Finetuning a smaller model leading to better performance seems like a significant finding that'll lead to a lot of companies fine-tuning their own internal "ChatGPT"s