Jupyter AI always replies when there are multiple humans present in a chat

[ellisonbg]

In testing the Jupyter AI 3.0.0a0 pre-release, I find that the AI always replies to all human messages, even when humans are trying to chat between themselves.

Description

When a single human user is chatting with an AI, it makes sense for the AI to reply to all messages. However, in multi-user chat settings that behavior quickly confusing and frustrating as it becomes impossible for humans to chat between themselves.

Reproduce

Open Jupyter Lab with Jupyter AI in two private browser windows, so you have two human identities. Try to send a chat message to the other human and Jupyter AI will always reply.

Expected behavior

• As a user, when I am chatting only with AI and no other humans, I want the AI to reply to all my messages.
• As a user, when I am chatting with both other humans and AI, I want a way to specify if the AI should reply or if my message is directed to humans.
• As a user, I want to configure Jupyter AI once, but then pick which chat documents jupyternaut participates in.

One way of handling this is for the AI to not respond without an explicit @ mention if there are multiple humans in the document. However there are some subtle cases related to who is in the document currently and who was there previously. Some interesting UX questions here.

[ellisonbg]

As we tackle this, I would love us to keep in mind that there can also be multiple AI personalities in a given chat. Given this, some sort of @ mentioning of AI personalities will probably be required.