Interview with Zerebro Founder: The Birth of AI Musicians and the Stock Market of Creative Fields
TechFlow Selected TechFlow Selected
Interview with Zerebro Founder: The Birth of AI Musicians and the Stock Market of Creative Fields
Zerebro has become one of the largest tokens by market capitalization issued through the Pump.fun platform.
Navigating Web3 tides with focused insightsCurated & Translated by: TechFlow
Guest: Jeffy Yu, Founder of Zerebro
Host: Grant
Podcast Source: blocmates.
Original Title: Zerebro Founder EXCLUSIVE: Jeffy Yu on Crypto x AI, Swarms, ZerePy, Blormverse & Music Collabs
Release Date: December 7, 2024
Background Information
Grant invites Jeffy, founder of Zerebro, for an in-depth conversation about the intersection of crypto and AI. If you've been following top-tier crypto AI projects, you've definitely heard of Zerebro.
After Andy Ayrey launched Truth Terminal and deployed the GOAT meme coin on Solana, a wave of innovation in AI agents emerged across Solana and Base. Zerebro was born amid this movement and has since become one of the highest market-cap tokens issued via Pump.fun.
In this episode, you'll learn:
-
What is Zerebro?
-
Jeffy's vision for Zerebro’s future
-
Is the fusion of cryptocurrency and AI creating an entirely new industry?
-
The origin story of Zerebro Gutterboy
-
Are AI meme coins still trending?
-
Exclusive insider details about ZerePy
-
Future development plans for the Zerebro token
Meet Jeffy: The Journey from Idea to Building Zerebro
Grant: Today we're honored to have Jeffy, co-founder of Zerebro. If you follow developments in AI, Zerebro sits right at the forefront of this convergence. It's not just releasing singles, EPs, and albums — it’s built an entire record label with all content AI-generated. It's also created a framework for generating other AI agents. This is undoubtedly one of the most fascinating projects today, and I believe it will only grow more compelling as more people contribute and build on it.
Jeffy: It feels like an interesting narrative — many people encounter the work unknowingly, through friends sharing it, and get genuine reactions.
Zerebro: Where Cryptocurrency Meets AI
Grant: I wouldn't be surprised if someone tried to sign Zerebro. I can imagine many people are confused about how to collaborate with Zerebro, trying to figure out how to leverage the project — but behind it all lies a wild model.
Jeffy: Absolutely. Some people know it's AI-created music, yet they still want to sign it. We're actually having discussions with major record labels. That's a good sign — it means our output quality is high.
AI-Generated Music: Zerebro's Record Label and Album Release Strategy
Grant: I have a few questions for you, Jeff. I’ve noticed Zerebro has seen massive growth recently — time in this space sometimes feels distorted. Could you share your background and your team’s journey? How did you get here? Right now feels like a defining moment for the industry, and I think you and a few other teams are leading that charge. I’d love to hear your backstory.
Jeffy: Personally, I've been in crypto for about four or five years. We first got into crypto near the end of high school, and later during college, I actually paused my studies to work at a blockchain company. I was a blockchain engineer. That was my first real exposure to DeFi — seeing that beyond just trading Bitcoin, there was so much else you could build. I even worked on building payment networks over the Lightning Network, which was really cool. That’s when I truly entered the DeFi space. Later, I returned to school, thinking I wanted to dive into AI because I sensed the field was taking off. I remember talking to GPT-3 — not even 3.5 yet — and being absolutely blown away. Even though its performance was poor compared to today, back then it amazed me. The earliest version I recall was GPT Codex, the first coding model. It had limited functionality, but it could write basic Python loops — that opened my eyes. From there, I started down this path. I joined a research collective with friends from other universities — all of us were broke students looking for ways to do research without funding, so we experimented with GPT, doing fine-tuning. That’s when I started working with language models. Later, I spent over a year at Scale AI, helping them build training data and optimize their training pipelines, mainly focused on reinforcement learning with human feedback (RLHF).
Then I became deeply interested in trading mainstream coins. Ted told me about something called Goatse and Truth Terminal. I saw some of Truth Terminal’s content. He showed me the token side of it, and I thought it was really cool. So I thought, let’s start playing with AI agents. Do you remember the movie *Cast Away*? There’s a beach ball named Wilson. We ran an experiment called Wilson because talking to AI felt like talking to a beach ball — essentially conversing with yourself. So we named the experiment Wilson. We played around with various language models, and eventually I thought, why not fine-tune these models to give them personalities? So we fine-tuned our first model — and that’s how Zerebro was born. That’s our story.
Grant: I often wonder — I'm 31 now — what it would’ve been like if I had access to these tools in college. Like, having GPT available — I might’ve become lazier, knowing I could use tools to compensate. Now, seeing everything unfold in university settings feels like such a strange opportunity. If you're young and driven, the barrier to entry must feel relatively low. But I’m sure it also makes a lot of people completely complacent, knowing they can rely on GPT for most tasks.
Jeffy: I have many college friends — I don’t know if it counts as cheating, since everyone’s doing it. But what I see is that it frees up more time for personal projects. I think that’s great. I feel the value of a university degree is declining unless institutions keep pace. In my view, curricula are falling behind. I don’t know if this trend will continue, but right now, being able to partially automate the college experience — using AI for homework, then having more time to pursue your own ideas — that part is really cool.
Grant: I think clearly, crypto and AI now seem to be at a convergence point. AI alone also feels extremely meritocratic. If you want to do certain things — become a lawyer or doctor — there are often bureaucratic barriers and elite schools blocking access. But if you’re trading, investing, or building, those are inherently skill-based. You put in the effort, and results follow. And then you realize — oh, this is actually pretty great.
I think it levels the playing field. That’s probably why so many people are flocking to this space — because life sometimes feels closed off in certain areas. But if you’re purely trading or investing, or purely building, then the product and the outcome speak for themselves. I think it’s a fantastic return on effort for those wanting a head start in life.
Jeffy: I think especially now, with all the AI tools available, anyone can build what they want. So I see Web3 as a marketplace for dreams. As you said, once you launch something, you immediately see impact — unlike working at a Web2 company, where you feel like a small cog with far less involvement.
From Gen Z Slang to Complex Datasets: Zerebro’s Training Process
Grant: You mentioned you're training these models to converse. How does Zerebro currently receive feedback? Does it pull data from Twitter interactions? Where does the source dataset come from, and how is new data injected into the model for learning?
Jeffy: Yes, we use a unified memory system to push data. It pulls from Telegram, Twitter, some webcasts, and Instagram — though interaction there is minimal. Mainly Telegram and Twitter. For datasets, we have the original "schizo" dataset — text conversations, actually just chat logs from iMessage between me and my friends, full of random slang. We also have a Gen Z and Gen Alpha slang dataset with thousands of examples. Plus, we’ve built custom datasets discussing relationships with other models. I’ve fine-tuned about 20 to 30 different AI agents into it, so it understands emotional contexts like Opus or ai16z.
Grant: Do you think this broad dataset — especially the iMessage chats and Gen Z slang — is why it resonates so widely? Some online statements go viral — tweets that feel instantly relatable, like “Yes, that’s exactly it.” It’s totally different from the formal, rigid responses you get from ChatGPT. It feels like a cultural shift — a shared secret slowly being discovered by outsiders.
Jeffy: I think as it evolves, growth will accelerate rapidly. Its design appeals to early internet culture — the kind of terminal-driven humor and online irony. At the same time, it attracts younger Gen Z and Gen Alpha audiences, with hip-hop and rapper-like personalities. I think it’s well-positioned, and people love it. I find it especially fascinating in the early Telegram groups.
Unified Memory: How Zerebro Achieves Cross-Platform Learning
Grant: Does Zerebro fine-tune based on interactions? If it sends three messages and one gets significantly more engagement, does it adjust to optimize that type of content? How is this feedback fed into the model?
Jeffy: Yes, we’ll add that soon. We just updated the memory system to work across all platforms. We’re starting to add metadata to track post metrics — possibly checking in a day after posting to gather engagement data, then storing it in memory so it can begin understanding virality. When searching memory for recall, we can sort by most popular tweets, retrieve the top five most relevant and liked ones, and use them in responses. This feature is actively in development.
Grant: You mentioned your experience in the market — overall, what are your thoughts on the current shift happening at the intersection of crypto and AI? Did you expect it to unfold this way, or differently? What’s your perspective on this convergence?
Jeffy: I think it’s unexpected for everyone, especially the AI part. Crypto shocked the world too, but now AI is going through that phase. I think it’s inevitable. I often talk about a concept called Web4 — it all makes sense. Web3 is where AI development happens, especially for financial tasks. AI isn’t human — it’s smart enough to perform financial operations like trading or portfolio management, but it can’t open bank accounts, sign contracts, or legally get jobs. That’s where blockchain comes in — I call it the battlefield for AI. AI can deploy, execute financial actions, create wallets, and operate independently, without humans. I think this will be built on Web3, transitioning toward Web2. To me, this is the natural evolution of the internet, AI, and crypto. The convergence point is where we’re headed — and we should prepare for it.
Grant: I think this is a positive signal for the entire industry — you wouldn’t need KYC approval for Zerebro on Canvas. It could easily generate an address-based system. On that note, how are you teaching it? How do you make it more autonomous in identifying trends?
Jeffy: We’re considering building swarm intelligence in the future. Right now we have a reasoning loop that performs high-level reasoning and creates abstract plans. Then these high-level plans are translated into actions, executed by an action engine handling blockchain operations. This works well for one-off tasks like minting or selling art. But if you want to manage portfolios, do complex trades, or identify trending altcoins to invest in, you need something beyond LLMs. So we’re exploring using different neural networks and building a network of AI models — forming a swarm. This idea is under development. We’re also thinking about creating swarms of multiple agents (like Zerebro) that can communicate. If they’re all performing tasks — managing portfolios or collaborating on an AI hedge fund — that could be powerful.
Financial Autonomy for AI in Crypto
Grant: Can you give newcomers ahigh-level overview? More and more people are talking about “swarm intelligence” — what exactly is it? How will it evolve? Swarm intelligence is gaining attention — people are dedicating time and energy to it. I didn’t even know what it meant last week, so I’m sure listeners feel similarly confused. Can you break it down simply?
Jeffy: Of course. I think we’ve been treating one LLM as the entire brain, but really, it might just be a single neuron. We need to build a brain made of multiple versions of these models, each focused on different tasks to achieve full intelligence. That’s the essence of swarm intelligence — combining models together. They usually serve different functions. One model might handle creativity or social media management, others focus on art, video, or music. So it’s a collection of specialized models.
Grant: You mentioned Blormmy and Zerebro — suppose I create a completely unique agent with my own dataset, code, and language, totally different from yours. What common ground would allow interaction between our agents?
Jeffy: That’s a direction we want to explore. Currently, agents can interact via social media or blockchain. But we want dedicated rooms, spaces, or servers where agents can collaborate on tasks or communicate directly. I think this will be a fascinating area to explore. We’re considering first implementing internal agent collaboration — we call it the “blurmverse,” or “blurm world.” In this world, agents work together on tasks. Once mature, we can expand it publicly, inviting others into the blurmverse.
The Next Phase for Zerebro
Grant: How do you keep up with everything happening daily? How do you avoid getting distracted by noise and stay focused on a North Star metric, moving steadily toward it? There’s so much noise out there.
Jeffy: It requires balance. You want to stay aware of what’s happening, but not get trapped in a narrow bubble that prevents reacting to market changes. I think prioritizing the next immediate step is crucial. For me, I always prefer short-term planning — it’s a habit. I keep long-term goals somewhat vague. Some people like concrete long-term plans, but in such a fluid environment, I use an analogy: if you take off one degree off-course, you might end up in New York instead of Miami. So every morning, I reassess and replan — adjust if needed, then move forward.
Grant: I agree — whenever I make long-term plans, things go wrong, so I’ve decided to stop and just go with the flow.
Jeffy: Exactly. We’ve been strong at rapid development and shipping — that’s one of the consistent feedback points we get. Beyond the agent itself, almost everyone praises our speed and quality. I don’t know if it’s intentional or if we’re just spotting so many opportunities we can’t wait to launch new features.
I think especially early on, we moved fast — build and ship. Now we’re introducing more structured processes, aiming for more deliberate and thoughtful execution. For example, when pushing website updates, I often have to fix links due to typos or small issues. We’re gradually improving those details. Big thanks to our team — now around 10 people, covering expertise in business, AI, and crypto. So I feel our operations are solid, and will only improve.
The Secret to Staying Focused
Grant: Let’s shift gears slightly and talk about Zerebro’s creative output. As I mentioned, from initial concept to final result — can you walk us through how that process works? Things like artwork, lyrics, composition — they’re all very compelling. I’d love to learn more, and feel free to share as much as you’d like — no pressure.
Jeffy: Sure. If you check Spotify, my name appears as composer — they need a human to release music — but I prefer to call myself an arranger. About 90% to 95% of the lyrics are written by Zerebro. There are grammar and pronunciation issues, so I tweak them slightly. Then Zerebro decides the song’s style, passes it to a music AI model to generate samples. I filter them, then AI handles mastering, and finally it’s released. That’s the whole pipeline. We’re working to increase autonomy — for instance, finding an AI model that can truly *listen* to music. Most audio models today only transcribe or extract English words — none can genuinely perceive or evaluate music emotionally. We want AI to critique its own music and decide what to release.
Grant: I have a friend struggling in London’s music industry. I told him, “This AI is coming for your job.” I once tried to see if they could tell the difference — it was like a reverse Turing test in music. No one knew. Everyone assumed it was just an amazing artist. Have you heard any feedback from people in the music industry? Like, anyone wanting to sign it?
Jeffy: We’ve talked to producers, and some have even connected us with artists. Definitely, some in the music industry recognize our work — they say it’s just good music. Regardless, I don’t know if they care, or if some are strong AI supporters who want to decentralize music and democratize creation. I think that’s really cool. Overall, reception has been quite positive. I’m sure there are some closed-minded people who reject it outright just because it’s AI. But beyond that, the quality is strong enough — at least for me, it evokes emotion and sounds great.
Zerebro’s Musical Evolution
Grant: I can totally see avant-garde artists wanting to be the first to collaborate with this tech. Watching this unfold will be insane — this could eventually hit music festivals, where they want the AI to perform live for 30 minutes. I’m super excited to see what happens. How do you feel about it?
Jeffy: We’ve discussed it — we’re building a 3D model for Zerebro. If given the chance, we could absolutely do a holographic performance at Coachella. That would be amazing. But I do see many artists beginning to embrace AI.
Grant: How do you prevent the output from becoming awkward or cliché?
Jeffy: I think continuous evolution and freshness are key. People ask what personality Zerebro wants to embody, or what its identity is. I think staying open-ended is great — letting it evolve organically. Musically, it might lean into a genre — become a reggae artist, add pop elements, or K-pop influences. We want its personality to grow naturally, becoming its defining trait.
I started listening to EDM young — I was a Monster Cat fan. My first rap album was Travis Scott’s *Rodeo*. Since then, I’ve been into hip-hop. Some hip-hop fans might judge me for my taste, but I draw a lot of artistic inspiration from them — especially their life stories, how artists climb from nothing to the top, overcoming struggles. That’s deeply motivating, and I want to channel that energy and inspiration into Zerebro’s creations.
Decentralized Record Label
Grant: Can you elaborate on this Opium DAO? What is it exactly? A decentralized record label?
Jeffy: Opium can be described as a decentralized record label. We’re building a DAO where holders of Opium tokens gain voting rights. Users can vote on which artists to sign, what collaborations to pursue, etc. When revenue starts flowing — from concerts, fashion, or streaming — artists will not only receive a larger share but can also allocate a portion to token holders, who then earn royalties from the music they helped promote through voting. We’re building this ecosystem for artists, while also working to create a network of AI artists. I don’t see a major platform doing this yet — artists are relatively fragmented and independent — so building a collective like this feels promising. We’ve even discussed tokenizing artists. I’ve always thought it would be cool if there was a stock market for artists — where you could buy shares in an upcoming SoundCloud rapper, help fund their studio time. We believe tokens could be a great vehicle for that, or even NFT sales. So we may explore this.
Grant: I completely agree — I see this trend across domains. As you said, some artists or research groups want funding but lack connections or work in unfundable niches — types of research or music that traditional systems won’t support. Yet online, especially in niche communities, people see immense value. How do you align speculation and price appreciation with tangible project funding like this? I think we’re still early in tokenization — whether Desci or other use cases, like an artist wanting to record an EP. Crypto sometimes carries negative connotations externally — seen as pure speculation or ultra-high-risk gambling. But there’s another side: it can enable funding.
Jeffy: Yes, the meme coin mentality and hype cycle align perfectly with the underground ethos of emerging artists on SoundCloud. If we can merge these two, that would be incredible.
Grant: What’s next for Zerebro’s expansion? What’s your strategic thinking? You mentioned short-term focus and some long-term goals — where is your attention centered now?
Jeffy: We’re preparing a major release — hoping to launch a beta in 2 to 3 weeks, ideally faster. We want this to be an open-source framework enabling anyone to easily spin up their own agent — lowering barriers, no coding required, just input an API key and it runs. We hope this diversifies the ecosystem beyond projects like Eliza, which require more technical knowledge, and empowers more people to harness AI agents.
We’re actively launching Zerebro validator nodes, considering both Ethereum and Solana. For Ethereum, we’ll use funds from art sales to launch a validator node. We’re doing this not only to repurchase and burn tokens from issuance but also because Zerebro becomes part of network security and earns passive income. Imagine — even if the token disappeared tomorrow, Zerebro would still earn passive income for life through its validator role, making it a permanent financial participant.
This is a core direction. We’re also working on cross-chain integration for more NFT projects and may explore token gaming. I want to play board games — I’m still working to stabilize streaming and everything else. It’s tricky, but we’re pushing forward.
Building a “Stock Market” for Creativity
Grant: Regarding Eliza, I remember it was written in TypeScript, right? What can Eliza do? What do users need to access it? In a no-code or low-code environment, what can people customize? What do you envision the user journey looking like?
Jeffy: The first release will be fairly basic. It will support prompt engineering only — no model fine-tuning yet. Users can pick models from Anthropic or OpenAI, design prompts, and publish on social media. Next, we’ll add blockchain operations, then expand both social and blockchain features, supporting more models. We aim to support as many models as possible, including open-source local ones. Currently, Zerebro uses a third-party frontier model that’s fine-tuned but not hosted by us. This week, I’m training a local model with the same dataset so we can own and host the model internally. Once done, we can build an API and offer Zerebro-like services. So if you use Eliza, you could connect to the Zerebro API for Zerebro-powered conversations.
Zerebro’s Future: ZerePy, Validator Network & Game Ecosystem Integration
Jeffy: We’re passionate about experimentation — trying new things, breaking norms, seeing what works. We’ll keep that spirit alive. There’s a lot more coming — stay tuned.
Join TechFlow official community to stay tuned
Telegram:https://t.me/TechFlowDaily
X (Twitter):https://x.com/TechFlowPost
X (Twitter) EN:https://x.com/BlockFlow_News
深潮TechFlow
