8080 ASM AI experiment

[roger arrick]

[scott baker]

I wrote a blog post up back when I was using GPT and Claude to do vintage computer programming: https://www.smbaker.com/can-an-ai-program-your-vintage-computer-for-you. In this particular post, I used it to generate Forth programs, but  at the time I also had it generating 8080 assembly and even 8008 assembly (which is kind of funny when an AI uses registers that don't exist in an 8008, or tries to push to the stack, but when I pointed this out, it apologized and corrected the code).

These days I use a personal Copilot account for all my vintage programming. It's actually a decent help with assembly. Even obscure assembly like 8008 or Z8000.

Scott

On Friday, 30 May 2025 at 13:23:57 UTC-7 roger arrick wrote:

[roger arrick]

---

From: intel-...@googlegroups.com <intel-...@googlegroups.com> on behalf of scott baker <smb...@gmail.com>
Sent: Friday, May 30, 2025 6:31 PM
To: intel-devsys <intel-...@googlegroups.com>
Subject: intel-devsys Re: 8080 ASM AI experiment

 

--
You received this message because you are subscribed to the Google Groups "intel-devsys" group.
To unsubscribe from this group and stop receiving emails from it, send an email to intel-devsys...@googlegroups.com.
To view this discussion visit https://groups.google.com/d/msgid/intel-devsys/f3951aa7-932c-4d1c-b7ed-98a7bf5c038dn%40googlegroups.com.

[Herbert Johnson]

[the code fragments are image files, which I won't retype just to make a
point.]

A more experienced programmer, would likely have asked for a routine
that is not passed a value in the A (accumulator) register, but in some
register useful as a temporary register. C or B are useful for that
purpose, as DE and HL are more useful as 16-bit registers and operands
on the 8080.

We know this stuff; but AI's 1) may not know and 2) they don't know what
they know and don't know. However, I may be wrong, or this idea may be
debatable. I do not pretend to be very knowledgable about AI's.

I also suspect, that a million people using AI's, and telling them what
their errors are, will make the AI's "better" by the metric of "whatever
people tell me is an error and is correct". I am reluctant to be an AI
teacher.

Regards Herb Johnson

--
Herb Johnson, New Jersey USA
http://www.retrotechnology.com or .net
preserve and restore 1970's personal computing
email: hjohnson @ retrotechnology dot com
or try later at herbjohnson @ comcast dot net