Well, a few month ago, before my personal situation changed a bit to the worse for now, I tried to use various LLMs (6) to try and give me a little boost with a problem I ran into with a long stalled project of mine, which ultimately should be a contribution to FreeDOS, but with the option to "modernize" some parts of it and make it natively run on Windows for example. Or Linux, etc. A GW-BASIC compatible interpreter. Yes, I know PC-BASIC by Rob Hagemans exists, but that requires a huge Python foot print. The issue at hand is that this BASIC (like the ones before it) is using MBF (Microsoft Binary Format) floats, while everything else (including its successor, QBASIC) is using IEEE-756 floats. But all information, beside the basic bitwise representation, and some 3rd party source code, which I know existed at some point, has vanished from the Internet. So I asked those 6 LLMs to write me a library for all math functions, all performed in MBF (so explicitly no conversion to and back from IEEE-756!), and all in ANSI-C (instead of the ASM86 used in the original GW-BASIC sources). One of those LLM outright refused ("Sorry Ralf, I can't do that. ;) ), three others produced not even remotely usable garbage, with the last two seemingly, after some persuasion and repeated correction, seemed to come up with viable solutions, though only ChatGPT's final answers seems to be somewhat trustworthy. Due to changes in my personal situation, I have not been able to even test those results yet, but at the very least, I got some hints on some of the gotchas I ran into when I tried to recreate those maths routines myself (that was well before the GW-BASIC 1.x assembler source was even released), so I think I will get this eventually work, once my own situation has settled enough. But then I wonder how much of the resulting code would be indeed attributable to AI, if there is any questionable/lack of attribution that would prevent the resulting interpreter to be one day published and become part of FreeDOS (development tools).

Ralf




_______________________________________________
Freedos-devel mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/freedos-devel

Reply via email to