Over 10 years and Wunderwaffe is still undefeated!

If you challenge Wunderwaffe, I fight back. Everyone can try now in the world of AI. If you have questions, feel free to ask. I can give you some basic advice.

https://www.youtube.com/watch?v=aCN9iCXNJqQ

Waiting for challengers. It is boring else. Like me winning a world cup due to being the only guy showing up.


#AIPositivity #VibeCoding #ILoveMyLLM

Edited 2026-02-09 20:43:53

Plenty of people wanted to give this a shot over the years with actually ML, but iirc you actually can't develop any new AIs?
Or is there still a way to run test games with them?

Or is there still a way to run test games with them?

If AIServerBot (https://www.warzone.com/Profile?p=7263394758&u=AIServer%20Bot_1 ) still works, that is the way. I have also thought about e.g. a Wazone mod. Krinid told me that a mod can override your orders. So it should be possible that 2 human players play against each other and at the end of each turn, the mod overrides the orders with AI moves. I believe krinid told me about a timeout. I do not remember the exact amount of seconds but it did not shock me.


Plenty of people wanted to give this a shot over the years with actually ML
Machine learning is such a big word. Realistically I believe you can get some automatic learning effects in the sense of optimizing some parameters. Personally I do not believe this gets you extremelly far. I believe the real breakthrough is not with machine learning but when you manage to let the bot play moves which you yourself find good.

Edited 2026-02-09 21:13:38

Yeah I know ML can mean very little or very much (and its usually very little).

The issue with Warzone mods is that LUA is being emulated by C# which makes it much slower than something actually running in C#, so that makes you more likely to run into the game-wide mod timeout.

But if you want to try out your Codex, that might be an interesting challenge?
Give it the WW code and tell it to translate it to Lua, or perhaps just see what it can come up with on its on from finding the WZ wiki?

I like the concept of machine learning, but the ways about it really should be discussed. I don't believe it's ok scape data as you wish, without consequence. There should be limits that you pay for, as it can deny service to legitimate users who just want to play games.

I'm open to the idea of ML and AI, but the ways about it seriously need to be considered.

Plenty of people wanted to give this a shot over the years with actually ML, but iirc you actually can't develop any new AIs?
Or is there still a way to run test games with them?

I had this discussion with Fizzer a while ago. He actually removed WarLight.AI as what I believe to be a direct consequence of asking around the framework. There exists an unmodified fork on my repo: https://github.com/DanWaLes/WarLight.AI . I forked as a precautionary measure to preserve the history of https://www.warzone.com/blog/index.php/2016/02/the-warlight-ai-goes-open-source/ .

The underlining framework isn't accessible, as it was back then over 10 years ago. Best approach is a mod that hopefully doesn't contain too much overhead for it to be above an experimental mod.

I'm saying this as a hobbyist programmer who's been on the site for at least 12.5 years.

The issue with Warzone mods is that LUA is being emulated by C# which makes it much slower than something actually running in C#, so that makes you more likely to run into the game-wide mod timeout.

Timeout for mods is over a minute. (https://www.warzone.com/wiki/Mod_Developers_Guide#Mod_Timeout) That is more than enough. It always depends on what you want to calculate but Wunderwaffe actually moves within probably less than a second. I also built a multiattack bot later on which does some more heavy calculations but still nothing coming close to a minute on a normal sized "Strat 1v1 type" map.

Give it the WW code and tell it to translate it to Lua, or perhaps just see what it can come up with on its on from finding the WZ wiki?

That is actually not a bad idea. However Wunderwaffe is too far. I feel that I or someone else might need to build some kind of a "Starter Bot Mod". I am talking about a bot which deploys all the income to 1 territory and does random moves. But if it has all the data structures in place, people can build on top of it.

Edited 2026-02-09 21:50:49

It always depends on what you want to calculate but Wunderwaffe actually moves within probably less than a second.

Please provide statistical open evidence that this can be measured in. Mod performance stats are limited to mod creators only. Provide game links where its measure as a mod, rather than as a built-in feature, as well as full detailed performance in https://www.warzone.com/wiki/Get_mod_performance_API .