Show HN: AI decision model plays Mortal Kombat II

Category: ai-ml

Tags: game-ai, benchmark, emulator, replay-verification, llm-agents

Score: 6.8/10 (Innovation: 7, Technical: 7, Documentation: 8, Utility: 5)

FightBench is a replay-verified benchmark that evaluates AI decision models by having them play Mortal Kombat II via game memory reads, not pixels. It freezes the evaluation conditions (ROM, state, seeds) and uses a policy table or live loop, with a public leaderboard verified by replaying logged actions. It's interesting as a rigorous, reproducible testbed for LLM-based game-playing agents.

Target audience: AI researchers, game AI developers, LLM evaluation engineers

Repository: https://github.com/aipdv/fightbench · Python · MIT · 2 stars

View on Hacker News