I know the author specifically did not use a rules engine in their simulation because of uncertainty on how it would affect it.<p>I do still wonder if adapting something like card forge for llm use would result in engaging gameplay with an llm.<p><a href="https://github.com/Card-Forge/forge" rel="nofollow">https://github.com/Card-Forge/forge</a>
I actually considered using card forge when I started this. I mostly didn't end up using it because of how much more work it would have been.<p>But also with a rules engine, you have to manually go though every step, and pass priority after every action.<p>I think it makes more sense to let an LLM play magic like a person would. On early turns it is acceptable to say "I play a land and pass" without going through every phase. And you can say "I tap all my land and play this card" without having to use a tool call and agent turn for every land tap.<p>Also card forge would not let you goldfish a deck. You must have opponents.
Awesome ! Does this use <a href="https://mage-bench.com/" rel="nofollow">https://mage-bench.com/</a> , or is it a separate project? I ran 4 local models in a tournament recently with mage-bench on an RTX 5090 ; Qwen 3.6 27B won narrowly over Gemma 4 .
No, I was not aware of that project when I made this.<p>I'll have to look into that project, but I also have an RTX 5090 and did a lot of testing with Qwen3.6 27B and Gemma 4 31B. I was not able to get it to play legal turns consistently. I had to keep expanding the system prompt and adding rules for edge cases. By the end, the prompt was over 10k tokens, and while it mostly make legal turns, it did not make good turns. And all the heuristics in the prompt degraded the performance and increased the cost for frontier models.
I love obscure benchmarks, and I feel like I can trust their results a lot more - afterall, they (probably) weren't benchmaxxed. RuneBench[0] is another good example (how well LLMs can play Runescape)<p>[0] <a href="https://maxbittker.github.io/runebench/" rel="nofollow">https://maxbittker.github.io/runebench/</a>
Sadly this benchmark removes the part of MTG that is most interesting: the opponent(s). Without opponents you simply don't have a game. You just have a rules engine - quite boring!<p>I think I object more to the decks used in testing than the machines' decisions. I do have nit picks though: This hand is quite poor and should be mulliganned: <a href="https://app.mtgautodeck.com/public/benchmarks/4bd9955b-ebe1-4fe1-bb87-52d5111c4025?simulation=518a11b5-8986-431c-9968-08ee1d468831" rel="nofollow">https://app.mtgautodeck.com/public/benchmarks/4bd9955b-ebe1-...</a>. The poor runout reinforces this decision.<p>This project is cool though, props for making it!
Admittedly, the mulligan phase system prompt is the weakest part of the project. I had to add heuristics to stop the LLMs from mulliganing down to just a few cards looking for a perfect hand. The scoring for the benchmark is mostly based on if the LLM could complete legal turns, not good turns.<p><a href="https://github.com/CallumFerguson/mtg-auto-deck/blob/a877c08def403be0227aa489ec4747aa5f5d1b39/mtg-auto-deck-server/llm/prompt-constants.ts#L31" rel="nofollow">https://github.com/CallumFerguson/mtg-auto-deck/blob/a877c08...</a>
Very cool. I’ve been daydreaming about whether LLMs can be used to reason through gaming decisions.
Looking forward to this metric being Goodhart lawed.<p>Like how the strawberry example was overtrained for, or how the pelican on a bike started being used in official release posts.
[dead]