"Chat GPT as Umpire?" Topic
5 Posts
All members in good standing are free to post here. Opinions expressed here are solely those of the posters, and have not been cleared with nor are they endorsed by The Miniatures Page.
Please be courteous toward your fellow TMP members.
For more information, see the TMP FAQ.
Back to the Wargaming in General Message Board
Areas of InterestGeneral
Featured Hobby News Article
Featured Link
Featured Workbench Article These are not the seasonal figures that you might give your mother to put on the shelf!
Featured Profile Article Thanks to the generosity of TMP readers, there has been much progress in building a new home for our staff editor and her family, evicted from their home.
Featured Book Review
|
Please sign in to your membership account, or, if you are not yet a member, please sign up for your free membership account.
 Extra Crispy  | 18 Dec 2025 12:14 p.m. PST |
I have yet to use ChatGPT or any other AI tool. But I wonder….is there any way it could be used as an umpire for double blind games? Anyone with real facility with these tools able to help out? |
ThunderAZ  | 18 Dec 2025 12:53 p.m. PST |
I've used AI to umpire before. In general it has worked OK, it not perfect and if you are a rules lawyer, AI will give you high blood pressure… given you prep the AI session with ALL of the correct information prior to kicking off the game. I've found that if there is any complexity to the rule system, AI can make a lot of errors. I use it to play test my rules. Even if it misunderstood the rules, it always provides useful feedback. It is also fantastic at reading through my rules and suggesting changes and improvements. |
| Andrew Walters | 19 Dec 2025 10:33 a.m. PST |
AI, in this case LLMs, makes stuff up. This can be very, very useful. But it doesn't understand what's going on and cannot be made to understand what's going on. So it's not going to keep track of where your ships are and where my ships are and roll for detection and then let one or the other of us know what's happening. If you ask it to it will produce the output you ask for, but not via the method you describe. If you ask it how it produced the output, it will make up an answer to *that* question, and it may in fact describe the process you requested that it use. But the process it describes when you ask it what process it used is *not* the process it used. So it will not reliably follow the rules, even if it can parrot them back to you and claims to use them. On the other hand, and I am 100% serious here, you can describe the javascript app you want and it will create that app. The app it creates *will* work just as you describe. I've done this a bunch of times. It's impressive how complicated your description can be and it will still deliver a working app. Then you test it, discover your description was not quite what you intended it to be, or you discover a missing feature, or you want the appearance modified. So you have an ongoing conversation and ask for changes and you get them. For modest sized apps such as one to umpire a double blind game this is not hard. It may take twenty minutes or more to describe the app, but it works. Provided, of course, you can accurately describe what you want. A lot of people can't. I think Claude is best here. Play with the LLMs, it's interesting and thought provoking and you will find uses for it. |
| Andrew Walters | 19 Dec 2025 10:45 a.m. PST |
Paste this into the (free) Claude app: Write me a web app using HTML, CSS, and Javascript, no libraries. This app will umpire a double blind naval wargame for WW1. Two players will each enter the locations of their fleets, the app will look for locations where both players have fleets and let the players know that there will be combat at these locations, but will not otherwise let the players know where the enemy's ships are. The appearance of the app will have a WW1 naval theme; ie everything is grey with some bolts here and there and some rust stains. There will be four tabs. The first tab is just a welcome and instructions. The second tab is where the first player enters the locations of their fleets. When the user switches to the second tab it will not show its contents until the password "firstplayer" is entered. Then it shows an editable text field. The player will enter the location of their fleets as a series of lines, each line with an x and y coordinate such as "3,4" or "3 4" or "3N 4W". A warning will be shown if the lines cannot be parsed properly so the player can fix them before switching back to the first tab when they are done. The third tab is like the second except this is where the second player enters their locations. The password here will be "secondplayer". When the players click on the fourth tab a function will run that looks for common locations in the lists in the text areas on the second and third tabs. The app will announce that there is combat at those particular grid points. |
| Andrew Walters | 19 Dec 2025 10:49 a.m. PST |
This will probably produce something not quite right. If you wanted to make it useful you would spend some time tweaking this input. But you might never have to touch the actual code. Obviously you could take this further, add detection ranges, let the players choose their own passwords, allow players to name their fleets for clarity, store the positions in localStorage so it will be preserved between sessions, etc. But you see that just a few minutes work can create a simple and useful app. The key to taking advantage of AI is understanding what it's good at (writing code, parsing text, making stuff up) and what it's bad at (facts, judgement, having a clue what's going on) and then using it appropriately. Kind of like every other tool since we figured out how to sharpen rocks. |
|
