Use internal compagny context to produce response


I am new to GPT-3 and find this model amazing. I would like to make a project to make an assistant in a production plant.

I have a word file that contains the instructions of a software. I would like to be able to give the text of this word file to GPT-3 to be able to ask questions later.

For example: “Starting the equipment”: “the vision equipment is started automatically when the capper is turned on via the main switch. After a few seconds, the vision program will display the main page.”

I would like to provide him with this information and then be able to ask him for example:
"how do I turn on the vision system? "

For now, I have tried fine-tune GPT-3 with the data as:

{prompt: chapter title, response: text of the chapter}
in my example it gives: {prompt: “Equipment startup”, response: "the vision equipment startup is done automatically when the capper is powered on via the master switch. After a few seconds, the vision program will display the main page.}

Unfortunately this approach gives me totally random results when I ask a question to the fine-tuned model.

Has anyone been able to experiment with a project like this or have any ideas on how to make it work?

Thanks for your help! :slight_smile:

Welcome to the forums!

You might find this thread helpful…

Finetuning is for Structure, not Knowledge

1 Like

Thanks for your quick response! I will continue on the path you gave me. I will come back to the topic when I have done more tests.