Hello guys!
First of all I wanted to clarify what language model and AI of such type are. These are just mathematical probabilities of words that will fit next after previous ones in given context. So the context must be given, there is no sense in using unprompted AI. In current project I had a goal to save users' RAM, so I did it like this: LM reloads (dumps it's own context and frees memory) every dialogue pair from player to NPC.
How AI knows about all the things around then if it does not have memory?
All the data is stored in the variables and prompted with each player message as a context to the AI (LM). So on the one hand every time it is like you write to it for the first time, but on the other it in fact "knows" what it should.
Is it connected to anything outside of the project?
No it is not, it is local solution, running on your computer, no internet used and no connections established. It is safest and most private of what is possible now.
So I explained a little, but maybe it is not enough, please write me, I will explain in a separate post how it works in the project in details.
But the question I wanted you to ask, if it is local model and it literally does not even "knows" anything as it's memory is dumped every time, so it is not a ChatGPT powered solution, it is different. How much do you hate AI in the project?