As part of our efforts to improve the personalization of our AI system, I came across the “name” property ({“role”: “user” / “assistant”, “name”: “John”, “content”: “…”}) that can be added to each message. In our use case, we could add the user’s name and the AI chatbot’s name.
My question is: Does anyone have any experience (positive or negative) with this feature? Are there any concerns about using it? From my understanding of how large language models are trained, it seems they didn’t use the original names of participants in their training data. They likely used designations like User/Assistant. Therefore, maybe it’s better to maintain that approach.
The AI takes to and understands the name field. Here’s a chatbot “address the user by name” that the first introduction response is to a name “instructions”, and then the prompt has a different name field.
I didn’t expect it to be asking “what’s your name” off the bat, but the language the AI receives, user:David from the extra parameter, takes care of it knowing.
Name is not as versatile as it could be, because you cannot prompt the AI to have a name when it responds to you, which could switch roles or fine-tuning. It also cannot have spaces.
You can see also that user/assistant has semantic meaning being picked up on that limits the AI possibilities also.
