The Ford Barn - View Single Post - Artificial Information -the new How To Authority???

alexiskai · 07-09-2023, 02:08 AM

Quick update on my AI project now that Code Interpreter has been released to OpenAI Plus users. I had hoped that the Code Intepreter function would be able to parse PDFs of books and extract text, but so far I haven't been able to get it to do that. However, GPT-4 is able to write Python scripts for me that will do that on my workstation.

So my first task has been to convert images of The Restorer magazine to text files. As you may know, MAFCA sells the entire back catalog of The Restorer from 1956-2006 on a flash drive, but it's all images – not searchable. I was able to get the AI to write a script that converts each issue of the magazine from images to a single text file. Of course, all of the photos are stripped out, but for the moment I'm just trying to create a corpus that will be suitable to train the AI.

My next phase will be to work through the service bulletins, which I also have as PDF already. Then on to other books, like the JS, the Les Andrews blue book, and the McRee engine rebuilding guide. These I'll have to scan myself.

Once I finish those, I'll start figuring out how to process what I have into a database to train the AI on.

All of this content will live on my workstation, so not even a shadow of copyright issues yet. We'll cross that bridge when we get there.

07-09-2023, 02:08 AM	#48
alexiskai Senior Member Join Date: Apr 2018 Location: Mebane NC Posts: 2,377	Re: Artificial Information -the new How To Authority??? Quick update on my AI project now that Code Interpreter has been released to OpenAI Plus users. I had hoped that the Code Intepreter function would be able to parse PDFs of books and extract text, but so far I haven't been able to get it to do that. However, GPT-4 is able to write Python scripts for me that will do that on my workstation. So my first task has been to convert images of The Restorer magazine to text files. As you may know, MAFCA sells the entire back catalog of The Restorer from 1956-2006 on a flash drive, but it's all images – not searchable. I was able to get the AI to write a script that converts each issue of the magazine from images to a single text file. Of course, all of the photos are stripped out, but for the moment I'm just trying to create a corpus that will be suitable to train the AI. My next phase will be to work through the service bulletins, which I also have as PDF already. Then on to other books, like the JS, the Les Andrews blue book, and the McRee engine rebuilding guide. These I'll have to scan myself. Once I finish those, I'll start figuring out how to process what I have into a database to train the AI on. All of this content will live on my workstation, so not even a shadow of copyright issues yet. We'll cross that bridge when we get there.