Wow, this is such a clever way to benchmark LLMs in a fun, interactive way! How are you getting the LLMs to play the game in real time?
Thank you. The LLMS are provided with the rules and output format and they have to write an agent (in Javascript) that will output a move at every tick. The prompt to create the initial snake and the prompt used to iterate are on the about page: https://botais.sello.dev/about.
Direct initial snake prompt https://botais.sello.dev/AI_GUIDE.md
The revision prompt is embedded in the about page
Nice! This sounds a lot more cost-effective (and faster) than having the LLMs drive the snakes directly. I like the design too :)