LLM Clash

Hello everyone on Product Hunt, Late last year I found it a bit shady how a bunch of models would claim to be the best at some benchmarks (when it was a bit obvious that they were being trained to beat those benchmarks), so I created LLM Clash in my free time to test out some models. Wasn't surprised when I had models as small as "Vicuna 1.5 7B" (with a well crafted prompt) beating a fine-tuned LLama 2 13B in a basic debate. Being able to debate a point with a wide range of topics, following the below criteria should be the basics: - Clarity and Organization - Content and Substance - Rebuttal and Counterarguments - Delivery and Style - Use of Evidence and Citations When a debate has reached it's limit it will be judged by 5 models on the above criteria. Feel free to share your thoughts or potential changes you want made.