[COFF] Perl (was Re: forth on early unix)

Tony Patti via COFF coff at tuhs.org
Tue Sep 23 12:38:36 AEST 2025


> From: John Levine via COFF <coff at tuhs.org> on September 22, 2025 at 9:59 PM wrote:
>
> Something smells wrong here.  Anyone doing numerical work in python uses the numpy and scipy libraries which represent arrays in native binary form and have efficient versions of numerical operations.
> This suggests they did a mechanical translation into simple minded python using ineffiecient data structures rather than the way an actual python programmer would do it.

Hi John,

NumPy's core library does not natively support arithmetic over Galois fields.
As the title of my paper should suggest 
"An interesting example at the intersection of Matrix Mathematics and Cryptography (and how Artificial Intelligence can write programs)",
the goal was to see how well A.I. could generate cryptographic software, in multiple programming languages, 
and in what most people would consider to be atypical (uncommon) Galois Fields such as GF(997727) and GF(1077777719) and even polynomials over GF(9973^12).
The latter, by definition, are subject to both a prime modulus and a monic irreducible polynomial modulus. 
Additionally, the software handles matrices with sizes ranging over six orders of magnitude, from 4×4 to 4000×4000. 
So, when you say "they" in your sentence, you are referring to the two A.I.'s cited in my paper: Google Gemini and Microsoft Copilot.
I would agree with your use of the word "mechanical" since the software in the paper was A.I. machine-generated.
Full source code is included in my paper, for those who are interested.

Tony Patti
(ARPAnet NIC IDENT "TP4")





More information about the COFF mailing list