Music recommendations is a fools errand, I’ve never experienced or heard of a system consistently recommending it well. From my experience it’s 50:50, when it’s good it’s good then half the other time it’s Ed Sheeran
Audio analysis is one of the easiest problems that ML can deal with. The problem is, how can you use a pretrained LLM for discovering newly-released music? And how do you train future models without a source of new data?
> Audio analysis is one of the easiest problems that ML can deal with.
Maybe, but that doesn’t tell me anything about LLMs. I’m not saying that it’s a particularly hard problem, I’m surprised that an LLM specifically would be good for this purpose.