I’ve been working my way through the issues of the People’s Computer Company and typing in their programs. Many of these are in the HP dialect, and it serves as a great source for real-world HP programs. There are also a number from Mac Oglesby in the original Dartmouth format, including Motie and Rescue. There’s a couple in 2200 dialect, which is weird, and some odds and sods.
I’ve just finished the January 1976 edition, which had the original Star Trek and Mac’s code, so while there’s still a couple more issues to go I am going to link to it now:
The later issues, starting around v3n2, were converted using the aid of ChatGTP. If you tell it the image is of BASIC code it really narrows down the grammar and does an amazing job of figuring out that this is PRINT not PR1NT, and it’s 2280 not 228o. However, it also will confuse a font that has a bar in the zero with an 8, so it gets line numbers wrong… and then it knows that line numbers have to be increasing, so it goes ahead and renumbers the rest of your code!
I’ve tried my best to check things and get the numbers and spacing correct, but I expect there are errors and would love to get your pull requests! I’ll try to have the rest uploaded over the next week.
Update: I’ve been experimenting with alternatives, and Gemini AI trivially outperforms ChatGTP at this task. It did a 100% perfect conversion on code from Rescue, the same code where CGTP renumbered everything even after telling it not to and having it hand back the exact same code. It also did it much more quickly, perhaps half the time.
Definitely worth looking at, I’m going to try it with the really bad printing from What to Do and see how it goes.
Interesting. I’m still struggling with good and bad ways to make use of the new LLMs - and of course they keep evolving. Anything where you have a way to check the correctness of the output is firmer ground.
OCR of difficult sources could well be a good use case!
Wow. Good job.
I‘ve been typing in some games from Mac in the past year & make them run on msBasic on my KIM-1.
I recall some issues of PCC were printed so bad that I couldn‘t read the listings (on the scan).
Now I have all early PCC‘s physically. So if you struggle with something, I can make a HQ photo if you want.
Yes, I’d love it if you could scan two bits from the September 1976 edition… the one on archive is very poor quality. I managed to pull planets off of page 16 ok, but Journey to the Center of the Earth on page 18 and Hats on page 31 I can’t even read, let alone scan.
Actually, I’m trying to scan planets from the same issue now, and it’s unreadable in a few spots too. If you can scan that first page of code for that that would help - the OCR got all of the DATA wrong and when I zoom in to fix it I can’t read it either!
Hehehe, those were the issues I was indeed thinking of. They repeated the print in Nov-Dec. 76 page 43.
I made photos of all possible listings, and uploadedd them herE:
Toilet paper describes it fairly well. I also have 5 of the “newer” PCC’s up to Jan 78. I never scanned them. Are you interested? I would set up my camera again and make pdf If you need them.
At some point, yes, but for now I’m doing 101 and then I’ll try What to do after hitting return again. The copy of that is really poor quality too, so I’m not sure how it will go. You wouldn’t happen to have that would you?
I also tried out the AI the other day to input a hexadecimal dump of a Japanese magazine. From what I’ve tried, Gemini judges the dump most accurately.
But it still makes mistakes with 0/8/B, and D and C.
It’s 10 times faster than typing it in myself.
(It would be better if it could check the checksum and correct it, but I couldn’t tell it to do that.)
Imagine how much better the BASIC world would have been if someone thought up adding checksums to the listings from the start. Proofreader didn’t come out until 1983, almost two full decades later!
I can’t tell if I should be impressed or disappointed by Google Gemini. At first it seemed to really understand code that I gave it, to the point of correctly guessing at the purpose of my program. But when I asked it to translate from Python to Java it made dozens and dozens of type errors and suggested impossible “solutions” that indicated a lack of understanding.
It comes as no surprise that Google would have a vested interest in creating artificial coders that rival their very best human coders. Nowadays many Google SWEs (rhymes with “sneeze”) use an AI-augmented IDE that is constantly making suggestions. The only complaint I’ve heard is that the IDE doesn’t realize you’re done. It keeps suggesting more code to add to your already complete program.
