I created an arena where 14 LLMs battled each other in 314 matches of Street Fighter III.
This was quite a fun project to create an actual chat bot arena, and gives rise to a new type of benchmark. Claude models were the top performing especially haiku
its interesting it would say "Jump Cancel," which i gather wasn't an allowed move from the prompt, but which is an advanced SF/fighting game technique.
2 comments
[ 3.4 ms ] story [ 21.6 ms ] threadThis was quite a fun project to create an actual chat bot arena, and gives rise to a new type of benchmark. Claude models were the top performing especially haiku