Music recommendation is such a hard problem. There are all these seemingly obvious relationships you can map between bands to create a big graph that looks good but that almost never captures what goes on when a human with deep music knowledge recommends music. Often the best recommendations have no obvious relationships to the band you like.
I played around with this tool a bit and didn't find it any better then other more traditional music discovery tools, that is to say not very effective.
For example, I entered John Zorn and was recommended a bunch of John Zorn's bands. I entered The Residents and got The Pixies :/
I think its more effective to click around on Music Brainz and Wikipedia.
You seem knowledgeable about this.. Care to test my old project for music recommendation? I built it by looking at co-occurrence of artists in Spotify playlists, which gives me word2vec-style vectors, and then its just kNN.
No login needed, just enter some artist names and see what you get:
Cool app. One small complaint is the chatty tone of the recommender engine. In particular, I find it a bit disingenous to have an LLM tell me "Ah, I love <X>!".
EDIT: I also notice the recommendations are totally different when making the same query in a different session. I'm not sure if that's intentional? I expected at least some overlap with the previous time I asked.
Weird. It hallucinated one for me, and recommended an album that doesn't exist. Flashbacks of 2023. If this is your app, you might want to consider adding a validation layer that performs a review before publishing an output.
Thank you for this - I've just discovered something I'll be listening to all week. Is there any chance I could prompt it again with the bands I liked and didn't like from the list, possibly asking for something more refined?
typing in what kind of music you like and discovering music was the entire premise of Pandora and it was pretty successful.
You can easily comes up with "song/band sounds like X" but it'd be a lot harder to do with movies and TV shows because beyond genre there's a lot of variation in what makes something good. Acting, direction, lighting, story, effects, actors, etc. Failing on any one of those things means I'm less likely to enjoy a show, but being similar in any one of them could cause a match.
That said, if you were to ask for specifics it might be more helpful. Recommending movies based on tone, or style, or story elements might still be interesting but I think you'd still run into the problem where it may not easily result in something you're likely to enjoy.
I did try out the OP's recommender. It seems to misunderstand which genres strongly fit the band.
Guadalupe Plata is super-gritty mex rockabilly but the AI slotted it as delta blues.
Darla Farmer's only album is story based prog - like The Dear Hunter w/ a Diablo Swing Orchestra tone. AI called it a "hazy, intimate vibe".
All Them Witches scored better but the recommendations were all bands I know (Colour Haze, King Gizzard, King Buffalo).
I played around with this tool a bit and didn't find it any better then other more traditional music discovery tools, that is to say not very effective.
For example, I entered John Zorn and was recommended a bunch of John Zorn's bands. I entered The Residents and got The Pixies :/
I think its more effective to click around on Music Brainz and Wikipedia.
1) Imagine the timeline of musical history. If you don't have a clear idea of it, Wikipedia is a good place to start.
2) Pick any genre/period you don't know well. (For example, medieval music, or swing-era jazz.)
3) Look up the main figures of that genre/period. (For example, Guillaume de Machaut, or Duke Ellington.) Wikipedia is good for this too.
4) Listen to a sample of their most well known pieces. YouTube is good for this.
5) Repeat. Feel free to go down rabbit holes as you want.
No fancy tools needed, just your mind and the internet. This will give you interesting music for many years, and improve your musical taste a lot too.
No login needed, just enter some artist names and see what you get:
https://blog.jonas-klesen.de/artist2vec
It would be great if somebody could reverse engineer their recommendation algo
EDIT: I also notice the recommendations are totally different when making the same query in a different session. I'm not sure if that's intentional? I expected at least some overlap with the previous time I asked.
Great idea though! I got inspired to listen to some stuff by it, even though it wasn't really what I wanted to find.
4 of the 5 links didn't even work. The other sounded nothing like what I described.
> "lofi home recordings with no electronic elements and harmonically complex"
Yielded a bunch of things I would have expected but after 5-6 'dig deeper' clicks there were lots of interesting artists I hadn't heard of before.
> "contemporary classical with no electronic elements and small chamber arrangements"
Lots of interesting results I hadn't heard of before.
> "obscure indie bands associated with Olympia, WA formed in the 1980s"
Very good list. Does the time/location-bound results very well.
I think it's a good companion to sites like RYM as a middle ground between 'best of' ratings lists and personally curated lists.
I’d be curious about the data source here though. Custom curated? Relying on LLM World Knowledge with some prompting?
if so, what about something more visual like movies and tv shows?
You can easily comes up with "song/band sounds like X" but it'd be a lot harder to do with movies and TV shows because beyond genre there's a lot of variation in what makes something good. Acting, direction, lighting, story, effects, actors, etc. Failing on any one of those things means I'm less likely to enjoy a show, but being similar in any one of them could cause a match.
That said, if you were to ask for specifics it might be more helpful. Recommending movies based on tone, or style, or story elements might still be interesting but I think you'd still run into the problem where it may not easily result in something you're likely to enjoy.