The trend is, not every one has the capability to train a Large Language Model. So the majority will focus on fine-tune or prompt engineering :)