We’ve come a long way since primitive AI music generators in 2022. Today, AI tools like Suno.ai allow any series of words to become song lyrics, including inside jokes (as you’ll see below). On Wednesday, instant engineer Riley Goodside tweeted an AI-generated song with “sad girl with a piano performing MIT license text” and it’s been widely circulated online in the AI community. started doing
The MIT License is a popular permissive software license created in the late 1980s, often used in open source projects. “My favorite part of it is ~1:25. It nails ‘warranties of marketability’ with a beautiful Imogen Heap-style glissando and then immediately pronounces ‘fitness’ as ‘fist’,” Goodside wrote on X.
Suno (meaning “to hear” in Hindi) was founded in 2023 in Cambridge, Massachusetts. It is the brainchild of Michael Shulman, Jorge Coxco, Martin Camacho, and Kenan Friberg, who previously worked at companies like Meta and TikTok. Snow has already attracted big-name partners, such as Microsoft, which announced the integration of the first version of Snow Engine into BingChat last December. Today, Snow is on v3 of its model, which can create temporally coherent two-minute songs in many different genres.
The company did not respond to our request for an interview by press time. In March, Rolling Stone’s Brian Hiatt wrote a profile about Snow describing the service as a collaboration between OpenAI’s ChatGPT (for songwriting) and Snow’s music generation model, which Some experts believe that training may have been conducted on the recording of copyrighted music without a license. Permission of the artist
It’s exactly the kind of service that irked more than 200 musical artists last week enough to sign an open letter from the Artists Rights Alliance asking tech companies to create music. Stop using AI tools that can replace human artists.
Considering the unknown evidence of training data, the ownership of generated songs seems like a complicated question. Snow’s FAQ states that music created using its free tier remains the property of Snow and can only be used for non-commercial purposes. Paying subscribers allegedly own songs created “while subscribed to Pro or Premier,” subject to Listen’s terms of service. However, the US Copyright Office last year ruled that purely AI-generated visual art cannot be copyrighted, and while the criteria for AI-generated music has not yet been resolved. Yes, it may even eventually become official legal policy.
Moon Shark Song
Benj Edwards
Using the service, Snow has no problem generating unique tunes based on your prompts (unless you provide your own) and sets those words to the stylized genres of music it uses in its training data. Prepares on a set basis. It also dynamically produces sounds, although these include audible distortions. Sound output is still indistinguishable from high-fidelity human-generated music, but given the pace of progress we’ve seen, that bridge could be crossed within the next year.
To get a feel for what Snow can do, we created an account on the site and asked the AI engine to create songs about our mascot, the Moon Shark, and savages with CRTs, two inside jokes at Urs. Interestingly, although the AI model performed the task of creating an original song for each subject, both songs begin with the same line, “deep in the digital domain.” This is likely a sample of any hidden prompts SnowChatGPT is using to instruct it when writing lyrics.
Listen is definitely a fun toy to experiment with and is undoubtedly a milestone in creative AI music tools. But it’s also a feat tainted by unresolved ethical issues surrounding the removal of a musical work without the artist’s permission. Then there’s the issue of potentially replacing human musicians, which isn’t far from people’s minds when it comes to sharing their listening results online. On Monday, AI influencer Ethan Mollick wrote, “I’ve had an AI song in my head all day. Serious milestone or good?”