Really cool stuff. I was avid mtger and drafting is really hard.
I wondered how it would do with some agents/mix-of-models magic, a ft model or maybe long enough context with data to cover basic metas and make advices on longterm strategy -> not making shots but giving hints (for the next try and pick: either high mana create or land) comboed with more short-fused model that makes actual calls according to advisory.
Hm, I am thinking this could be fun thing to reproduce for league of legends picks (they should be vastly easier tho, feels like xgboost or some other ml would suffice).
A year ago, I was thinking about something like this, but I don't have the competence for such a project. I've always wondered what would happen if I taught a model to play mtg, and then fed her the best decks and asked her to build her own.
Hey, what GPU model are you using? Can I also fine-tune it on Dataoorts: https://dataoorts.com/gpu/ for T4 GPUs?
Really cool stuff. I was avid mtger and drafting is really hard.
I wondered how it would do with some agents/mix-of-models magic, a ft model or maybe long enough context with data to cover basic metas and make advices on longterm strategy -> not making shots but giving hints (for the next try and pick: either high mana create or land) comboed with more short-fused model that makes actual calls according to advisory.
Hm, I am thinking this could be fun thing to reproduce for league of legends picks (they should be vastly easier tho, feels like xgboost or some other ml would suffice).
Anyways, great post, super fun read!
A year ago, I was thinking about something like this, but I don't have the competence for such a project. I've always wondered what would happen if I taught a model to play mtg, and then fed her the best decks and asked her to build her own.
do one with yugioh
sick