Fortran meets AI

Story time

My father studied at Faculty of Electrical Engineering at SS Cyril and Methodius University in Skopje in the late 70s.
I studied at the same faculty and university 30-something years later and got my Master's degree at Faculty of Computer Science and Engineering.
While back, during my studies I found one textbook from the courses he took, Fortran IV. Here's the photo of it:

Some of you reading this might know what Fortran is (programming language), but I presume almost none of you seen problems solved in the language or punch cards. Yes, punch cards. Back in the days, people used to code on punch cards and that was included in this course. Punch cards meant that every line of code in a program had to have its own card. So if your program had 1000 LOC, then you had 1000 cards. How that looked, more in this video:

Most interesting thing: there was an exam inside it, his exam from almost 50 years ago! Here it is (fyi, the language is Macedonian):

As you can see they weren't solving important problems back in the day like, recommendation algorithm or even more important and harder to grasp, CSS. It's mostly useless math.
Even though I've known about this exam for quite a while and wanted to solve it, never got to it. Until now.

Today

With everything that's happening in the field of computer science and AI, writing code and solving problems is the easy part. But an exam 50 years ago in Fortran? Would that be a problem for today's best models? You've guessed it right - it won't.

I've made a small showcase GitHub Page with the solutions to these 5 problems, called Fortran meets AI you can visit here.

The same problem is solved by Claude Code (Opus 4.6), Codex 5.4 (by OpenAI) and MinMax 2.5 from China.

Results

• Codex had no problem in reading, recognizing and solving the issues. The model was provided with the images, translated it to English and created two .svgs. It produced solutions in Fortran, Python and Kotlin.
• Cladue Code had no issue to resolve this, provided the solutions for the problems in Fortran, Python and Kotlin.
• MinMax 2.5 the free version provided by OpenCode's OpenZen (which I've used) cannot read images, so I used the .svgs that Codex created. Had no issues
• I've added Python compiler (it was the easiest) so solutions can be compared in terms of compile speed and simplicity. And with that you can clearly see that Codex gave "uncopilable" solution on the second problem (SyntaxError: '(' was never closed). The only model that did this, which, to be honest, I didn't expect.
• Even more surprising, Claude Code got the first task wrong, MiniMax did it most optimal, made mistake in Problem 2.

In details:

Last, not least, Claude code was the only one to include the input card layout in the solution, for example:

Words of wisdom

If you got to here by reading, you deserve a smart sounding sign off to this article. From this stand point programming and software engineering in the late 70s looked pretty bleak and like nothing that can guide our lives in such an important manner as today. It took new technologies and discoveries to place it as one of the guiding forces of current and future progress. To that note, maybe today's state of SWE will look as bleak and simplistic as Fortran in the 70s. Matter of fact, I'm sure of that and that day is coming rather sooner than later.