"Claude can write new rules!" Topic

I conducted an experiment. There is a very fine company that produces some rules I like. The line of rules in question are very simple, but good: interesting flavorful, playable. They are all battle-specific. This line covers the 20th century, but each game is for a particular battle/campaign. There are two that cover WW2 in Europe, but for whatever reason none for WW2 in the Pacific.

I photographed the pages for the two WW2 games – no scanning, no OCR, I just took phone photos of the pages – and gave them to Claude. First I asked it to resolve two somewhat involved combat situations, to make sure it could interpret the tables and procedures correctly, which it did. Then I asked it to create a set of rules for a particular WW2 Pacific campaign. It did it. I made a couple comments, we went back and forth a little, Andy version 4 with not much work I had a perfectly good set of rules! The historical flavor was there, the game would certainly be playable, there are interesting player choices in force selection.

Then I asked it to take everything it had learned and do *another* battle, which it simply did. Just like that.

Now these were simple rules. I plan to experiment with some more complicated rules soon. This was also just extrapolation. It was more than writing a new scenario, since new equipment and units needed to be stated up.

I have a couple other projects to try out like this. the other projects will be started from scratch so I will be able to share the results, which I don't feel I can do with these, since they are heavily derivative of someone else's excellent work.

Another question plaguing me is should I share this with the company involved? On the one hand, they might want to use this as a starting point for producing two more sets of rules with minimal work on their part (certainly not zero, but far less than starting from scratch). But on the other hand some people are very emotional about AI and they might just get mad. So I'm just going to keep my mouth shut.

I am confident, though I haven't tried it yet, that if you gave Claude a set of rules and several scenarios it could make up additional scenarios that would be playable and well-researched. They might not be balanced – AI still doesn't truly understand things, but if you point out imbalances it can quickly fix them.

And in the meantime, if you, like me, like collecting and reading rules, we have a source of new material.

I have to use AI for work. Claude has proven capable of producing new material from source material and a few prompts. On the other hand, CoPilot is dumb as a box of rocks.