spunky2342

You are very right, using a generalized LLM in combination with good music search and metadata tools works very well for this sort of thing. I know because I built a platform that does this. The big limiter isn't the tech, it's what the rights-holders will allow. They maintain tight control over their catalogs because renting that intellectual property is their entire business model. This makes them very cautious about letting any actually useful AI near their music and metadata.

Spotify is definitely not trying very hard, the author is justified in complaining. I had a very similar experience. They are in a perfect position to be able to make something amazing, but they (and other streaming companies) are somewhat limited in what they can do because of their music licensing relationships. I was able to build something 100x better than their DJ using an off-the-shelf LLM and some well-crafted search/metadata tools. It has no problem doing what Petzold wanted. It's a much better way to interact with music than what the general public has access to now, and I would love to commercialize it, but the rights-holders (UMG, WMG, Sony, etc) are very protective of what they own. If you want to know more reply here or reach out to me at info@tunistry.com.