Can I change or correct responses?
Yes, admins can review and correct responses from the admin dashboard. When a correction is made, you have the option to send an email notification to the employee with the updated information. Learn more in Correct the Lattice AI Agent.
While this is useful for on-the-spot corrections, for more lasting improvements, we recommend updating the source documentation in your Knowledge Vault. This ensures the agent learns from the most accurate information for future queries. Please note that responses generated from specific user data cannot be corrected in this way, as they are tailored to the individual.
What if I have different policies for different locations? Is it possible to maintain multiple knowledge vaults so that I can ensure that employees belonging to certain groups only see content relevant to them?
Admins can add permissions for each document uploaded to the Knowledge Vault based on User Groups, so employees only have access to documents that are relevant to them. This would allow for location, function, and role-specific content.
How do I know the AI isn’t hallucinating?
We've built in several safeguards to ensure accuracy. For responses based on your company's documents, the agent provides citations that link directly to the source material, allowing users to verify information. However, like any generative AI, there is a possibility of generating incorrect information, or "hallucinations." We encourage users to use their judgment and cross-reference critical information, just as they would with any other AI tool.
Is there a limit to the number of documents that I can store in the Knowledge Vault?
There is no limit to the number of documents you can upload to the Knowledge Vault. Admins can bulk upload up to 10 files at a time. The maximum size for a file is 25 MB, and accepted file types include .txt, .md, .html, .doc, .docx, .csv, .xls, .xlsx, and .pdf.
Can we sync documents from knowledge management sources?
It is possible to sync documents from GDrive, SharePoint, Box, OneDrive and Dropbox. If you are an HRIS customer, you can sync HRIS documents to the Knowledge Vault.
Can I control the tone of the Agent?
Admins cannot set a global, permanent tone for the agent. By default, the agent responds in a capable and helpful assistant tone. However, individual users can influence the tone within their own conversations. For example, a user can prompt the agent to respond in a more casual, formal, or even bulleted-list format for the duration of their interaction.
Is my company’s data safe?
Lattice is SOC2 Type II Certified for Security, Confidentiality, and Availability, and commits to continue to responsibly collect, process, and protect customer data in accordance with their existing Data Processing Addendum. Lattice may use anonymized, aggregated data and feedback collected from customers to further develop and improve Lattice AI. (See § 9 of the AI Addendum.)
What data is being used to train Lattice AI?
Lattice AI is built on OpenAI, but OpenAI does not use our data for model training. Lattice interfaces with OpenAI via its API, Lattice has not and will not opt-in to allow OpenAI to use customer data to train or improve its models.
If someone interacts with the AI Agent in a specific language, will the AI Agent respond in the same language?
Yes. The AI Agent is designed to be multilingual and will respond in the language used in the query. This applies even if the language is not one of our officially supported languages on the Lattice platform.
Will the agent always produce the same response to a question?
No, like other generative AI technologies, the Lattice AI agent is non-deterministic. This means that if you ask the same question multiple times, you may receive slightly different wording or phrasing in the response. However, the core information and meaning should remain consistent, assuming the underlying source data has not changed.