Do you know what most people really like? Themselves. More particularly, seeing how awesome they are. The absolutely best use case for AI may be just that: “Hey Siri, show me how awesome I am“.
Viewer discretion advised: You will see a lot of faces of myself. You are welcome.
During the past month, we have seen a use case of AI design take off: Avatars. Two one-man companies sprung up seemingly over the weekend, offering to turn your face into an array of stylized scenes for $25: AvatarAI.me and ProfilePicture.ai.
I gave AvatarAI a spin, and the results are hilarious. To be fair, I own very few photos of myself, and I’ve seen people have even better results.
The magic of Open Source
In What do you wish existed, I wrote about DALLE and Midjourney – the two AI Models to generate images from text. That was in August, 3 months ago, or 30 years in AI time.
Since then, a third entrant has considerably shaken up the space: Stability AI released a free and Open Source model called Stable Diffusion (you can play with Stable Diffusion here). Emad Mostaque, a former hedge fund manager has spent $600 000 and hired a team of brilliant scientists to train this model and release it entirely for free. It is not as polished as Midjourney and DALLE, but it’s a perfect Open Source story: The community quickly embraced it and started improving, productizing, and introducing new use cases.
In particular, Google Research has extended Stable Diffusion to release (again, Open Source) DreamBooth. DreamBooth is a way to “tweak” Stable Diffusion with about 20 photos to produce variations of these photos in different styles. It works for People, Dogs, and Product Photography.
AvatarAI.me and ProfilePicture.ai are running DreamBooth, on top of Stable Diffusion in the cloud with Astria. Pieter, founder of AvatarAI summarized this at the bottom of his landing page:
There is also a trick to make Midjourney create an avatar for you without prior training:
The future of AI companies
It’s easy to take this piece in a predictable dystopian direction, crying out that “AI will pamper our egos while artists starve and also get morbidly obese like in WALL-E“. I trust that New York Times will stand up to this editorial challenge, so I’ll leave it to them.
For years now, the future of AI was both “settled” and a little dystopian: The giants of the Internet (Googles, Facebooks, TikToks, etc.) will gather proprietary data (which we hand off to them in exchange for “free” products) to train their models, which in turn give them further advantage to squash any competition from the little guys.
It was so fun to see those assumptions turn out false! Stable Diffusion was trained on images freely available on the Internet. Instead of relying on proprietary data to learn how to generate them, the researchers added random noise and train the model to “pull images out” of noise.
Not only did the dataset not turn out to be an moat, but the very expensive-to-train model itself was also released for free! Now every script kiddie can have a head start on Google in bringing an AI use case to the market.
I think this is the future we are heading towards: There will be a few “general purpose” models like Stable Diffusion (or whatever the Open Source version of GPT-3 will be), and the money will be made in slightly tweaking those to a specific niche, using some domain insights and finding novel and creative ways to augment computer-human interactions.
Go go script kiddies!
A thing I’ve read
Like Script Kiddies, Online Creators have a bright future ahead of them. Paul Millerd shares why he is excited about the years ahead for Online Creators:
I believe that we are in the early days of what will be remembered as one of the greatest times to be alive for hyper-curious people who are willing to be creative, connect with others, and share their ideas online.
His piece is a love letter to connecting with other people over ideas. Writing online is a multiplayer sport and sharing your own ideas is an invitation to play:
Sharing Your Unique Interests Online Is Good. Full Stop
Throughout history, people have risked death for the ability to share their ideas. Now almost everyone has unfettered access to the internet and most people are sitting there and thinking “eh, I’m good.” People will look back at us and wonder what the hell was wrong with us
The only true way to be a creator is to be a “professional dyletant”. If you take the “creator economy” too seriously, you will just create a job to hate for yourself. Nat Eliason published a fantastic piece about getting insanely rich in the creator economy, exemplifying why I frankly hate this term:
Comb through all the videos of people whose success fills you with jealous rage. Take notes on what they’re doing and how they’re doing it.
We want a strategy that we know will work, which is why your goal is to feed the algorithm. To figure out not what people like but what TikTok and YouTube and Twitter like so they put you in front of the scrolling masses who can subscribe to you. You thought you were breaking out of having a boss? Hah, no, your boss is now the codebase of hungover 20-somethings in San Bruno. Good luck getting a promotion.
It will feel awkward at first, but eventually, that little nagging voice in the back of your head saying, “this doesn’t feel authentic!” will get drowned out by an endless stream of dopamine
People like paying for the sense of progress, and they value the information more just because it’s behind a paywall.