated topic isn't what you had in mind or if you'd like to further workshop it, you can modify the topic, also using natural language.

We also recommend you carefully review generated topics for accuracy, either in the authoring canvas or code editor.

Feedback you provide regarding your level of satisfaction with generated topics is used to improve system quality. You can provide feedback by selecting the thumbs up or thumbs down icon for responses generated through Copilot. You can also include more feedback in free text.