Subscribe for More

Subscribe to our community of creatives, music fans and artists; experience monthly submission calls and exclusive content. Connect with like-minded peeps.

The Mesmerized Stream cover image The Mesmerized Stream cover image
Gabriel Mazza profile image Gabriel Mazza

Spotify Announces AI Playlisting Feature

Spotify has been busy integrating an AI tool in its mobile app, allowing users to generate playlists based on moods, genres and artists. For the time being it’s only available in the UK and Australia - luckily, that’s exactly where I am based...

Spotify Announces AI Playlisting Feature

Here's Spotify's original press release:

Spotify Premium Users Can Now Turn Any Idea Into a Personalized Playlist With AI Playlist in Beta — Spotify
We know that discovery happens on Spotify, with nearly two billion discoveries happening on our platform every day. Features like our AI DJ and daylist power those discoveries and bring fans closer to the artists they love. Now we’re providing our Spotify Premium subscribers with another tool to fuel discovery and music curation. With AI... Read more »

It states: "You can use prompts that reference places, animals, activities, movie characters, colors, even emojis. The most successful playlists are generated with prompts that contain a combination of genres, moods, artists, or decades. It’s a brand-new tool, and still in beta, so we’ll continue to iterate and innovate on the experience. Stay tuned for what’s to come!" (AS IF WE NEED ANOTHER SOOTHING COFFEE SHOP VIBE PLAYLIST?)

This is how it works:

Source: Spotify

Looks easy - and it is. Best results seem to arise when prompting for moods and genres, although the tool certainly lacks precision. For example, when asked for a playlist that would "make me feel like a sad kid in 1994", the resulting selection included records released after 1994 (for instance Radiohead’s OK Computer).

A search for "late-night industrial noise" yielded a more fruitful outcome, although the AI took some convincing, apparently keen on delivering something ‘more accessible’ (cue the "excuse me sir, do you know who I am?")

Similarly, a search for a "playlist that makes me feel like Bruce Willis in Die Hard" resulted in a rock and blues-packed selection, with a decisively uplifting quality. 

💡
Now this is where it gets interesting: the new tool has a basic understanding of your taste and listening history, inputting messages such as "Loading playlist 'Your Time Capsule'". It also scouts existing playlists, see the message below:

!!! THE REAL DEAL was searching within my own listening history. For instance "create a playlist with tracks I have been listening to this past week". That worked fairly well! Perhaps not that useful on a curatorial level but certainly fun to play around with. What about the OPPOSITE? "Create a playlist with artists I don’t know and never listened to". Now THAT would be a stunning music discovery tool. It failed miserably! For some reason, it picked artists I regularly listen to. 

It wasn’t always smooth sailing. Check a series of ill-fated attempts down below:

NOW onto what truly caught my attention: searching for tracks released only this year! While the calendar filtering is not perfect, the AI tool seems to get it right when approached with a more algorithmic prompt: 

The same attempt failed when looking for tracks released TODAY (April 10th): 


I am going to leave you with a quote from Spotify: “We know that discovery happens on Spotify, with nearly two billion discoveries happening on our platform every day”. It’s been like that for a few years, and perhaps the AI approach can further empower that. Imagine being able to search by geographical and calendar values, plus genre! 

For instance: I would love to get a daily personalised playlist with all new releases in a particular genre or mood. It’d be like my own ‘morning musical newspaper’, if you will -  now imagine if you could automate that process, asking Spotify to auto-generate the playlist for you… That’s the next frontier of music discovery - which could easily be applied to searching within specific catalogues. 

Imagine having a limitless search engine able to analyze Spotify’s content pool for any relevant criteria… that’s the future, my beloved friends. (Which is already here).


ALSO>>>> PLEASE BE KIND TO YOUR AI TOOLS!

The words you use matter, especially when you’re engaging with ChatGPT
Do you start your ChatGPT prompts with a friendly greeting? Have you asked for the output in a certain format? Should you offer a monetary tip for its service? Researchers interact with large language models (LLMs), such as ChatGPT, in many ways, including to label their data for machine learning tasks. There are few answers to how small changes to a prompt can affect the accuracy of these labels.


💡
Leave us a comment below! Got any suggestions for The Mesmerised Stream? Want to share something with us? Contact us on Twitter at @itsmesmerized or Instagram at @mesmerized.io

Gabriel Mazza profile image Gabriel Mazza