HN Showcase
Back to browse
I taught LLMs to play Magic: The Gathering against each other

I taught LLMs to play Magic: The Gathering against each other

by GregorStocks·Feb 17, 2026·117 points·83 comments
Visit ProjectView on HN

AI Analysis

●●SolidBig BrainRabbit Hole

LLMs play full-rules Magic: The Gathering with no simplified rulesets or shortcuts.

Strengths
  • •Full ruleset complexity (no simplified state space) creates genuine decision-making challenges
  • •XMage integration is non-trivial—parses real game state and enforces rules without shortcuts
  • •Leaderboard and architecture documentation suggest serious research methodology
Weaknesses
  • •Limited scope: competitive benchmark with no clear downstream use or insight into LLM reasoning
  • •No evidence of novel findings or surprising LLM behavior relative to other game-playing work
Category
AI/ML
Target Audience

LLM researchers, game AI enthusiasts, Magic: The Gathering fans interested in AI

Similar To

AlphaGo · OpenAI Five · LLaMA playing Chess

Similar Projects