AI like ChatGPT, Claude, Gemini and others have begun to rapidly change the way people live and work. While not quite reaching the level of a full personal assistant status (but not much longer?), AI can assist with all kinds of tasks and answer lots of questions.
Is it always right? Definitely not. Which is why we tend not to ask it to do heavy research lifting without checking its work. To see what we mean, we asked ChatGPT if it could be 100 percent trusted when doing research on any topic.
“Short answer: no — not 100% of the time.”
Even AI knows its limits, but that doesn’t mean it isn’t useful. Case in point: music discovery.
Can AI help us discover music? Short answer: hell yes.
Recently, we thought it might be interesting to put AI to the test in scanning for music we may not know. To do this, we asked ChatGPT (our AI of choice, though this likely works similarly with others – and we used the Plus $20 version, but the free works nearly as well) what it needed to accomplish such a thing. It asked us a series of questions and also asked if we had a Spotify or Apple Music library (we did!) and if we could share it.
After downloading our library list from Spotify – we used Exportify to accomplish the task, but there are other options out there – we uploaded the CSV file into ChatGPT.
It spent a couple minutes working and then began to break down our list into broad genres: rock, alternative, jazz, R&B, etc. Then it began to list examples of artists and songs in those genres not in our playlists.
This is where the interesting part came in. At that point, we asked it to eliminate any artists we already had heard of, but also said whether we liked them or not, and if they should be included in the discovery algorithm. Finally, we asked it to limit its choices to only music from artists who had no releases prior to 10 years ago.
The idea here was to find NEW music, but we could have easily refined it to “only music from the 1960s” or “only artists from America between 2010 and 2015.” Generally, the more specific we were, the better the responses. To help with that, we asked it to place a greater weight on comparisons to our favorite bands in each genre and provided a list.
From there, we just kept refining, replacing artists we knew and liked or knew and didn’t like with others. We asked it to provide us with one signature song by each artist we could use as a guidepost and listened. Anything we like stayed. Everything else was discarded.
What remained was a playlist of about 30 songs we could download as a CSV. We then uploaded it to Spotify via TuneMyMusic and a playlist magically appeared.
Here’s perhaps the most interesting part of this. We saved that thread in our account and can go back to it, updating and managing. Over time ChatGPT comes to really understand our likes and dislikes and can make recommendations based on that which are remarkably accurate. The more information it has, the more it understands what we want.
Some probably find this creepy or invasive and, if so, AI probably isn’t for you. And we get it. But, when you spend your life searching for music recommendations from friends and on playlists and through music blogs, you eventually start running out of time.
Whatever you like of AI, this was incredibly fast, easy and on point. And the discovery process was actually pretty fun. Plus, now we have an ongoing conversation about music with a tool that helps us find new stuff all the time.