Print

Please login or register.
1 Hour 1 Day 1 Week 1 Month Forever
Login with username, password and session length

[Protools5LEGuy]

Clean Room reverse engeniering AI/Human leaded code for 7 on x86

https://www.youtube.com/watch?v=rJRlHKQqX2M

Interesting toolset.

[lartu]

It looked kinda buggy and barebones in the video, but wouldn't it be cool to install it onto a fat, gray early-2000s laptop if it was fully fleshed out!

[Jubadub]

Yeah, very old news by now, but the key takeaway regarding that project is that it is not Mac OS whatsoever. It is just AI slop at its purest in trying to create a new OS, so it is very much literally not Mac OS.

Official GitHub page for it: https://github.com/Kelsidavis/System7

I would like to see Mac OS reimplemented and reversed-engineered as much as the next Mac OS user does, but this project isn't it.

What annoys me is that it is most inappropriately titled "System7", leading to all this confusion. A real effort to put Mac OS on x86, on the other hand, as well as all other CPU architectures, would be interesting.

[laulandn]

It's...interesting, but just because I like the idea of anyone doing anything to get MacOS to run on x86 natively.  But you're completely right, this isn't MacOS at all. 

It's completely and utterly backwards how I personally would try to do it.  I'd start with the Universal Headers and actually USE them.  Instead of using ANYTHING resembling the structures Apple used, these are completely new and the result of asking AI questions.  Things are there, but not in the right places and very very weird.

They aren't even trying to keep the same names for the include files, so there's "SystemTypes.h" instead of "MacTypes.h".  This makes it pretty much impossible to compile any real Mac program to run on it...which is what I'd really want to at least try.

So, compared to other "rebuild an ancient OS from scratch" like ReactOS, Haiku, or AROS it doesn't even look compatible with real source for programs written for the original OS.  So it is extremely hard for anyone, the people making it or anyone else, to test that anything REAL actually works.  It's like a closed world.

But...still...an interesting artifact...I guess...

[Protools5LEGuy]

I wonder if the  1.1Gb modell trained by Alex Hoopes https://oldapplestuff.com/blog/MacinAI-Local/ expert  in MacOS era would spitt a better result

The Custom Model: 100M Parameters of Macintosh Knowledge

The custom model is a LlamaForCausalLM with 14-18 transformer layers, 512-640 hidden dimensions, and 8-10 attention heads. It was trained in three stages:

Stage 1: Continued pretraining on a 1.1GB corpus of Macintosh-specific text. The corpus includes Inside Macintosh (Apple's developer documentation), MacWorld and MacUser magazine archives, BYTE Magazine (Mac-filtered), vintage Usenet discussions, programming reference books, and Softalk magazine. 94,185 entries, cleaned to remove non-Mac contamination. Training ran locally on my 3080ti server.

Stage 2: Supervised fine-tuning on ~5,800 instruction pairs in ShareGPT format. This includes action classification (launching apps, opening control panels), AppleScript generation, Mac OS how-to procedures, OS-aware responses (different answers for System 7 vs OS 9), and hardware specifications for 500+ Macintosh models. Another few hours on the 3080ti.

Stage 3: Direct Preference Optimization (DPO) with 300-400 preference pairs. The model learns to prefer correct answers over known hallucinations, and clean responses over ones with token leaking (garbage bytes after the actual answer). 30-60 minutes.