Remember a few months ago the American stock markets had a hiccup when a Chinese company brought out an AI called DeepSeek. I'm just looking at YouTube videos here and it turns out you can use it offline if you download it. It's about 5 or 6 gigs to download.

There are about 6,000 papers, so that will be maybe a gigabyte or two of papers, giving a total of about seven or eight gigs for the AI and all the papers.

So I'll write a program using wxWidgets that will build and run on MS-Windows, macOS and Linux, and you'll ask it a boolean question that will be applied to every paper, such as: "Does this paper mention adding a new CV-qualifier to the language?". So then the program will iterate through the papers one by one and ask the AI the same question for each paper, and then you'll get a list of what papers answered true. All the hard work will be done by your own CPU and RAM, and you won't need an internet connection.

If you have 32gigs of RAM then you can copy the program and papers into RAM before running it so that it's a lot faster (I used to do this in the 1990's with MS-DOS using RAMDISK.SYS to play Quake and Doom).

I haven't downloaded the AI yet but if it's a pre-built x86_64 executable then I'll see about integrating Qemu into the program so that you can run it on arm64 desktop PC's (e.g. Apple Silicon).