Thanks for the great information at the 3.0 preview yesterday. I think the 3.0 calendar will be even better than our current version which is working quite well. Thanks. We have a couple of questions about the AI implementation in 3.0. Do you publish or disclose which LLM backend(s) you are using to generate the event titles and description suggestions? Also, I know our calendar instance is displaying publicly visible information. Still, we are wondering if you have a policy on whether the data submitted to the AI is used to train the LLM? You might have mentioned this in the presentation or previously. I apologize in advance if you did. These are questions we need to clear with our IT Risk team to make sure we can use the AI features in our instance.
I think this new page from our documentation answers most of your questions, but in brief: Yes, we’re using the gpt-40-mini API and our system prompt is shared (and configurable). We only send to OpenAI that system prompt + the user-provided description, and their policy states up top "We do not train our models on inputs and outputs through our API." Source: Key concepts | OpenAI API
Thanks Jon, I hear you! We had a Calendar-only preview this week (more to come in your inbox shortly about this) since we’re knowing that CMS sites have a bunch more to them and will have longer testing/rollout timelines. More from us on that very soon!