A user asks about data usage and training practices for Discourse AI, particularly with Google Gemini LLM. Falco explains that Discourse AI is flexible, allowing users to choose their AI provider, such as AWS or Google Gemini, and that data sent to the LLM provider depends on the chosen service. Falco notes that sensitive data may be sent if included in posts, but all interactions are logged for auditing. For stricter data control, users can self-host an LLM. The user’s specific questions about Google Gemini’s data handling and training practices are directed to Google, as they are outside Discourse AI’s control.
I want to use Discourse AI. I have some questions regarding this.
What data is being utilized by Discourse AI (and any LLM like OpenAI or Google LLM)? Does it include any sensitive data?
“Discourse AI”, the plugin, is agnostic of the AI model provider. For every feature, you can pick the provider that aligns best with your needs, whether it be privacy, feature set, pricing, etc.
If you are using, let’s say, AWS for the AI provider, then you are subject to AWS Bedrock terms of service regarding data being used for training, etc.
We have dozens of features in Discourse AI, and they will send data to your preferred LLM provider. For example, the summarize feature will send the topic posts and ask for a summary. If your posts contain sensitive data, that will be sent.
We log everything that is sent to the database, so you can audit it anytime.
And if you have stricter data locality requirements, you can run an LLM on your own servers to keep data under your control.