FAQ: Generative AI

Loopio’s Generative AI functionality helps you generate high-quality responses with trusted library content. This document is meant to provide answers to common questions around specific Loopio features that leverage Generative AI, and our use of the technology in general.

What is Loopio’s approach to Generative AI?

A thoughtful, user-driven approach to AI means placing control in the hands of people like you. After hearing from our customers, there are three main factors that are critical to building a Generative AI solution that works across various organizations.

The three pillars of trust for Loopio’s approach to Generative AI are:

• Trustworthy sources: ensuring that the referenceable sources of generated information is within your team’s Library and not the internet, so you know the information is relevant and reliable
• Human ‘copilots’: our interface is intuitive so that SMEs and other teammates can easily review and improve generated results
• Customer choice: at both the account and individual level, customers have the option to be able to opt-in or out of AI

How can I use Generative AI to respond to questions?

Go to Project Workspace. Select the question you want Loopio to answer, then click the ‘Generate Answer’ button, and your entry will be generated for you! Loopio will craft a unique answer for your RFP, based entirely on the trusted information you already have in your content library.

The answers can then be modified as necessary and then captured through the Reviews from Projects workflow, to be picked up by Magic next time around.

What information does Generative AI use to generate a response?

Generative AI utilizes your library content to generate a relevant, tailored answer to a question. Loopio’s search engine is utilized to retrieve the most relevant content for generating an answer.

The Search engine considers signals such as entry popularity, recency, and keywords that were used to find entries and then passes up to 10 top entries along to Generative AI.

The Entries retrieved during a search are only taken from Stacks that the User has access to based on their Library (Stacks) permission level. This ensures that every user can only generate answers based on their permission level.

If the answer cannot be found in the Library, will Loopio’s implementation of Generative AI draw upon publicly available information to formulate an answer?

Loopio’s implementation of Generative AI will use your Loopio library content, in conjunction with OpenAI models, to generate the most accurate and relevant answer. Rather than leveraging online sources, the functionality will only search within your trusted existing library of content.

Loopio’s generative AI will either respond with generated text when the model is confident, or an indication that the AI cannot answer the question given the content in your library.

What is Loopio's recommendation for where Generative AI functionality should be applied within the overall response process?

• Run Magic with strong settings (e.g. “Strict” Magic Strength Setting)
• Click ‘Generate Answer’ on any question that didn't have a strong match
• Review and wordsmith the overall response for the final draft as usual

Does Generative AI learn from the user's input for questions in the future?

No, the feature does not "learn" directly from the generated or edited responses. The user must add an entry to their Library in order to enable Generative AI to reference the content.

This serves two purposes:

1. It allows the response to be used as a valid entry on Magic matches
2. It enables Generative AI to reference the added entry for future response generation

Given the current state of AI, human oversight is necessary to ensure that the user thoroughly vets generated entries before saving them to the Library.

Is there any data stored by the GPT provider (OpenAI) around what library entries were used for the generated answer?

Loopio customers' data is not sent to OpenAI unless they use the Generative AI feature by manually clicking the "Generate Answer" button available in the Project Workspace. This button is only visible when the feature is enabled.

Only a subset of the data (max of 10 library entries ) needed to answer a question is sent to OpenAI at any given time.

This data is retained by OpenAI for abuse and misuse monitoring purposes for a maximum of 30 days. After 30 days, data will be securely deleted from the OpenAI platform.

A limited number of authorized OpenAI employees, as well as specialized third-party contractors that are subject to confidentiality and security obligations, can access this retained data solely to investigate and verify suspected abuse.

How does Loopio’s Generative AI ensure accurate answers?

If Generative AI cannot generate a response to a question due to lack of relevant content in the library, it explicitly communicates that the answer cannot be generated, ensuring transparency in its capabilities.

Given the current state of AI, human oversight is still necessary to ensure that the user thoroughly vets generated entries before saving them to the Library.

How did Loopio test the Generative AI functionality?

The functionality underwent rigorous testing during an extensive Beta phase with 25 customers, with a focus on delivering a high level of customer satisfaction. This thorough testing process enabled Loopio to refine and fine-tune the Generative AI functionality, ensuring it met the needs and expectations of its users.

Can I limit Generative AI functionality to specific users/roles?

When configuring roles and permissions in Admin Settings, anyone who is an admin can grant or restrict access to generative AI for specific user roles. This ensures that this feature is used securely in your organization.

What Does Generated Answer Mean?

The Generated Answer indicator will appear throughout Loopio any time Generative AI was involved in creating an Answer, any time in the Entry's history.