Applies To:
- Users
- Pinnacle Admins
Common Causes/Issues:
AI Chat empowers you to solve problems and access resources efficiently.
Solution Overview:
The following represents questions and answers specifically pertaining to the AI Chat feature set in Pinnacle Series. The answers do not reflect other functionality.
Technical Specifications
| Question | Eagle Point Responses |
|---|---|
| Was this developed in-house? | The Pinnacle Series chat widget was developed in-house but leverages the Microsoft Azures OpenAI Service models. |
| Was this built on an AI framework? (i.e., ChatGPT, Copilot) | It was built using the Azure OpenAI framework, using the GPT 4 model available within the service. |
| Where will it be hosted (consider all its elements)? | Same hosting location as the overall Pinnacle Series solution. |
| Amount of computation used for training it? | The model was pre-trained by OpenAI and provided through Microsoft Azure. Eagle Point did not conduct or directly control the training process. Azure OpenAI services manage and optimize all computations necessary for inference rather than training, as the model is already trained. |
Data Handling
| Question | Eagle Point Responses |
|---|---|
| What types of input data (technical information, personal information, synthetic, open-source, commercially sensitive, etc.) do you expect to be processed by it? Are user entries the only way to input? | The input data primarily consists of user queries, which may contain general technical information or operational inquiries. The system is not expected to process sensitive personal or commercial information. User entries are the primary mode of input. |
| Will outputs be combined/merged with other data sets? | Outputs will be blended with Pinnacle Series content via our standard search index. |
| Will the solution retain input data and outputs? For how long? | Currently, it does not retain input or output data. However, as we continue to enhance the use of AI, taking advantage of conversions to train responses or possibly build content continually will be considered and could change the use of retaining data. |
| Who will have access to it? | Once turned on within the Admin Portal > Settings > General options, the chat feature is available to all tenants' users. |
| If it is provided by a third party (as opposed to entirely developed in-house), will users' inputs and outputs be used to train the model? | No, Microsoft Azure OpenAI does not use customer data to train or improve the model. All input and output data will remain isolated from Microsoft and OpenAI and will not be used in further model training. |
AI Chat Data Privacy and Compliance
This section provides reassurance about how AI Chat handles data securely and responsibly.
| Key Aspect | Details |
|---|---|
| Data Processing | User queries are the only input processed; the system does not handle sensitive personal or commercially sensitive information. |
| Data Retention | Input and output data are not retained unless explicitly enabled in future enhancements. |
| Compliance Standards | Compliant with: CCPA, GDPR, SOC 2, SOC 3, & CSA Star. |
Transparency, Legal, and Compliance
| Question | Eagle Point Responses |
|---|---|
| Does it comply with any AI Industry Standards and Regulations, and which ones? |
Compliant with: CCPA, CSA Star, GDPR, SOC 2, & SOC 3. References: |
AI Chat Resource Isolation
| Question | Eagle Point Responses |
|---|---|
| Does Pinnacle Series AI Chat share resources or data with other customers or the Eagle Point vendor? |
Pinnacle Series AI Chat does not share resources or data with other customers or the Eagle Point vendor. |
| How does AI Chat handle model deployment? |
AI Chat operates on a single deployment of OpenAI’s GPT 4 model, meaning all customers use the same model instance. |
| Is user input stored, used for retraining, or repurposed in any way? |
No user input is stored, used for retraining, or repurposed in any way. |
| What technology does AI Chat use to ensure relevant responses? |
AI Chat utilizes Retrieval Augmented Generation (RAG) to pull relevant information only from content that a user has explicit access to. |
| Can AI Chat generate responses based on another customer’s content? |
The system never generates responses based on content from another customer’s tenant. |
| How does AI Chat ensure access to information remains isolated? |
While all users share the same model deployment, access to information is strictly isolated, ensuring responses are generated only from content that the user is authorized to view. |