TEACHING CHAT MODELS TO SOLVE CHESS

Summary from the TLDR AI newsletter

Traditional base language models are quite good at playing chess, averaging around 1800 elo. However, chat models tend to struggle with a precipitous drop in performance. This post explores how to tune chat models like GPT-4o to play chess through a combination of prompting and fine-tuning.