DADABOTS and the Endless AI Livestream: Music That Never Stops
DADABOTS is a music technology project by CJ Carr and Zack Zukowski that trains neural networks on audio recordings of specific genres and lets the models generate new music directly as sound. Its best-known experiment, launched in 2019, was a YouTube stream of heavy metal created by a neural network, running around the clock; it played day and night without repeating itself exactly. The project attracted attention because it skipped sheet music altogether: the network learned from raw audio and produced raw audio, including drums, guitars and vocal textures, all at once.
This article explains how the project worked, why it matters for discussions about AI and creativity, and what music students can learn from it.
Learning straight from sound
Many earlier AI music systems worked with symbolic data, such as MIDI notes or scores, and needed synthesisers to make sound. DADABOTS used a model architecture called SampleRNN, which predicts audio one tiny sample at a time based on what came before. Trained on hours of recordings in a single style, it learned the textures of that genre: distorted timbres, rapid drum patterns and abrupt changes.
What it captured well
Sound colour, energy and short-term patterns that make a style recognisable.
What it struggled with
Long-term structure. Songs did not develop like human compositions with verses, choruses and planned climaxes; ideas appeared and dissolved.
Why a livestream?
A stream that never stops turns a technical demonstration into a public experiment. Listeners could drop in at any moment and hear something new. The format also highlighted a striking feature of generative systems: they can produce unlimited quantities of material. That raised questions about value. If music can be generated endlessly, what makes a particular piece worth listening to?
The creative and ethical debate
The project sparked discussion on several fronts.
- Authorship: the creators designed and trained the system, but the moment-to-moment music came from the model.
- Training data: models learned from existing recordings, prompting questions about consent and credit for the original artists.
- Purpose: was the stream art, research, entertainment or all three?
The team engaged openly with musicians and described their work as collaboration with artists in some projects, which helped frame the debate around permission and respect.
What the experiment reveals about music
Listening to endless generated audio makes the role of structure obvious. Human music usually tells a story in time: tension builds, themes return, sections contrast. The stream sounded convincing for seconds and wandered over minutes. That contrast is a useful lesson for any student.
Compare generated and composed music
Try this comparison at home. Listen to two minutes of any long generated audio stream, then two minutes of a short human-composed piano piece you know well. For each, write down when you noticed a new idea, when something returned and when you felt a sense of arrival. Most listeners find many new ideas in the generated audio but few returns or arrivals. The human piece usually shows a clear plan, which is exactly what a performer needs to project.
Lessons for pianists and students
Study form, not only sound
Recognising introductions, sections, repeats and endings in pieces you practise makes your interpretation clearer.
Notice what makes a style
Pick a genre you enjoy and list its surface features, such as rhythm patterns, chord colours and textures. Then list what happens over a whole piece. You will see that style lives at both levels.
Use variety with intention
When improvising, repeat an idea before changing it. Listeners need recognisable material to follow a musical journey.
Where AI audio went next
Since the project, audio-generation models have become far more capable, producing clearer sound and longer structure. Many musicians use such tools for sound design, demos or inspiration. The open questions DADABOTS raised early on, about authorship, data and value, remain central to the conversation.
FAQ
Q: Who created DADABOTS?
A: CJ Carr and Zack Zukowski, who describe it as a music technology and art project.
Q: How did the livestream make music?
A: A neural network trained on recordings generated new audio continuously, sample by sample.
Q: Did the music have song structure?
A: Only loosely. It captured style and texture well but struggled with long-term form.
Q: Why is the project important?
A: It showed early how AI could generate raw audio at scale and raised questions about authorship and data.
Practise with focus using TIGA ALPHA
TIGA ALPHA is an AI piano teaching app built by a Thai team, with lessons in Thai, English and Mandarin. Its CEO, Teacher Aum, has 20 years of experience in piano teaching, and that music teaching background shapes every lesson, while piano teaching with AI gives you feedback each time you play piano. Use it as an online piano lessons app for short daily sessions, follow it as a complete AI online piano course, or combine it with a piano course or piano learning course led by a teacher. Whether you want to learn piano from zero, learn piano online after a long break, learn music alongside your children or simply play music for fun, this online piano teaching application keeps your next step clear. New users get a 7-day free Premium trial, then 799 THB/month. Open the piano teaching app below and pick a first lesson that fits your week.
- Download and start: TIGA ALPHA landing page
- In-person lessons in Thailand: TIGA Studio
- About the company: tigaalpha.com
Endless music is easy to generate; music with direction still needs a player who knows where it is going.