dadabots

Artist

dadabots · dadabots.com/bytebeats ↗

Based

Formed 2012, Boston, Massachusetts. Based in Boston and Sacramento, US

Practice

Neural synthesis, generative music, live coding, machine learning research, hackathon culture, and on-chain audiovisual systems

Now

Co-developing AI music tools, live neural synthesis performances, open research, and long-running generative livestreams

Raster

raster.art/artist/dadabots ↗

In the set

♫ ByteBeats ↗ · 512 editions

Links

Full bibliography ↗

dadabots is the collaborative project of CJ Carr and Zack Zukowski, a duo whose public language oscillates between band, research lab, hackathon team, and machine learning studio. Their work begins from a refusal to separate music from software: they make tools, train models, build listening systems, and then perform through those systems. Profiles of the duo trace their origins to Berklee College of Music and Music Hack Day MIT in 2012, grounding their practice in musicianship before it moved into code, neural networks, and live generative performance.¹

The project's notoriety first formed around neural synthesis rather than image generation. dadabots trained models on extreme musical idioms such as black metal, math rock, punk, and free jazz, then released albums and livestreams that sounded like degraded genre memories being reconstructed by an alien ear. Their research paper Generating Black Metal and Math Rock: Beyond Bach, Beethoven, and Beatles describes the use of a modified SampleRNN architecture to generate raw audio directly in the time domain, an approach that mattered because timbre, distortion, density, and space are not decorative in these genres, they are the composition itself.²

dadabots make the machine audible not by hiding its errors, but by letting its artifacts become musical material.

From neural synthesis to bytebeat culture

The early dadabots position was not simply that AI could imitate music. It was that machine learning could expose hidden assumptions inside genre, authorship, and style. Their second research line, Generating Albums with SampleRNN to Imitate Metal, Rock, and Punk Bands, presents album generation as a proof of concept for machine-assisted production, including generated artwork and titles, while still distinguishing between fully automated and human-curated approaches.³ This distinction remains central to their mature work: the system is never just autonomous, and the human is never simply outside it.

Their public FAQ described dadabots as a cross between a band, a hackathon team, and an ephemeral research lab, a formulation that helps explain why their output appears across papers, livestreams, collaborations, competitions, talks, music releases, and NFTs.⁴ The press page and project archive document a practice that treats publication and performance as part of the same circuit: papers lead to code, code leads to streams, streams lead to collaborations, and collaborations feed back into new methods.⁵

♫ ByteBeats, created with KAI and released through Art Blocks Factory on 2 April 2021, condenses this history into a blockchain-native form. Art Blocks records the collection as 512 unique artworks by Kaigani Turner (KAI) and dadabots.⁶ Verse describes the project's premise with unusual precision: generative music in 40 to 140 bytes on-chain, where bytebeat expressions synthesize audio and video through bitwise operations.⁷ The work therefore belongs to both generative art and generative music, but it does not illustrate one with the other. Sound and image share a common computational source.

Live on Art Blocks: bytebeats-by-dadabots-x-kai ↗

The on-chain instrument

Bytebeat music has a compactness that suits blockchain art. A short mathematical expression can produce a stream of samples, often crude, buzzing, rhythmic, and surprisingly musical. In the dadabots and KAI implementation, that compression is not nostalgia for early computing alone. It becomes a way to ask how little code is required for an artwork to feel alive. Each edition of ♫ ByteBeats is therefore a tiny instrument, an encoded audiovisual score, and a collectible output of an on-chain system.

The project sits at a productive distance from dadabots' neural work. SampleRNN and diffusion models are heavy, data-hungry systems that learn by ingesting large corpora. Bytebeat code is small, explicit, and procedural. Yet both methods share a fascination with emergent excess: a tiny formula or a trained model can generate more output than a human could reasonably precompose. This is the bridge between ♫ ByteBeats and the later dadabots language of infinite livestreams, prompt jockeying, and genre engines.

dadabots' long-running livestreams made this tension visible to a broader audience. Press coverage of Relentless Doppelganger described a 24/7 AI death metal stream generated by a neural network, an event that turned research into an ambient public spectacle.⁸ Other outlets framed the work through the uncanny persistence of the system, noting that the software could churn out technical death metal in real time while listeners gathered around the stream.⁹ These reports sometimes sensationalized the work, but they also registered its central innovation: dadabots changed AI music from a file into a situation.

Collaboration, performance, and institutions

The duo's work has repeatedly entered institutional and interdisciplinary contexts. The Bell Labs and Reeps One collaboration We Speak Music positioned machine learning and generative audio as performance tools, not merely studio utilities, and connected dadabots to histories of experiments in art and technology.¹⁰ A related account of Reeps One's project described how Dadabots could model a vocal essence, producing speech-like output that sounded like the performer without simply repeating recorded language.¹¹ In these settings, the voice became both source material and contested identity.

Their public interviews also clarify the ethics of their practice. In the FAST45 interview, Carr and Zukowski present themselves as musicians, machine learning researchers, programmers, and hackers, emphasizing the hybrid nature of the work rather than claiming machine autonomy as a spectacle.¹² In a later Art Blocks Marfa interview, dadabots linked their roots in metal, punk, and electronic music to their continuing interest in lo-fi sounds, live neural synthesis, and the Art Blocks community.¹³ These statements matter for ♫ ByteBeats, because the project can be misread as a technical novelty when it is better understood as a compact expression of a larger performance philosophy.

dadabots have also appeared within broader debates about AI music. TIME placed them among artists exploring AI as a creative collaborator, quoting Carr on the value of weird, broken, and unfamiliar sound.¹⁴ Pitchfork's wider account of musical AI supplied the cultural background against which dadabots emerged, a field concerned not only with composition, but with imitation, authorship, taste, and the limits of machine style.¹⁵ The dadabots answer to that field is not polished automation. It is noisy, unstable, genre-literate, and deliberately excessive.

After Art Blocks

Since ♫ ByteBeats, the duo's work has continued to expand. The AI Song Contest pages document dadabots entries across multiple years, including accounts of human-AI process, model building, genre fusion, and open-source commitments.¹⁶ Their music archive lists projects such as PROMPT JOCKEYS, PHỞ QUEUE LIVE, Bandschleifer, Nuns in a Moshpit, and I Throw My Things Down The River, showing how the project moves between contests, livestreams, collaborations, albums, and experimental tools.¹⁷

The later technical lineage reaches Stable Audio. Stability AI's 2024 announcement describes Stable Audio 2.0 as a system capable of generating structured songs up to three minutes long, while the 2026 Stable Audio 3 paper, co-authored by CJ Carr and Zack Zukowski, presents fast latent diffusion models for variable-length audio generation and editing.¹⁸ The continuity is striking: from raw SampleRNN experiments, to byte-sized on-chain instruments, to diffusion-based music systems that run at consumer scale, dadabots have repeatedly worked at the point where research infrastructure becomes cultural form.

Within AB[500], ♫ ByteBeats is significant because it insists that Art Blocks history is also a history of sound. Its 512 editions are not silent images with musical associations, but audiovisual computational objects. They connect early bytebeat minimalism to NFT-era on-chain generativity, and they connect Art Blocks to a longer experimental lineage of hackers, musicians, neural networks, and live systems. dadabots' contribution is to make the artwork behave like a machine you can listen to.

Bibliography

The references cited in this essay are listed below. For the complete bibliography see the dadabots links page ↗.