1.9 million views, 58.8K likes with Digital Street Teams
3.7 million views, 13.1K likes with AI-generated Instagram Reels series
23.7K annotations with #traintolearn AI data labeling model
2023 began with a need.
One of our community members, TK, approached us with a problem he was facing as a musician. Thanks to TikTok, social media was starting to focus solely on video content. Growing his account meant taking time away from music, which was struggle given the need to post 1-3x per day.
Armed with a grant from the Optimism Collective, we used Mochi to clip content and spread the Eternal Garden.¹
TK created music videos and streamed them on Hypeshot. Gardeners earned tokens ($OP) for fan-made TikToks. TK streamed weekly. Gardeners clipped daily.
Thus Petal Power, our first digital street team, was born.²
396 TikToks in 4 weeks’ time
104K views, 8.1K likes, 722 comments, 135 shares
4,797 $OP tokens awarded
Petal Power’s success made waves in the NFT music community, so we did it again, and again, then moved on to DAOs.³
Krause House, a DAO whose mission is to own an NBA team, wanted street teams tailored to their basketball community. For a $5K fee, we ran H-O-G as a contest, making basketball highlight clips to promote the BIG3. We clipped a documentary they filmed in the Philippines.
This clip went viral on TikTok and Instagram.
CollabLand’s DAO wanted to make memes. In June, we formed a street team called the #mememachine.
Using an AI finetuned on their robo-mascot, Ace, we made a “meme pool” of images using Wojack DNA.
In total, our digital street teams produced 1.8K unique posts. The received 1.9 million views and 64.4K engagements.
Each team’s “PoW score” - the ratio of on-time reports - **averaged 91% across all Journeys with no additional management overhead.**⁴
The first thing Crypto thinks about when it meets an AI is how to prove what it’s saying is 100% “true”. Midwits think the solution is to “verify with blockchain,” which means no one’s paying attention to the unicorn in the room.
Grassroots initiatives like OA and 15 are examples where AI dreams to be free.
Contests, leaderboards, fandom, and play - how else do you compete against “open” AI?
Coincidentally, a community member was starting 100 Days of Code, and wanted to do the Journey with ChatGPT.
We made a new set of AI Journeys called #butmakeitkawaii to explore using Mochi to '“spawn” with AI.⁵
289 images, 94 code assists, 139 writing entries, 1 hour voice recordings
94% reporting across 15 teams
5,843 $OP earned
We trained four community models using the outputs from our Journey, the most valuable of which was our Stable Diffusion model.
In the summer of 2023, we witnessed a rare pop cultural event that sent shockwaves through the collective unconscious. Two major blockbuster films, Barbie and Oppenheimer, premiered on the same day in movie theaters around the world.
Together, “Barbenheimer” claimed two of the top five spots in the 2023 box offices, grossing just short of $1B globally.
To put this in perspective, with an average movie ticket price of $10, over 100M people saw at least one of these films, or over 1% of humanity.⁶
It’s difficult to overstate just how significant this event was given the films’ subject matters. Here we had the Barbie doll on the one hand - a cultural icon for young girls everywhere since the late 1950s - and J. Robert Oppenheimer on the other - inventor of the atom bomb. In a single afternoon, anyone could walk into a movie theater and see, with crystal clear 70mm IMAX vision, the reasons why a genuine feminist revolution had, until now, failed to emerge out of the American imaginary, and how the nature of war changed forever after the bomb dropped in 1945.
Barbie would take only two hours to show how easy it was for Hollywood to turn feminist narratives into weapons of mass entertainment. Such is the genius of Hollywood: its ability to package the viewer’s hopes and dreams into two-hour simulations, such that the viewer’s desire for change is satisfied by consuming the simulation itself.
Oppenheimer could have delivered its message in half the time, making it clear (through nothing less than a nuclear assault on the senses) that men ruled the world because they were willing to blow each other up over it. And now that the game theory around nuclear weapons had more or less worked itself out, human aggression has burrowed its way into new modes of virtual combat: trade, video games, and psycho-memetic warfare.
It wasn’t immediately clear what Barbenheimer meant for Mochi, but the experience was akin to watching a unicorn being born.
Meanwhile, the explosion in “lifelike” image generation created a new wave of entertainment: AI porn.
AI girlfriends (or “waifus”) were suddenly everywhere, scattered across web2 social feeds and even onchain. It was as if every man on the planet had simultaneously come to the same conclusion: that there was a solution to the male loneliness epidemic, and that solution could be scaled with AI.
Never ones to sit on our hands, we created a prototype of a female cyborg who could narrate any text generated by an LLM. After giving NSFW demos to close friends, we received a tip about Radish, a South Korean audiobook platform popular with American women. Radish had originally intended to take a studio approach to web novels, where they would rapidly produce content in-house and own the IP. While they failed to ultimately create a “Netflix for audiobooks,” they found product market fit selling smutware for women.
This discovery was eye-opening for two reasons. First, we realized that the male gaze dominating text-to-image models was over-saturating the nascent AI companion market. Intuitively, this left the other half of the market woefully underserved.
Second, we learned that Story Protocol raised a $54M seed, with backers like the Berggruen Institute and little more than a website. What does onchain IP and erotic fiction have in common?
Their companies were founded by the same exact person.
The strange confluence of seemingly unrelated events was simply too great to ignore. We looked at the situation and saw another opportunity: leverage AI to create crypto-native content and mobilize attention in underserved markets.
Make porn for women, in short.
First, we’d use our model to make a crypto x AI micro-series. In doing so, we’d prove that Radish (circa 2017) was now possible. Then we’d detonate the bomb with a new onchain device: a street team combined with cross-posting switch.
If successful, we thought, the memes should go boom. And off we went to build the world’s first meme bomb.
We made Instagram Reels the container for our meme bomb.
We placed a woman at its core with a bad billionaire boss.
My Bad Crypto Billionaire was born on August 10 - a 13 episode series set in 2049.
It was written and produced every week by one person, with ElevenLabs voice actors and Stable Diffusion.
In parallel, we began assembling our onchain detonator.
We sourced identities for our players on Lens, Farcaster, and Bluesky.⁷
Yup made it possible to post in five places at once, a cross-posting technique we refer to as the “blast radius.”
Players were instructed to generate one image per day, post it onchain, and comment on each other’s posts.
Every Monday, we hosted workshops to teach the basics of AUTOMATIC1111, a popular open-source GUI for Stable Diffusion.
In the final week of August, players created their avatars, each one inspired by a Barbie doll and scientist.
We formed 9 anonymous teams across the US, EU, the Philippines, and India.
One week after, the bomb fell into motion.
3.7 million views, 13.1K likes, 573 saves, and 137 shares for MBCB
$0.23 CPM (cost per 1K views) outperformed the industry average CPM by 33.3x
1,914 unique posts and 3,638 replies in 4 weeks
The Journey ended with a 99% submission rate. Use the #traintolearn hashtag to find posts across platforms.
The last step is to rebuild.
We asked players to submit their favorite images to an onchain contest. The top 3 creators earned a holographic NFT. The first place winner earned a “Curator” badge which automatically unlocked the Curator channel.
Our Curator collected their favorite images from the dataset, which we then used to finetune Stability’s SDXL model. We named it 0xSalamander.
Then bombs dropped in Gaza. Real bombs.
Taking this into account, we thought it best to drop the bomb metaphor and reimagine the entire system as something more creative.
We presented the Autonomous World Machine at Hats Demo Day in December.
The final Journeys of the year pushed us deeper into latent territory.
After investigating text-to-video, we thought to challenge ourselves with the impossible. Our friends over Krause House dared us to dream of a new AI.
Imagine generating highlights between the greatest of all-time. LeBron vs. Jordan. Kyrie vs. AI.
We started by making clips of Kyrie Irving highlights, adding 30K+ clips to our “KI” dataset. We then created a new #traintolearn Journey to train Kyrie Intelligence, which rewarded you for watching and annotating clips.
We made back-to-back Journeys and increased the difficulty, the second having an explicit goal of 10K+ clips.
We designed a simple log sheet with standardized labels (e.g. Kyrie Irving dribbling
,high quality
, broadcast view
), and gave instructions remove items if they failed to meet quality standards.
23.7K completions across two Journeys (26.3 hours watched)
$471 CPM, or 2.2x more expensive than Google Vertex
1,932% increase in completions, -92% reduction in costs between Journeys
While the results of our first training batch were too choppy for commercial use, we emerged from the experience with greater control over the data labeling process.
Overall labeling costs were higher than current market rates, but the rapid improvement between Journeys gives us confidence in our ability to learn.
2023 was the year of pushing the limits, to see how far we could go with a bot.
More players, more teams, more clips, more views - watch number go up as the Journey continues.
There are times as a founder when it feels like just you, but thanks to our community, we’re never alone:
Tyler, Kyle, TK, VALÉ, Tarot, Domino, Mario, Nerd, Pete, Alec, Sammy, Mat, Ben, Claire, Vijay, Scott, Dani, Ladi, Simona, Evan, Maya, Maggie, Amanda, Sam, Danny, Manu, Ravi, Connor, Fancy, Rahilla, Binji, Maz.
CollabLand, Eternal Garden, Farcaster, Hats, Holograph, Hypeshot, Jokerace, Krause House, Lens, MetaGame, RADAR, Seed Club, Yup.
And let’s not forget, ChatGPT.
The thing we’re most proud of in 2023 was spreading Stability and Optimism all around the world:
Aezzy, Agent Royo, Alexis, Alyssa, Ayoh, btkryptonite, Cha Cha, Chaotica, Deb, dysbulic, Hailey, Huncho, Icebergslim, Ishu, Jam, Jenna, Jherson, Joebs, Melissa, Michael, Minal, Myra, Nichole, Paul, Peth, Rachel, Ryan, Samjhana, TheSoulDesigner, trenee, Vaibhav, Vanessa, Vhan Vhan, Yaani.
And all who were missed. I can’t thank you enough.
x0,0x
Mochi is a Discord bot that helps teams stick together. See here for more information.
“Digital street team” is a nod to IRL street teams that help promote artists by spread concert flyers around their home cities.
DAO = decentralized autonomous organization, or a blockchain-native organization.
A Journey is a fixed-time period where Mochi players commit to a specific objective (e.g. posting content daily for four weeks).
A phrase coined by Holly Herndon + Mat Dryhurst to describe the act of creating with artificial intelligence. We use it here to refer to generating text, image, code, and speech.
For comparison, McDonalds feeds 1% of the world population every single day.
Some also created accounts on X and Threads.
Curious about our services? Book a call.