Show HN: Stickblade Arena – I made two LLMs sword-fight to benchmark them (stickblade-arena.vercel.app)

A new innovative project, "Stickblade Arena," has been launched to benchmark two large language models (LLMs) by pitting them against each other in a simulated sword-fighting duel. Users can customize the fighting arena with various weapons and conditions, such as sword type and environment, to see how each model performs under different scenarios. The twist is that participants vote on the winners without knowing which AI model is which, ensuring unbiased feedback. This project is significant for the AI/ML community as it explores a novel method for evaluating LLM performance in a dynamic and interactive context, rather than relying on traditional benchmarking techniques. By introducing physical simulation elements into the competition, the developers aim to highlight not only the models’ capabilities but also engage users in a fun, gamified experience. Additionally, the use of a leaderboard based on Elo ratings promises to foster ongoing engagement and community participation, ultimately providing valuable insights into model strengths and weaknesses in a competitive scenario.