Bettr
Humans make art bettr
The problem Bettr solves
"The next big thing will start out looking like a toy"
The ghibli moment caused attention on AI not seen since the release of ChatGPT.
In my haste to quash the luddites and proclaim the virtues of AI, I joined in on the dunking on artists and creatives on x/twitter thinking I knew better.
Watching the arguments unfold in my mentions, something caught my attention. Someone claimed that if Mike Tyson had simply asked an artist to create his ghiblified portrait (whilst holding a dove), they would have done it for free just to avoid the creation of 'ai slop'.
The 2nd time someone in my mentions on x expressed the same idea, I really started to pay attention. My thought was not about artists performing slave labour of course, I had a better idea.
Currently, I have several conversations per week in which I try to explain to people who are somewhat sceptical of AI capabilities that due to the non-deterministic nature of generative ai models, their performance does require some practise and skill. I believe that for those that have adopted these tools and learned to use them, there is an arbitrage that can be capitalised.
My theory is that humans are better WITH ai. We're building Bettr to give artist that opportunity to sell their work with near 0 transaction fees (using base) which can allow much smaller transactions which is perfect for ai generated content. This is an example of a new kind of labour that I believe will emerge: Human plus AI labour.
For example, if a startup founder needs their logo redesigned but can't pay tens of thousands to a design industry, yet doesn't have the skill to produce a world class logo with AI, they can post a listing on Bettr, with a starting image (that they may have attempted themself with AI), some guidance on the intention and a small bountry. Artists can then post their submissions and the winning submission can receive the bounty (currently this is determined through social dynamic eg the submission with the most likes; however this can likely be improved with experimentation.)
You can think of this as akin to Fiverr but calibrated to 2025. The problem we are solving is matching humans who can't use AI skillfully with those that can for a price that makes sense.
Challenges I ran into
I started building this around 4 weeks ago as my exams were beginning for my masters' degree. This has prevented me from properly launching and testing the water. I will do this next week after my final exam.
The cost of transacting is the most obvious obstacle we see to any usage. That's where Base comes in. To present a benefit to buyers and sellers of this marketplace, the arbitrage needs to be relativley small in comparison to a subscription to a generative AI product (eg $20 per month for ChatGPT plus). Stripe was a little cumbersome in terms of UX and fees so I will be focusing on Base integration next week and will implement it shortly after launch.
Speed, simplicity and minimising cost for me to build were my main constraints. After a small amount of planning with a mixture of language models, I generated a PRD for Cursor to create a react web app, using supabase as a back end.
Due to my intention of spending $0 on building the site except for purchasing the domain (for $9) and using my existing cursor subscription, I had to get a little creative with the reward allocation system. Running a cron job through supabase, we sweep our db for any listings that ended in the 5 mins prior to the execution of the job. We then allocate the prize for each of these listing by selecting the submission with the most likes. To ensure no discrepancy with winner selection, we also round each listing's end time up to 2s before the next 5 minute interval, insuring each listing ends around 2s before the cron job allocates the winner.
An obvious challenge lies in the idea that as models improve, there will be less capacity for humans who use them to have an impact on the output. I am unconvinced of this as the stochastic nature of the models, combined with the ever increasing expectations of users, seems to have shown so far that there is always more that can be done by those on the bleeding edge using the models skillfully. Even if this particular use case became less suceptible to skill, pivoting or expanding into adjacent media seems vaguely trivial.
I also note that my presence on X in e/acc has the potential to upset some artists. As a result, I have removed this from my profile.
Tracks Applied (2)
AI
Stablecoins
Technologies used
Cheer Project
Cheering for a project means supporting a project you like with as little as 0.0025 ETH. Right now, you can Cheer using ETH on Arbitrum, Optimism and Base.