In my previous devlog I told about having a problem with making my AI understand Natural Language.

I have combined my possible paths and created a 4th path.

Now, I will take a open-source (MIT Licensed ) LLM (name is Phi-2, developed my Microsoft), Then train it to be useful to my AI. Then I will sync/ combine my trained LLM with my AI. By doing this my AI can perform tasks on naturally said command and It will be easy for the user to use.

Today while working on Was-Assistant.
I realised that my AI will not able to understand naturally said commands with my pre-trained commands.
that’s why I am thinking to take any one of be below paths:

Path-1:

I will add my AI with a open-source LLM and then develop it.

Path-2:

I will train it to catch key words. Like it will be trained
 to turn on Light if it get any command with a line
 consist of those two words ("light" and "on").

Path-3:

Make to to work exactly on it's given exact command. 

I came up with the idea in path-2 while thinking about marks schemes. I may take the path-2, because I like it most. My may also take path-1. For path-3, I dislike it.

I haven’t decided yet. But it is more likely I will take path-2 or mixed.

I want to create an AI, which can think itself (in its specific boundaries), work as my schedule manager, manages my PC better than be up to root level. And an AI where data never leaves from PC.

Was-Assistant.

For a while, I am working on a personal assistant which directly works form PC without any external dependencies.

My aim on was-assistant is:

An personal AI which can think itself, use the 
resource available on PC and help in real time.
 An AI which can think itself and work as a
 personalised assistant. 
Like setting up alarm, managing schedule, and do
 task automatically. And also adapting with the flow. 
[Any more thing can add up as time goes].

Currently it is like a structure and running on pre-trained commands.