Hi,
First and foremost according to new forum community rules here is a warning:
!!! This project and thus this related topic is about an AI (more specifically ChatGPT) collaboration !!!

Preface:
I have never used AI Coding tools in my personal hobby projects so far since I like coding and in case of my projects there is no 'rush to market' that would justify such tools.
But I was curious how AI Coding would behave in situations where there are no known working solutions or best practices. We all know that AI Coding is very efficient when the problem domain is well known and the AI can use modern environments/frameworks/languages. There is no doubt that it can write the 1000th CRUD Web application in C#/Javascript without problems.
Also it can write legacy DOS based tools like FTP/File commanders etc. as can be seen here on Vogons.
But what happens if AI needs to write never existing legacy tools where really genuine new approaches are needed?

The idea:
My Thinkpad T430 always boots to DOS with full LCD brightness. Of course you can always press Fn+F8 multiple times to set the backlight level but e.g. Norton Commander interprets the Fn+F8 combo and changes the sort order of files.
So it would be more convenient to set the proper backlight level from CONFIG.SYS/AUTOEXEC.BAT. This is not a trivial task since on the T430 you cannot do this with the help of the known legacy interfaces and using ACPI from DOS is not without problems. But this is a very focused and well defined problem so I thought I gave a chance to AI to give me some solutions.

First I tried Google/Gemini. It was very confident at first and gave me solutions that either did not do anything or simply rebooted my T430. Finally Gemini declared that Lenovo disabled all possible routes to set the backlight level from DOS and the only possible solution is to alter VGA palette registers to fake lower brightness level (not the backlight).

Then I tried ChatGPT. The experience was similar at first but ChatGPT was more persistent and did not give up. And finally it could make a working solution 😀
It was rather funny that at the 1st 'confident' phase it even gave me compiled executable files but when I reported that the binaries rebooted my PC it switched to 'careful' mode and it was not willing to give me compiled binaries anymore 😀.
I did not use an AI enhanced IDE like Codex with ChatGPT but used the interactive console. It was very interesting experiencing how it reacts to failures or even to successful test results. The whole conversation can be read here:
https://chatgpt.com/share/6a636504-b994-83eb- … 65-4788ac1a851e

And finally here is the repository where the source files and the working binaries can be downloaded from:
https://github.com/Falcosoft/T430BLC