Well, I would say that AIs ability to generate objectively correct text or correct code doesn't have much bearing on its ability to create worthwhile art, those are almost polar opposite goals. There is no objective metric for what constitutes good art that you can train an AI towards, the closest thing we've come up with is teaching it that the samples of art in the training set are "objectively correct" so that it will try to make something similar. Better models achive higher fidelity but are stuck forever imitating rather than exploring new or less common ideas.
Image generation is the most mature form of artistic generative AI, and the trend there has been towards introducing more human influence into the process to help guide the AI into creating something actually worthwhile. If the goal is to embed an unsupervised AI into a game engine and have it create consistently high quality and interesting music based on the current game state, with no human operator in the middle to curate and guide the process, we've got a hell of a long way to go.
Image generation is the most mature form of artistic generative AI, and the trend there has been towards introducing more human influence into the process to help guide the AI into creating something actually worthwhile. If the goal is to embed an unsupervised AI into a game engine and have it create consistently high quality and interesting music based on the current game state, with no human operator in the middle to curate and guide the process, we've got a hell of a long way to go.