#Is there a way to specify in the LLM conversation instructions which user is interacting with it?

1 messages · Page 1 of 1 (latest)

exotic horizon
#

This can help avoid needing to specify the room every time when using the prompt "Turn of my light", as it would assume "my" means the light in my bedroom associated with my user.

river warren
exotic horizon
#

So we already include that metadata to the context of conversation?

#

I don't have room-based zones yet with something like bermuda, but i would still like it to know what i mean by "my bedroom"

#

My user is called John Doe (for example)

#

and the zone of the room is called "John" or "John Bedroom"

#

shouldn't it be able to infer that?

river warren
#

you dont need to know where YOU are, just that sattelite your speaking to is in the bedroom. so if anyone speaks to that device in the bedroom and asks to "turn off the light" they probably mean the light in the bedroom

exotic horizon
#

I see, I am using the Assist on my phone or in the browser on my desktop, so not sure that would work?

river warren
#

oh right

#

you could add it to the prompt of the llm that any time you refer to "my light" it means light.XYZ
then if you need the same thing for multiple users you could have seperate llm connections and seperate assist pipelines that each could use

#

i dont think the username is not provided in context to the llm

exotic horizon
#

Damn, that really sucks ngl

#

The instructions aren't per-pipeline though

#

they're in the actual conversation entity from the integration of the Google AI thing

#

It can be a template, but I don't know what is available in that context that I can reference in the template

exotic horizon
#

and then in some way pick a pipeline dynamically for the user?

#

But I think that feature doesn't exist

river warren
#

for the most part its designed around voice satellites which wont have a user attached to them