We have all been targeted by ads that you look at and wonder “how do they know I am in the market for that product?” The answer is that the AI/ML models that big tech companies have trained on our personal data are incredibly accurate and powerful.
There are two problems with this:
The first is that “our data” which they use to train their models actually belongs to us but for two decades now, we have been giving it to big tech companies.
The second is the models that are trained on our data belong to big tech even though they are trained on our data.
It doesn’t have to be this way and I don’t believe it will be this way for much longer.
Web3 will help.
Let me explain.
If you go to zora.co, you will find a social feed that feels like Tumblr, Instagram, Facebook, etc that you can scroll through and like the things you see. But there is one difference, liking is called minting on Zora. You don’t just tell the creator you like their work, you send them a tiny bit of money and you get to own a copy of the work.
It may not seem like much, but the difference here is that you own one of the things you liked and you paid a tiny bit for it. If the creator gets a thousand people to do what you did, which is not that uncommon at places like Zora, they make a nice bit of money on their work.
And the collector is building their own data set that they own. It is on the blockchain and it belongs to them.
The next obvious step is for companies like Zora to offer collectors the ability to train models on their collections. This turns their collections into training data sets. But these are training data sets the collectors own. Not training data sets that Zora owns.
It won’t be long until we have open-source AI/ML models that we can run on our phones. These will be our models and we can train them on our data sets.
Consider this blog post. You can collect it too. There is a green button on the upper right of this post that says Collect. When you click that the same thing happens here as what happens on Zora. I get a tiny bit of money and you get your own copy of this post.
Below is a screenshot of an Ethereum wallet I have connected to this blog. You can see some of the collecting transactions from this blog over the last week or two.
So what is going on here?
1/ Writers are getting paid for their work
2/ Readers are building a data set that they own, On Chain. Not on Facebook.
The next obvious step is for us to have our own open-source models that we train on these collections we are building.
These open-source models will help us write, find new things to read, and more. It can inspire us to start a new company, invest in a new company, listen to a new song, find artwork to hang over our fireplace and many other things we want to do.
Going back to Chris Dixon’s words in yesterday’s post:
in the long run, we are still going to need an economic covenant between AI systems and content providers. Al will always need new data to stay up to date. The world evolves: tastes change, new genres emerge, things get invented. There will be new subjects to describe and represent. The people who create content that feeds AI systems will need to be compensated.
There is a way forward that works for writers, readers, collectors, creators, and everyone.
It starts with us owning our work and allowing others to pay us to collect our work.
The thing that makes me so optimistic about this is that it is not some dream. It is happening right here on this blog. The tools we need to change the way the world works are already here. We just need to start using them.
My next post will be about how we get more people using these tools.
Over 38k subscribers
Squinting at the crypto x AI Agent hype theres a kinda new, kinda old, still fuzzy angle on a "big idea" I've been excited about for over a decade: all the world's media onchain. something like... AI agents form the demand side of a market for onchain media. Agents "mint" to own media for training or some other objective. This demand for low-value mints drives onchain as the standard "port of entry" for all internet media hard part is thinking through which agents/media types to start with, and what that market/standard looks like. if you have thoughts please HMU :) A few blog posts on this topic below from @fredwilson, @jacob, yours truly.
https://jacob.energy/ai.html
https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too
https://jessewalden.com/all-media-onchain-crypto-as-a-port-of-entry/
@fredwilson.eth
why is mint the new like? any good ELI5-ers out there, or write -ups? (i thought someone wrote something can't find it)
May it was this recent @fredwilson.eth post! https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too?referrer=0xaeA4A0dEDb94BA5b2b8ED9477A8a54379C584542
This is perfect. Ty.
bringing people onchain is the ultimate public good feeling inspired by @fredwilson.eth ⌐◨-◨ https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too
What we see now with frames makes it much easier for people to imagine how the vision for on-chain becomes really what is going to change the way the internet works On chain is the new On Line https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too
AI hype is at all time high but what I like most about it is the shift in UX Blunt tools to guide AI feeds -> nuanced tools Former excels at passive entertainment but lacks at intentional paths. This overlaps with @fredwilson.eth’s post https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too
Cookies that I own and drive to guide my experience across the web seems inevitable. Climbing the DIKW pyramid could be a paradigm shift. Corps know actioning is more valuable than raw info. How do we give consumers actionable knowledge by default?
The current default is unlimited info without many accordances for how to use it or to guide it Culturally, this is why you notice a strong love/hate relationship with the web for many. Corps have less of that view because at scale the metrics are still growing and their models are aligned with their business outcomes
Hi Casters. I promised a follow up to yesterday's post on AI being the native business model for content. Here is that follow up post https://avc.xyz/minting-is-the-native-business-model-for-web3-and-maybe-ai-too
Agreed on the thesis. Biggest challenge is how to make this behavior desirable to a mainstream audience.
I don’t remember social media being desirable to everyone at the beginning. Even now it’s quite tribal, some like IG others FB etc. I suppose at some point the UX improves so the experience of using say Zora is basically the same as IG except you can collect / be collected which is inherently more powerful imo.
Bring the creators and content they love onchain. It’s that simple and that hard.
Curious, do you think that the supply side (onboarding creators) is more important / urgent than the demand side i.e onboarding users ?
They're very closely correlated and I think strong content drives more users. Recently I think platforms have put too much effort on supply (creators) and not enough on the experience for collectors (demand)
Make it desirable to a niche audience first, then the next, then the next.
I remember using steemit in 2016 (before I lost my key) - I loved the idea of owning my content, being rewarded and also rewarding others vs. rewarding fb etc. 8 years later ‘on-chain UX’ has improved a tonne and is good enough for those who will work things out. When do you think it will be ready for the masses?
I appreciate this might be in the next post, but are there any examples in the wild of a median user revealed preference to own their “own” interest graph data (or even do micropayments)?
Users might not care (yet) but creators do. Users that care about a creator's content will happily pay for it. Other users might not pay directly for the content but they pay with attention which can be monetized and creators can be compensated for it. The model hasn't been cracked yet but we can experiment and learn
Agree with Point #1 Disagree with Point #2 - Users complain about paying for Creators stuff (and any stuff really) all the time. #3 is true in theory, but again that's why I asked for examples in the wild.
Re. #2: Anecdotal data but there are a few podcasts I listen to that were dropped by their networks so they started a Patreon model to let listeners pay. A couple of them actually quit their day jobs and focused on podcasts since they can now make a living out of it. (They make more from Patreon than network + ads).
Re. #4: YouTube's rev share is an example of this. YT is the web2 platform with the lowest take rate. YouTubers get a cut of the Premium subscription and the ads based on video watch-time and such. This type of model could be replicated in web3 perhaps with even lower take rates.
FC ?
I'd like that to be true, but we don't really "own" our data here in a robust way (can pay for temp storage, but actions expire and follows aren't in the protocol) We don't do micropayments at any kind of scale Also there are only 2500 DAUs
follows are on the protocol, just channel follows aren’t yet
I think *owning* in this context has to be understood in relation to *not owning* For example, getting kicked off X with no recourse... now don't you wish you *owned* that account, even to some extent? I agree that ownership doesn't equal value and that rights must also be more clearly defined to have meaning
say it louder
something something "we're so early" But ya, I actually think the missing piece is what value results from our contributions to these models. So less about "owning our model" and more about owning the value that model can create.
I'm starting to think of Farcaster like "/costco internet". You pay to get in but it's a better deal
Great post. Aligns beautifully with what I've been thinking about this past couple of years. "My next post will be about how we get more people using these tools. " Looking forward to it. I've been exploring this as well.
100% agreed
Great essay! This could also work the other way around: Creators today add a “collect” button and receive tips from eager supporters. Consumers could also add a “pitch me” button and get approved creators directly feed them interesting results based on their interests, for a small fee given to the user.
The step I struggle with is where user controlled data (onchain) automatically leads to user controlled AI models. Why couldn’t a private company-controlled model train on public user data?
This is true today, but changes when onchain privacy controls become widespread. If I can control who can see my onchain profile and history then I do own that data.
That is true. Private state would make a difference
I had the same thought, Alok.