Usage of "assistant" role

[inner token]

Hello, just learning how to use gpt-3.5-turbo in a Discord bot coded on Node.JS (using https://www.npmjs.com/package/openai package), looking through the documentation (https://platform.openai.com/docs/guides/chat/introduction) it says there is three "roles" to be used on ChatCompletion, i kind get what user and system does, but what about assistant? It says it can be use to "help store prior responses". Can i use that as a rudimentary "session" manager?

For example, in my project i could use Discord's channel id as a unique identifier to make all conversations to stay in the same topic in a conversation in certain channel?

Thanks a lot for all help

[glass rover]

if i get what you mean;

user -> us, it is our input
system -> context
assistant -> bot response

[inner token]

i see

[warm yarrow]

I'm just appending chatgpt answers as 'assistant' to the conversation, and it seems to be working.

[inner token]

the previous ones?

[warm yarrow]

you can test by asking the AI to clarify something that it responded with

[inner token]

will try to do that in my bot project

thanks a lot!

[warm yarrow]

yea so every time I create a new prompt it gets larger and larger, usually alternating 'user','assistant' over and over

[inner token]

interesting

[warm yarrow]

I don't think the server keeps track of the conversation, it re-reads everything every time you send a prompt.

[marble ice]

correct, assistant is chatgpt, user is you and system is context

for memory you have to append every new message to a message list