What Is Decentralized AI?
Asked whether patients send their glucose readings up to FLock, Jiahao Sun says it works the other way round. Your data stays local with your iPhone, so you don't have to worry that it's being uploaded to anywhere else.
In this episode · 22 chapters
- AI, blockchain, and what decentralised means here
- Where FLock started
- Chips got cheaper, models got smaller
- Send the change, not the data
- The risk of handing your data to an enterprise tool
- 23andMe, and what happens when a company dies
- Where federated learning actually gets used
- A philosophical choice
- What blockchain makes harder
- The glucose app, and how FLock works
- The museum analogy
- Real world use cases
- Trading desks, unpacked
- Baby scans, and the retail app
- The safety layer
- Five to ten years out
- Can hallucinations be removed?
- The carryover question
- Hot buttons
- Staying out of your own way
- Lessons from Thinking on Paper
- Google Docs, and the cost of switching
You tick the terms and conditions box when downloading the software, which means that you forfeit everything already.
Jiahao Sun, CEO of FLock.io, pairs federated learning with a blockchain: part of the model trains on your device and only the change travels back. He draws on his years as RBC's global head of AI to explain why banks banned GPT at first.
He covers trading desks that cannot share data, a planned app that renders a baby's face from an ultrasound scan, and why he doubts knowledge graphs can keep pace with large language models as a fix for hallucinations.
Jiahao Sun, founder and CEO of FLock.io, argues on the Thinking On Paper podcast that decentralised AI can compete with big tech by training models without anyone handing over their data. FLock combines federated learning with blockchain incentives: “Your data stays local with your iPhone”, and a small piece of the model comes to the device to learn.
What is decentralised AI?
Decentralised AI trains models without anyone handing over their data. Jiahao Sun explains that FLock combines federated learning with blockchain incentives: the data stays on your device, and a small piece of the model comes to it to learn.
What is federated learning?
Federated learning trains a shared model by sending back only the gradient, the record of how the model should change, instead of the raw data. Sun says Google developed it in 2017 for typing on Android, but it stays centralised because users must trust Google not to take the data anyway.
Where does the name FLock come from?
FLock's name joins federated learning and blockchain. Sun says the public ledger takes the place Google or Apple hold in their own systems, so anyone can check that no data was sent to a third-party server, and FLock itself does not have to sit between clients.
How do training tasks and staking work on FLock?
Anyone can post a training task on FLock with a token bounty, such as a medical company that wants a glucose monitoring model. Contributors must stake tokens, meaning lock them up as a deposit; Sun says anyone who submits random data or tries to steal data is penalised and “slashed out for the training”, while honest trainers share the reward.
Who has FLock worked with?
Sun says FLock has worked with University College London Hospital on glucose monitoring and eye clinic models, and with trading firms such as GSR whose desks cannot share information. It also calculates credit scores for Web3 users without them revealing their wallet addresses.
Can knowledge graphs fix LLM hallucinations?
Sun, who spent four or five years as a research fellow at Imperial College London working on graph neural networks, doubts that knowledge graphs (structured maps of verified facts) will fix LLM hallucinations. He says they cannot keep pace with the brute-force growth of large language models, and that in finance only Bloomberg had enough data to make one work.
Where they disagree
Jeremy points out that users still let something onto their phones, and in the backstage review he doubts people will trade convenience for privacy: “I just don't see the convenience and security seesaw changing”. Mark agrees that people will switch only when the incentive outweighs the shortcut.
Links from this episode
Questions
What is decentralised AI?
Decentralised AI trains models without anyone handing over their data. Jiahao Sun explains that FLock combines federated learning with blockchain incentives: the data stays on your device, and a small piece of the model comes to it to learn.
Who is Jiahao Sun?
Jiahao Sun, founder and CEO of FLock.io, argues on the Thinking On Paper podcast that decentralised AI can compete with big tech by training models without anyone handing over their data.
What does FLock.io do?
Jiahao Sun, CEO of FLock.io, pairs federated learning with a blockchain: part of the model trains on your device and only the change travels back.
Can decentralised AI compete with big tech?
Jiahao Sun, founder and CEO of FLock.io, argues on the Thinking On Paper podcast that decentralised AI can compete with big tech by training models without anyone handing over their data. FLock combines federated learning with blockchain incentives: “Your data stays local with your iPhone”, and a small piece of the model comes to the device to learn.
Transcript
Jeremy Gilbertson disruptors and curious minds. Welcome to another episode of Thinking on Paper. Today's big question, how do we build AI that is decentralized, privacy preserving and fair without sacrificing performance? Imagine a world where anybody, not just big tech, can train AI models, control their own data and get rewarded for their contributions. That is what FLock.io is building.
and we're super excited to talk to them, but we wanna really dig into does decentralized AI really work and what does it mean for the future of AI development? That's what we're breaking down today. My name's Jeremy, this is Mark. We get into this every week. Mark, talk about our guests and kick off the episode, sir.
Mark Fielding Yeah, thank you, Jamie. It's gonna be a fun episode today. We've got all the hot buttons coming later. We've got our backstage breakdown of the show. And I don't think that the world has ever needed FLock AI IOs. It does today. It's the 27th of March, 2025, and the world is reeling from the upgrade announcement from OpenAI on ChatGPT and their image production. And I've just spent far too long.
looking at everybody in the world posting pictures of their face, I'm all for ties into a Ghibli animation production. And I think it just highlights where we are with the power that these centralized AI companies wield. And perhaps even more importantly than that is the.
IP data, our information. Privacy means nothing, doesn't it? So we're going to get into all that, the solutions with our guest, Jiahao Sun is the founder and CEO of FLock. And they are hopefully going to solve some of these AI conundrums for us and create a decentralized AI that we can all use, that's cheap, affordable, available and mine. My stuff is mine.
Jeremy Gilbertson She had welcome to the show, welcome to the show.
Jiahao Sun Hey, hey, thanks for having me. Hey, Jeremy, Mark.
Mark Fielding Thank you for being here. Thank you for thinking on paper with us.
AI, blockchain, and what decentralised means here
Jeremy Gilbertson So we're talking about this intersection of AI, of blockchain, of decentralization. Let's start from a high level. Why does any of this matter? Why are we even thinking about bringing a decentralized approach to AI?
Jiahao Sun Yeah, just go back to the example you just mentioned about recently. Everyone's so hyped about the Ghibli version of their face and posting on Twitter. But then have you ever thought about the thing behind this is you're actually sending your personal photo over to an API and sitting somewhere, I don't know, in SF or somewhere. And then they render the photo out and then send back to you.
During the process, of course, you have a beautiful photo, but then you've already sacrificed your own personal photos. If that's sensitive in some ways, then yeah, that'll be risky. that's where, initially, why people were thinking about, can we make models or can we make AI training more decentralized? And we don't have to connect to a server that I don't even know where it is. And I can put everything, put all my data securely, locally in my own environment.
That's where the start of the conversation, the start of the thinking comes through. Yeah, and FLock is one of the earliest actually in the world to doing this.
Mark Fielding I was, I hadn't even really thought I was so angry about everybody using one of my favorite movie producers, writers, creators who create some of the most beautiful worlds that I've ever known in my life and just using it for this frivolous nonsense. But I hadn't actually thought of them sending their face via an API to some data center somewhere. My brain hadn't even worked to that level next. And I don't think that many people using it even think about that.
Jeremy Gilbertson Well, here's the thing too. Mark and I talk about this all the time. I use Waze as the example, right? So I can get from A to B very simply. I can avoid traffic. I can do all this stuff. But guess what? Waze knows exactly where I am at every single point of the day. But I'm willing to give up that security for this convenience, for this ability to make my face into a Ghibli, right? So how do we... Let's talk about the culture side of that, right? Like, is the world ever gonna be...
ever going to understand that piece of the puzzle.
Jiahao Sun Yeah, that's cool part of it. At least from the representation of the final result, is. So that's also cool part of AI. You don't actually know what AI is actually thinking because they might not think in the same way as we do. as long as the output of the production is what you need, then we can assume, OK, they understand it. Because we are essentially thinking just like thinking on paper. We're actually thinking.
on behalf of the totally different silicon-based creature, which is totally different from us, carbon-based. So yeah, maybe there's something that we can never actually understand.
Jeremy Gilbertson Silicon based creatures and Carbon based creatures. We're actually reading a wonderful book called irreducible in our book club right now We're two chapters in and we're actually getting to the bottom of that. So Stay tuned listeners listeners for that. So one one last piece of why this matters is An accessibility to the technology I think that is that is really important. So AI is pretty expensive, right? You know other companies, you know Deep deep seek brought it down significantly over the last few months or weeks or whatever it is. But what you guys are doing provides a bit of accessibility to building models too, correct? Talk us through that a little bit.
Where FLock started
Jiahao Sun Yeah, to make it more accessible by people who are nowadays, we actually just announced and released the version that can actually run on a MacBook chip. it's like, the reason why like AI, actually, I'm so much in love with the whole progress recently years about AI and technology development is something that you thought doesn't really make sense maybe two years ago. It actually becomes very doable maybe in a year or two time.
Several years ago, we started to flock, we thought maybe we can only just run fine tuning locally. By fine tuning, we mean that we don't actually train the whole model from scratch. We just do a bit of inferencing and the final several layers. And then that can happen still not on your personal device, but on a decentralized data center. That maybe to the most can be the most efficient way of decentralized training. Then by time, we just realized, the chips are actually super powerful now, that's first of all. And secondly, the algorithms, the advancement of the research that we do and also the whole industry do, right? There are tons of ways that people are trying to tackle the problem of we have to have an expensive GPU, right? So people are trying to compress the model, basically making the model smaller and compress the gradient, basically meaning that making the communication between models smaller and even like optimizing the whole training process. So
It can actually host the whole training in smaller devices. And today it runs on chips. Maybe in the future, we can imagine something like your phone or a Raspberry Pi can actually run not only one, but a cluster of them, Can run a large-density model. And similarly for the DeepState case as well, that's also very, very inspiring as well. Like nobody have ever thought that any, you know.
let's say second or third tier company nobody heard about before that can use a very limited resources of the GPUs that they have to actually have a similar quality model compatible to the OpenAI ones. yeah, mean, before DeepSeek, I thought it's basically impossible to happen. But now, like, yeah, it happened. And actually, that gives every other company
Jiahao Sun know, alongside OpenAIR, a good signal that actually we don't have to brute forcefully just put all the resources all together. There are tricks that we can do. There are ways that we can actually reduce the cost and still make the high quality AI. So it's amazing. It's amazing age that we are living right now.
Chips got cheaper, models got smaller
Jeremy Gilbertson So it's like this intersection of the chip and the hardware cost coming down just a touch and being able to do more with less, which what technology always does, but also compressing the models and like you said, compressing the gradients, which I want to kind of click on that for our listeners too. That's the interactions between the models that generate the good stuff we want to have coming out of them, right?
Jiahao Sun Yeah, yeah, because the model when the model created, it's like a standard model, right? But then we want this model to work towards your way. So we can think of model as a like small baby. You want to it to the way that you want it to be, right? So you need to teach it. You need to teach the baby different languages, stories, and then it will repeat to you in the same way. So whenever there's a wrong doing, you need to reinforce it to tell him, okay, you need to change it, make it better. So the way that you communicate with this baby is basically the gradients that passing through to the model to make sure, okay, you are going to the wrong direction. I'll pull you back a bit. You are going to the correct direction. Yes, I'll give you more incentives. Go straight in.
Jeremy Gilbertson So gradient, gradient is parenting.
Jiahao Sun Exactly, yes. But then for the large-language model, that means a huge effort on parenting, let's say, right? So that will result in the data to be huge. And then traditionally, if we think about decentralized model training, if you have to send over this huge parenting order book over the internet, that's super, super heavy, right? It's not efficient. So that's why people... yeah, okay.
Jeremy Gilbertson Like Apocalypse Daddy. Mark has a book on parenting, yes. Sorry to interrupt.
Send the change, not the data
Jiahao Sun Yeah, in other words, yes, exactly. So that's why people are thinking of different ways to make it lighter. For example, I don't have to tell you everything. I just give you some very light comments and then in your end, you decode it well to make sure you understand the full command of it, something like that. And there are also people trying to make it like what we call a collaborative layer where the instruction is just like a textbook and everyone has that textbook.
so that I don't have to send you everything. I can just send you some very light codes. Turn to page 127, that's the order. So you read the order yourself and then that becomes your grade. So yeah, tons of different ways people are trying in order to lower down the cost that we have to communicate between different nodes or different models. So that makes things even more efficient, yeah.
Mark Fielding How efficient can it get in the next six months?
Jiahao Sun the next six months. I would say six months maybe a bit early, but you never know, right? But in my theory, so today, let's say an Nvidia 4090 card is great, right? Great for games, of course. Also great for motor training and inferencing, right? I think that one in some frame of about two to three years will be equally powerful maybe to the phone that you're having today. So if you think about the...
but there's some models that you can run on your graphic cards today may just be as accessible and as easy as your phone in about two years time, then yeah, that's the direction all the startups are trying to build up for. And especially within that two years, there might be some interesting miracle happening in the technology side itself. only, as as Jeremy just mentioned, not only just the hardware is getting powerful, but also the algorithms getting light-weighted, getting more complex.
and get more efficient. So yeah, we'll see this actually coming maybe even sooner than two years time.
Mark Fielding I want to get into flock, but before we just get into flock, what does all of what you've just said mean for these giants of AI for open AI for grok for Gemini like this shrinking down of the power and the compute that you need to create these language models. You're, speaking about the the sharing, I don't know how up-side it went, but what does it mean for these big mega-lists? Are they going to be victims or successors from this?
Jiahao Sun think still a general model that's been a foundation model that's been trained by such superpowers are still important, right? If we are talking about open source models, right? Because it's great foundation for all the decentralized scenarios to train from. Because if you imagine everyone had to do something or everything from scratch themselves, then that's a bit of pain. yes, for superpowers and Google, Apple, sorry.
not Apple, Google OpenAI, Microsoft. sorry. Yeah, it's great still to have them, especially for those who are contributing to the open source world like Lamar. It's great. It inspires lots of the challengers in the industry, which is way better than having their own just closed source model. So yes.
It's still very helpful and I see in the future the decentralized players still have to rely very much on the open source contributions, but then they will have their own alter or specified models onto each verticals in different industries, in different scenarios to make it more relevant to the use cases.
The risk of handing your data to an enterprise tool
Jeremy Gilbertson Makes sense. Let's stick with the big boys for a second. So, I've done some research. You've been pretty vocal about, you know, AI being very centralized, being too centralized. What, what is, what is broken about the current system and the current things that are happening? We don't have to, we don't have to look at each of the big boys specifically, but just in general, like what is, what is broken with the system in your mind?
Jiahao Sun Right. currently, there's actually a very live case. I used to work in RBC, Robb and Canada, as their global head of AI for the past 10 years. I know that the first maybe several... Sorry, the first time when GPT came out, all the banking industry banned the access to it because they can't risk...
their own financial data or internal data to be exposed to those centralized players. Even if they sign some legal agreement, say, okay, we're using you as a vendor for it. But because many of those data, can create such insightful derivatives and it's hard to trace, right? It's very risky for those banks to risk it. And similarly for healthcare industry, this turns out to be two very interesting industries for FLock to be FLock's clients.
So yeah, that's generally the case, but I think maybe it's too far away, not so related to everyone. I can bring up some very related examples. For example, you want to have your personal assistant, then that will require you to upload all of your chat histories with everyone so that they can understand you better, because otherwise it won't be relevant, right? But that will include your ex and some secret friends, so which you know it's going to be problematic if one day someone actually revealed it to the public.
Jiahao Sun So right, so there's always that risk that you are actually giving out that you don't want to. yeah.
23andMe, and what happens when a company dies
Jeremy Gilbertson So here's a very applicable thing. Mark, we talked about this at the very end of Nexus when we were finishing Harari's book about AI and humankind and all of that. The idea that future leaders, there's so much information on the people that are going be the future leaders right now are like 10 or 12 or 14 or 15, right? So they have this tremendous chat history. So if that chat history goes somewhere, and someone owns that chat history, they kind of can own that person in a way, which is like really scary to think about.
Mark Fielding Does that include signal chat history?
Jeremy Gilbertson fair enough. Boom.
Jiahao Sun Yeah, yeah, there's actually a quite live case right now about 23andMe because of their financing issues themselves, right? They have to sell the company to whoever to give the price, but then that includes that data set of almost everyone who tested, including me. I was like, no. Yeah, yeah, so your gene is of course very important and your private history, private data, photos, chat.
chat history is maybe equally important because your gene from bio perspective can maybe clone you and your personal data from AI person can clone you as well. So yeah, it's the same thing.
Jeremy Gilbertson I told my wife the same thing as soon as I heard about it. I'm like pull your stuff off delete your account like you know, because you're right. You're right that we had it we had tools for humanity on a couple of episodes ago and they used the the iris scan to create, know, personal identity trackability accountability kind of thing and they said the only thing more identifiable than your iris is your your genome, right? So people are like they don't think about that. They think about I got this cool chart and I know that I'm you know, 0.01 % Norwegian. I didn't know that. That's cool, right? Um, all right. I'm riffing on too long. Let's talk about federated learning. Cause I know this is one of the pieces of the puzzle and you mentioned banking. You have a history in banking. I've got a banking analogy. I want to throw your way and tell me if I'm right or wrong on this. saw this as I was doing my research. So the idea of federated learning can be broken down by a bank who does loans in three different sectors, say auto home and business, right? And you have these these three different sectors and three different data sets, right? So it's basically to prevent the data from the auto, the specific data, the specific information from the auto side interacting with the home side. But by pushing them all in together in this federated learning process, you get the benefits of the data working together without seeing the actual information. Is that close at all?
Where federated learning actually gets used
Jiahao Sun Yes, exactly. yes. It primarily being used in many of such use cases as you just mentioned, where different desks, can't mess their data up due to regulations or due to many of the reasons, right? For example, a trading desk, I'm trading the MEA equities, right? While the other desks is trading the Asia ones, which are not supposed to actually share your information across.
or I'm treating one client, you're treating another and we shouldn't cross our information. across the firm, let's say across different firms, it's beneficial for them to have a macro large-range model, let's say, that can give them great insights into the maybe daily news of today or what should we expect from the news, from market change.
by doing so that you have to input some of your internal data. And yeah, so that's exactly what federated learning is doing. Or we can put it even easily. It's not something new. actually very, it's been developed by Google in 2017, where in your input software on Android, all of your typing habits have been trained by traditional federated learning ways, meaning that your data
The raw data is not sending back. It's only the gradient, as we mentioned this term again, right? The gradient of those changes are actually sending back and aggregated between different users. So you don't need to worry that your data has been exposed to another Android user or Google, right? But then still, it's centralized, meaning that if Google wants, they can actually get your data because nobody knows that they're not doing it. And there are actually cases.
of a software company called SoGo, an input software company called SoGo who claim to be doing federated learning but then actually being exposed that they are sending back the raw data because it's cheaper, right? Nobody's regulating them. Why don't we just send the raw data? Because nobody's actually chasing us and oh, you have to do this, you have to do that. That's fine. You tick the terms and conditions box when downloading the software, which means that you forfeit everything already.
Jeremy Gilbertson And that's why they make the terms and conditions document like 64 pages in very small font. like if I use, I use Google docs to do a lot of writing and a lot of my research, it's not really mine, is it?
Jiahao Sun in some sense, yeah. I never checked that. So maybe you should read that Terminal Conditions.
Jeremy Gilbertson Yeah, yeah, maybe I should, maybe I should. All right, so.
Mark Fielding What you're doing is not.
Jeremy Gilbertson What's that Mark?
Mark Fielding What are going to do if it turns out that you don't know, which it will turn out?
Jeremy Gilbertson Going back to thinking on paper, Mark, thinking on paper, pulling it out of the digital ecosystem. Yeah. Yeah. I'm taking my ball and going home. Flock it, right? All right. So let's, let's talk about blockchain. We talked about federated learning a little bit. And then after this, this little bit, we're to go straight into what flock is doing. So with blockchain, there are, there are tremendous benefits, but there is additional friction, in the UX.
Jiahao Sun I like this. That's punchy, that's punchy, that's punchy.
Jeremy Gilbertson for users, right? So how do you balance the friction and in the user experience with the benefits of having data protected and all of that good stuff?
Jiahao Sun For us, it's easy that we just deploy the retail apps so the users don't have to understand everything around it. They can just use an app and then they will see clearly from the upload string and also from the blockchain logs because it's open to everyone, right? That their data is never sent out to any third party. yeah, so basically federated learning, blockchain, that's the reason why we call ourselves FLock. So federated learning and blockchain. So the reason we put blockchain here is just to replace the role that Google used to play for their Android users, or Apple is using for their own iOS users. So we have an open ledger. So we can see clearly who's actually playing EVO. And we can see clearly, OK, there's no one who's actually sending back my data to a third party server. So yeah, that's part of it. That's being orchestrating and governing the whole on-chain process when it comes to federated learning in a decentralized fashion.
And FLock as a company, we don't have to sit in the middle for every of the clients that we treat with, right? We just leave them on to the protocol. So that's the beauty of blockchain that we think is super helpful into this decentralized AI industry.
A philosophical choice
Mark Fielding Okay, so this is a, it sounds like a philosophical choice. You're not, you're not using blockchain because it's necessarily the most efficient way to create a product for the customer. This is the philosophy of Flockit to use the blockchain to keep third parties out.
Jiahao Sun Yeah, kind of like for all the technologies, right? Centralization will make it super efficient, of course, but then there'll be lots of flaws that you can't control, right? While in a democracy type of way, then things will be a bit slower, but at least you can make sure there's no wrongdoing in the whole process. So yeah, that's generally the decentralization versus centralization type of arguments, pretty simply, and why people want to do that.
What blockchain makes harder
Mark Fielding What does it make more difficult for you? The blockchain aspect of flocks.
Jiahao Sun Oh yes, in obvious ways, the efficiency of course, because if everything is just trained by sending all your data to a server somewhere, then it will be super quick, But then when it comes to decentralized training, the efficiency slows down a bit. But then that depends on different companies' strategy. They will have different ways to make it quicker by...
as I mentioned, compressing the data, compressing the gradient to make the communication more skipped. So you don't have to communicate every different epoch of the training. You can do maybe 10 epochs each time or 100 epochs each time so that the model can still converge but without needing to send in too much communications, which save a lot of time. yeah, lots of tricks have been done by people.
Jiahao Sun Recently, by researchers, recently even in media, I think they acquired many companies for the past several years for model compress, because they also believe in that by the time that they develop better graphic cards, they also need some techniques to make sure that the models can be compressed enough to host on their graphic cards. So they are trying on both ways.
Jeremy Gilbertson Who can get the fittest and the fastest? it the chips or the algorithms, right? And the software, right? It's kind of this balance, but hopefully they do it together. Well, let's talk about, let's talk about FLock specifically. And in our research, we found that you believe that FLock will democratize AI development. Can you explain that in simple terms to our audience without any kind of tech jargon and
Jeremy Gilbertson you know, what you feel that, yeah.
The glucose app, and how FLock works
Jiahao Sun Yeah, as I just mentioned about how Flark works, right? It's federally learning just like traditionally, but with blockchain for the orchestration and incentives, of course, naturally come from the web industry. So for every task on our platform, it will naturally just be a task raised by anyone. For example, I'm a medical company. want to do a...
glucose monitoring application, I want an app, but I can't send my client data to any third party, right? So I actually raise this as a task on FLock, and then the engineers on platform and the patients on the platform will jointly train the model without needing to send their data over. So that will train as a task on our platform, and it's been connected, all the nodes being connected by...
by blockchain. So the ones who contributed the ones who contributed either the gradients, either the model, the parameters or the data, they will get their share of the incentives by the time when the model was built. So that's generally how the model works. So someone raised the problem and then we put it as a task on the platform.
and the engineers, patients, or the users will join to make the model better. And once this model is better being deployed into the retail app, then people can just use it and grow it.
The museum analogy
Jeremy Gilbertson Let's kind of go down your analogy a little bit further. So say Mark and I are on the platform and we look and we see this task related to diabetes research or blood sugar monitoring or whatever it is. And Mark and I both are patients that monitor our blood glucose and we have an app on our phone, right? And we decide, okay, I want to jump in and I want to help this because...
Jeremy Gilbertson you know, either I believe in it or whatever, I would send my information up and Mark would send his information up, but it goes into this repository where no one can see Jeremy's blood glucose on March 27th was 80 and Mark's was, is that kind of where we're headed?
Jiahao Sun We'll actually know the other way around. It's actually when you open the app, the app that supports FLock, Your data stays local with your iPhone, so you don't have to worry that it's being uploaded to anywhere else. But then the model, a small piece of the model will be downloaded onto your phone to train your data. And it will get that gradient again. We mentioned the textbook of how the model should grow. They will send that change of the textbook to the other guy who holds the original book.
and then take your change of the textbook to change his own textbooks. And okay, now the book is a bit different and you train on my own data. So my book will change a bit. So that change, I will send over to the next one. So actually you are not even sending the textbook at all. Everyone have a original book and then we train our own data and we get a different way, a different shape of the book. And then we just send over the Delta, pass on to the next one, next one, next one, until we all converge on a book.
Jeremy Gilbertson Okay, so instead of me pushing my data out to you guys, you're deploying someone to learn. Not someone, but let's just personify this thing. Yeah, local model to jump into my phone and kind of be like, all right, let's see, how does this compare to what we have up here? Okay, here's the new piece of information that fits the gap. I'm gonna leave everything else. I'm gonna take that new delta, like you said, up and train the model. Is that it? Interesting.
Jiahao Sun Yes, exactly. And that can be super small because it's only the delta of the change.
Jeremy Gilbertson So, but you're still allowing access to something on your phone though.
Mark Fielding Is this like an AI? Sorry, Jeremy. Sorry, Jeremy. Sorry, Jeremy. Is this, I'm just having trouble following that. So is this like a alternative almost to ZK proofs and try extracting the information that's needed without extracting all of the information, just one little bit, which will give the whole picture without giving all the data. Is that what, okay. Thank you.
Jiahao Sun You can even think of this as embedding, but you can't really say that. yeah, like it's embedding your original data to make it into something that's actually not your original data, but it can be useful for others to learn from the model. So they make it insights of your data if we name it that. Yeah, exactly.
Mark Fielding But rather than a ZK proof is sending something out, you're inviting something in.
Jiahao Sun Yeah, yeah, which is a model. Yeah.
Jeremy Gilbertson So instead of going to the museum and taking the old book that's on display and then chucking it into this machine that's going to make more books or more things based on that book, you're sending someone to look at the book, kind of review it, and see if anything's different based on what they already know and then they leave the book alone.
Jiahao Sun Yeah, they leave the book, run away to another museum and then just write down the changes and the other guy will check it out. okay. There's a change. Yeah. Yeah. So you don't have to borrow the book from each different museums. Yeah.
Jeremy Gilbertson And then there's the reward and then there's a reward mechanism, Right.
Jiahao Sun Yeah, it is a reward for people who doing so. Yeah, yeah, you're contributing, you're helping. Yeah, incentives. Yeah, that's a yeah, that's also another big perk of using blockchain, which natively comes with incentives and it's fair because all the con all of them are countable on the chain, right? You can see clearly who has been interacting a lot with that model, who's not. And then, yeah, there'll be a share by simple calculation that you will know how much you should get. So yeah, that's another
Mark Fielding yes, the incentive. Let's get on to incentive.
Jiahao Sun piece of the democratization that we are mentioning because currently you are actually helping OpenAI to train their model but you are paying them 200 bucks. So it's not even fair.
Mark Fielding Well, okay. Yeah, I'm paying them. Okay, so one of my questions that I didn't want to word it so, so, what's it off so ugly? That's what's the adjective, the admin with that. Obviously incentives, crypto, tokenization of the blockchain. So these are the incentives. What role does the FLock token play? You mentioned, okay, if you want to use the
200 euros a month, think that OpenAI are talking about 2,000 and 10,000 and $20,000 subscriptions. Who knows what's going to be going on there? How much can I make? What's my incentive on FLock?
Jiahao Sun right. So that really depends on how much incentive the task creator is actually putting down. So for us, the whole blockchain mechanism is through the stakes, meaning that, for anyone who want to join to train the model, want to join to earn incentives, they have to put down a stake of their tokens to make sure that there's no wrong doings. For example, if there people trying to point us the whole model, right, because it's open, then if I'm just just a devil, right? I want to join the training. I just want to put some random data over and just maybe steal some data from others, right? So I'll be penalized and then I'll be slashed out for the training. But then for the rest of the honest joiner trainers, they'll get their incentives share of it. And then the whole pool will come from the one who actually raising up the problem, saying, okay, the one who say that I need a glucose monitoring app. So they will put down the incentive, say that 100K, 200K worth of the token to to bring it as a bounty on the platform, right? And people who saw that bounty, maybe it's larger or smaller, then they'll be incentivized or interested in, this is a huge bounty. I need to try, I need to join it because I know I can get a higher emission out of it. That's actually it.
Jeremy Gilbertson Here's something I just, when we started talking, you mentioned the idea of paying $200 a month, but also you're helping that thing that you're paying $200 a month. I just had an image of, you know, I pay a hundred bucks to go to the gym and I go to the gym and I work my ass off and nothing happens to me, but some other guy is getting bigger and stronger.
Mark Fielding That's what it feels like when I go to the gym.
Real world use cases
Jeremy Gilbertson Fair enough. Fair enough. Let's, let's talk about some real, real world use cases for flock. We understand how it goes. We also understand the why behind decentralized AI and the benefits of it. Let's talk about some real world use cases that you're super excited about some recent ones, maybe.
Jiahao Sun yeah, yeah, so we've been focusing on the business enterprise use cases in the past a lot, like two of the London hospitals, especially for glucose monitoring and the eye clinical ones. And then also the trading desks, as I mentioned, the financial use cases in trading, right? Because different desks, they can't just share information. So yeah, we've been working for such clients already, like...
to pop them ones like GSR and some of the bank web three protocols like request finance Recently, we're trying to push more retail applications, which you will see that announcements very soon As soon as the next next week actually so for the retail apps so for the retail user to get more Feeling of how flock works works on their own On their own phones or on their own MacBooks, especially now we have the version of flock application. Just test run with our top validators and then we'll push it very soon to all the Mac users. They can then use Mac to choose different tasks to join. So yeah, many of this.
Trading desks, unpacked
Jeremy Gilbertson Can we unpack the financial one with a little bit more detail? We don't have to share names or anything, but what specifically are they doing on FLock and what are the outcomes?
Jiahao Sun Yeah, so there are different parts of it. There are trading desks of it, and there are also the credit scoring part of it, which Jeremy actually just mentioned. For the credit scoring, simply that in many cases, users don't want to share, especially in the Web3 dynamic, right? People don't want to share their wallet addresses because that means you can trace many of their source of income and also the level of wealth that they probably have.
So we have FLock as a personal node for each of the users collaborating with us. So you don't have to send your wallet address or everything to a third party, but we can still calculate a credit score for you so that that score can be used as a verification for a lower collateral or a lower rate of the interest maybe when you do lending. So that's one part of it.
And the other part for trading desks like GSR, the top trading firms who's actually working with us, the trading algorithms that we're providing with them together is the ones that they don't want their different desks to share the information because for maybe for a certain token that you can't see the information from the other token, it's not even legal, right? So you have to make sure it's well segregated.
But still, we have the macro of the trading mechanisms that's being provided with insightful information or signals for their traders to make executions, or even for their trading boss to make executions.
Jeremy Gilbertson So there's like, there's a bit of a layer that you guys are the layer that kind of lives between the users of said platform and then whoever's performing, whoever's offering that platform to the users and you are this intermediary segregation layer maybe.
Jiahao Sun Yeah, to prevent any authority to actually holding all the data in one place, which is very dangerous.
Mark Fielding Yeah, very dangerous. After that, we've looked at the finance side of it. Could you impact the healthcare side of it? Because I think a lot of our listeners understand finance and crypto and Web3, but when they talk about AI and blockchain and then you bring healthcare into the conversation, especially the point of incentives, the healthcare part of this and how the incentives play a role in that healthcare scenario.
Baby scans, and the retail app
Jiahao Sun Yeah, so that's actually where we're to launch a very interesting retail app. Well, we've been collaborating with University College London Hospital and also North Island Hospital on virus research cases, right? But never really pushed any retail applications. So this will be the one interesting one you will see very soon. So we call it Baby 4D. It basically means that you have... So for those moms...
new becoming moms, right? They will have a scan of their ultrasound and then some of them in traditional hospitals, they are actually paying £100 to have a very naive rendering of the ultrasound image to a real baby face, which is quite fun. Like I don't have kids myself, but I...
My other co-founders, they just realized this is actually a huge opportunity, especially collaborating with the professors there, and they think, yeah, this is very interesting that we should do it. And this is exactly
Mark Fielding I could get like Ghibli Studio rendered versions of that.
Jiahao Sun Yeah, yeah, yeah. And this is exactly the case, right? If it's just your personal photo, maybe that's fine. Arguably, it's okay if you want to send over to a third party. under many regulations, even for the US under HIPAA, right? Your ultrasound data, that's healthcare data. You can't send it over, like just randomly to any third party. So yeah, so that's one of the use cases where I think the retail users will get it.
Jiahao Sun instantly, right? So now we have this healthcare data can also be used and tune the model and then to render the face for me. It's more in amusement, know, part of the, instead of the actual use because the face might not exactly look the same, but it's fun. It's fun for people to understand. And I feel how flock works, right? Without sending your data over, but we can still, you know, train a model together to make it the facing face rendering better.
Mark Fielding So speaking to somebody who works in biotech the other day, and he was saying that the biggest hurdle or challenge that they face is this protection of IP, obviously, and they want to share their research, they want to share certain results, but they also want to protect what, obviously, their IP, because in this it's worth millions, hundreds of millions, maybe even billions of dollars. And is what Block doing with blockchain and AI a possible use case?
Jiahao Sun Yeah, exactly, exactly. yeah. So by not sending those raw data over, yeah, now you can protect a lot of those IPs already.
Jeremy Gilbertson We also have another budding business opportunity built on humanity's lack of patience. You get back in the day where you just waited till the kid came out to understand if you were gonna have a boy or a girl and then you could do these tests to figure out if you have a boy or a girl and you these big reveal videos. Now you can see the kid. It's brilliant. I think it's cool. I think it's a cool idea to kind of envision on that. Let's...
Mark Fielding Well, actually, sorry, Jeremy's gonna jump in there. We can cut this out. But what you just mentioned, I was thinking about my own parenting journey. Won't bore you too much with parenting, Albert. This obsession that people have with the next moment, maybe next week I'll be happy. like, why don't you just wake up and your kid's 10 and you just forget, don't bother with all the difficult years because when they're 10, it will be cool.
Mark Fielding I think that what you just said just reminded me of that, of this obsession with next, next, next, and people are forgetting just living them.
Jiahao Sun Yeah, yeah, yeah, and they're willing to pay premium to just see it a second earlier.
Jeremy Gilbertson Everyone likes the behind the scenes, the sneak peeks, the exclusive access to baby geometry faces earlier. Well, let's, let's dig into the, the, not necessarily the flip side, but let's explore the idea when things become decentralized, centralized authorities tend to freak out a little bit, right? Because one of two things they start to see their potential control over something easing, but
There's also on the flip side a decent argument for if something is truly decentralized, how do you make sure people are using it the right way? How do you make sure it's not turning nefarious and those sort of things? So how do you guys think about that?
The safety layer
Jiahao Sun Yeah, the safety layer of it. guess, end of day, there will definitely be some kind of verification layers or verification mechanisms being lived on such decentralized AI protocols. Just like nowadays for building airplanes or large machineries, there's always a verification department of it, while that code base has been sitting there for decades.
You just need to pass all such verification processes in order to build your own robot or your own machinery. But of course, there's also drawbacks of it. In some cases, those code might be too outdated. They need to update or many of the the tricks of the hacks that they can maybe trick the results.
and then to make the verification looks good, but then still they will have their drawbacks or flaws that's been putting their own purpose. yeah, there's always challenge on such things, even for centralized models, right? There are people trying to reverse attack such models to make us say something that's not supposed to be saying. Yeah, and even for centralized ones, we've seen many cases where people are trying to hack it.
So I think it's just there for a bit centralized or decentralized models. It's just there. There's just been some bunch of people always trying to hack it. Yeah.
Jeremy Gilbertson Well, full clarity to you know, the larger models, right? You know, there's not a whole lot of governance or accountability happening right now with, you know, how they're training their data, how they're doing stuff either, right? So, but what happens is here's the spin is where like, these guys are decentralized, nobody's regulated, nobody's watching these guys. But I would argue since they're decentralized, they're part of an ecosystem, a community that's all has a vested interest in that.
Jeremy Gilbertson Wouldn't there be more accountability and oversight in that?
Yeah, yeah, that's a good point of it. Thank you. And I think so. Yes. That at least those models are being built by a group of yeah, people share similar interests. And the technology itself is used by them for. Yeah, for the sake of technology itself, right. So it's not. Yeah. Yeah. I for that. Yeah, for that. I'm super supportive.
And we've seen some different opinions on the authority about such similar situations as well. For example, the tornado cache issues right before it was an issue, but then it was being lifted to treat it as a technology itself. of course, I mean, for obvious reasons, some legit people don't actually use it. But the technology itself is less being chased by SEC and DOJ anymore. yeah, similarly.
Jeremy Gilbertson Shall we cast to the future?
Five to ten years out
Jeremy Gilbertson Let's, so where we're at with everything right now, centralized models, the starting of this kind of decentralized model approach that you guys are leading at FLock, what is this all gonna look like in five to 10 years?
Jiahao Sun Five to ten years, that's long. As I mentioned, everything happens so quickly in this industry right now, in Asia that we're living right now. So five to ten years, there can be many things happening already, like the quantum computing, maybe it's one part of it as well. So things might change again. Some of the traditional cryptography might have to change to something more resistant to the quantum.
attacks and also the computer itself, which we still haven't seen a very obvious blocker of the compute being growing. So it's just getting super crazy in the next following years. And also for the optimization or the algorithm side, I mentioned also before, tons of researchers, they're all trying to just bring down the barrier, the obstacles for people to actually...
access to those models. So tons of different tricks being published out. And it's actually one of the hottest topics in the top AI conferences. if you nowadays, if you publish a paper saying, I can have a better model by just piling up 10 times whatever OpenAI is holding, right? That's not going to be charming. It's just brute force, right? But then if you have a paper talking about, I'm using actually just one tenth of whatever.
of an ice holding to have a similar quality algorithm or model that they're building, then that's actually a phenomenon. So that's a trend I'm seeing on both ways, lower down the barrier, increasing the compute, then things will happen in all edge devices, maybe. So we can host such high quality data on the edge devices, not only just maybe computer devices or your phone, MacBooks, but also something like your wearables.
Jeremy Gilbertson What if we, what, what if we speak a little more, just real quick, Mark, what if we speak just a touch more tangibly on, on what is there an inflection point, a lever that you see happening in the next, you know, in that same time frame, five to 10 years where people will start to be more open to this decentralized approach versus centralized aid. Do you see any kind of major inflection point in there? If so, when?
Jiahao Sun I think for example, Apple is doing it in their own intermediate way. They have this Apple intelligence, but it's living on the private cloud, right? So it's definitely not good enough. It's still on their Apple cloud, iCloud, sorry. Just some way that they claim that cloud is yours, but no, it's not. It's messed up with all our others in same data center, right? But then yeah, the moment when that can happen actually locally on your phone, that's the moment that I can see.
such next level growth of the decentralized models.
Can hallucinations be removed?
Mark Fielding Could I get on such an impossible to answer question, I think in AI, but can I get your opinion on something which may or may not be in your ballpark to answer? Maybe for some of our listeners have experienced AI hallucinations and I've definitely, definitely, definitely noticed a huge improvement over the last eight, 10 months in hallucinations. can't say more about that.
Mark Fielding But I was reading some interesting Twitter conversations this week and maybe I'm just not using AI for a deep enough, technically profound enough use case to get the hallucination. So if people are using it for much more in-depth technical stuff, they get the hallucinations. I just wanna read a tweet from somebody, quote, I sense that a lot of people now think knowledge graphs will fix the LLM issue.
the issue being hallucinations, but no, they will not. They cannot, even in the case that knowledge graphs would prevent logical inconsistency 100%, there are a lot of text constructions that are perfectly logically consistent, but have zero relation to reality. So could you unpack that quote for me and beyond that talk about hallucinations for regular people using AI where we are?
Mark Fielding how we can remove them completely if at all.
Jiahao Sun Yeah, so that's quite an interesting question. Thank you for bringing this up. Actually for past four or five years of being a research fellow at Imperial College London, my original research focus is actually in graph networks and graph neural networks, neural graphs. So it was once a very high topic while people were very much interested in how to build a a great graph that can do all the culturalities. So once you have a question, we can just source back to see the culturalities and give you the answer. But then just overnight, GPT came out and nobody did it at all. I have two PhD students who just then had to change their thesis to make it more to date with a new larger language model, transformer type of technology.
technologies instead of focusing solely on the knowledge graph part of it. that actually just trying to give you an example about looking back to knowledge graph might go nowhere because from experience that we saw, let's say 10 years ago about deep learning came into power, right? There are still a of people who are trying to do a possibility analysis or probability analysis on the statistical level of the models, right?
But then where are these bunch of people? You don't see them except for some research papers they're still publishing things, right? Because in terms of applications, people don't not don't really, but people care. then because of the growth of the compute power and growth of the algorithm by piling up the compute power so quick that the causal reality analysis of such things can just cannot catch up.
So in the beginning, when deep learning becomes a thing, many of the hedge funds cannot use it as a proof for giving the alpha or explanation of the alpha to their clients. But then over the time, they give up, say, OK, it's fine. Let's do a black box back testing and give you few results. And that's just our AI deep learning alpha finder. And yeah, believe it or not.
Jiahao Sun similar things will happen for the LM stage as well. many people will claim that, graph neural network can solve the issue. But the thing I see is, this pace of developing graph neural networks or graph knowledge graphs will not be compatible to the growth of the brute force large learning model, large learning model, basically, because it will just go crazily and then go super large while it gets getting more explanationally harder for a knowledge graph to be constructed based on that newer model. So basically, the speed of the growth of the two fields just cannot match each other. So you can't really rely on the other to do the verification check for the other. yes. So theoretically, it works. And theoretically, yes, it's a great solution. But just practically, we won't see the two two sides of the match, know, be matching each other. You will see this as an example in the finance industry, there's only one company run out with their knowledge graph, that's Bloomberg, the only one that have enough data to treat the knowledge graph. Rest of it, even the second in the market, I don't wanna name the name, they failed, the project failed. I was the one who was dealing with them for the project, totally failed.
because not enough data, not enough research, not enough even computer storage for them to grow a similar product compared to the other one who's in the top in the world. Similarly, now, same age of the LLMs, and if you're expecting someone will come up with a knowledge graph that can actually facilitate or verify the large language model of the next day, to me, I guess it still will fail.
and people still get used to large-scale models, even if it's black box. Yeah. Yeah.
Mark Fielding So this tweet is correct then, or her assumptions are correct. Where does synthetic data fit into everything you've just said?
Jiahao Sun Yeah, that might change. But then the question is whether the data distribution in synthetic data is accurate. Because by the time of 2017, the last moment of human-generated data died. Because that's where the first version of GBT came out. there's tons of AI-generated data on the internet right now.
Again, there's a bunch of people who are just, bang, right? On Twitter, they create alternative histories by using large-amaged models. Now, who's the 47th president of the United States? They would just put Elon Musk, and then they'll put tons of resources to back such altered history. So think of one day in two years later, right? There's a new AI model to be built, and it's just extracting data from the internet.
for example, on Reddit, but they see more data talking about Elon Musk as a president instead of Donald Trump, then they'll put that as a ground truth. Damn, there's more information about it. So when we think about synthetic data, yeah, it's been polluted already, right? So you can't be 100 % sure that synthetic data that you are creating is the actual data that we're presenting, the human-generated data. So it can't just be altered already or polluted already by...
Jiahao Sun the AI that we created right now. So yeah, yeah. Yeah, yeah, like it's just no way back. Just like early days when people launching their satellites, we think it's fine. Elon Musk did a whole bunch of satellites around the whole globe, right? But then we realized it creates tons of garbage around this near space around us, but there's no way that we can clean them now.
Mark Fielding It's a vicious circle that leads to a dark place.
Jiahao Sun stem, stem thing here. So it's being developed. The area has been developed already and the pollution has already started and it's actually getting larger. And yeah, there's no way back.
Mark Fielding I just appreciate time. It's such a fascinating conversation that in your mind, that example of Elon Musk is president even if it's by proxy, but the Twitter masses confuse the future AI and the future AI believes that Elon Musk was the president in 2025, not Donald Trump.
Jiahao Sun how did we not get there? Yeah, yeah. So back to the question, I think Jeremy mentioned, brought up earlier, right, the verification protocol. So either the real data work of marks that we can put through to all the real generated data or ways that we can preserve the private data, right, not to be exposed to the public domain by using technologies like flock. I got it.
So yes, like some of your personal chat histories, never actually send it over to iCloud, right? So there's no way an AI can actually get those data and then generate some alternative history for you because you never actually brought it up. So there's tons of this too. Yeah, sorry.
Jeremy Gilbertson So instead of.
No, no, no, you're good. So friends and neighbors, instead of just chucking your data into these models, just allow, yeah, just say, flock it and just allow one of these little models to jump into your data, figure out the differences, and then just report back on the differences. And it makes it a little bit more palatable. And actually, you could get some money from it too. So who doesn't like that? Let's try to land the plane a little bit here, guys.
If someone listening, Jiha wants to get involved, wants to try this, hasn't really messed with blockchain much, hasn't really messed with AI much, what is a way that they could tiptoe into this and be like, what are these guys doing and how can I play around with it?
Jiahao Sun yeah, follow, well, just go on to our website, flock.io, and everything is there. Twitter, our blog, and also the tools that you can use right now to train the model, to join the training. Yeah, everything's there. So yeah, flock.io. Thank you.
Mark Fielding awesome it's yeah flock it I love it
The carryover question
Jeremy Gilbertson So let's go through hot buttons and we'll hit the carryover question, Jihao, and then you leave us a question and we'll give each other a high five and be about our day.
Jiahao Sun Yeah, I... Okay, so the question for the next guest, right? So as I know that the one is doing some of the space things, actually I have the question just I brought up earlier, right? So about all the garbage around the near space of the earth, right? Is there any solutions that we can actually solve that or we just have to leave it? And for the human beings to grow, technology to grow in the future, right? We will definitely have to...
launch a lot more satellites in the future, right? But then will that just be super jammed around the near space? And do we need to charge some congestion fees for that? Sorry, kidding. But yeah.
Mark Fielding I like conjecture. I don't mean to be a party pooper, Jeehow, but we had Philip Johnson on a couple of weeks who's the founder of Star Cloud and they're using Falcon 9, they're putting data centers in space. And we asked him the question about space junk and he was very dismissive. It's not a problem.
Jiahao Sun Is it not? Yeah, that's the thing. I don't know. I don't know. I'm not expert there, but from the industry I know, right, people who already put pollution of AI, AI garbage is, and that's already been polluted the whole industry already. So I don't know whether it's the same case for near space around our earth.
Mark Fielding It's not that loud, I mean...
Mark Fielding Thank you. Do you want to do a hot burn show? Maybe we don't have to.
Hot buttons
Jeremy Gilbertson Let's, yeah, let's do hot buttons. I'll keep the timer on my phone, because you're going to do all that stuff behind the scenes anyway. So maybe tee it up for Jihao, let him know what we're doing, and then we'll hit a hot button edit.
Mark Fielding Yeah, we should do this. I think we should do this earlier in the interviews because the idea of this is just to humanise you, to make our audience think, yeah, not only does this guy know exactly what he's talking about, but now I know him a little bit better in his personal life as well. So five questions, one word answers, 30 seconds. It's the Daniel Kahneman thinking fast, thinking slow.
Think fast, there's no wrong answer, there's no right answer. just, it's nothing about AI, it's nothing about technology, it's just, it's about you.
Jiahao Sun Okay, sure. Like, okay. Yep.
Jeremy Gilbertson Alright.
Jeremy Gilbertson Three, two, one, go.
Mark Fielding Football or Mountains or beach?
Mark Fielding Rock, reggae or jazz.
Mark Fielding remote work or office.
Mark Fielding Apple or Android? Yeah, which is most likely AGI or aliens.
Jiahao Sun wow, well, nicer, but AGI of course.
Mark Fielding Thank you. Well done.
Jeremy Gilbertson And that's it. Clock is done. Good stuff, my friend. Good stuff. So we did the carry over. We did the carry over question from you. What was the question that was left? We have a question left from our last guest, David Bianchi, who's a Web3 film producer, actor, writer, director, and poet. This is more on creativity, Jihao, or actually just kind of getting
Jiahao Sun Yeah, cool. Thank you. Yeah, that's great. Thank you.
Staying out of your own way
Jeremy Gilbertson what you need to get done related to what you love to do. What are you doing? His question was this, what are you doing to stay out of your own way?
Jiahao Sun wow, stay out of my own way. Yeah. I think it's always trying to explore the boundary of your interests, right? Trying to do something that's out of your scope. Like I was traditionally a trend AI researcher engineer. But then I choose to work in a finance industry, which I have no idea about before I joined the company. And then I was so long years in a traditional industry, then I tried to explore into somewhere crypto, which is also a...
totally new world to me. So for me, it's like stay interested in new things and new interesting things and then to develop some crossovers that can actually bridge your experience from the past to the future.
Mark Fielding Awesome. Curiosity. Love it.
Jeremy Gilbertson Wonderful wonderful. Well, Jihao. It was a pleasure talking to you today a couple of quick recaps We've learned that AI is a bit too centralized We also learn to maybe say flock it and try to fix that with decentralized coupling AI and blockchain we also learned a bit about federated learning which is an interesting way to Put put your information into something larger than you without giving up your privacy, which I think is super cool and real-world examples are in place with what you're doing and
And we look forward to tracking those. More importantly, listeners, what do you think? Would you contribute to this decentralized AI model if it meant you had more control and privacy and if you were incentivized to it?
Mark Fielding they should, can I just say that listeners, every time you use AI, whatever model you're using, just when you click enter, just think, just a little bit, just think of what you're sending, where you're sending it, and what that means today and in the future. And when you do that, I think,
Mark Fielding that perhaps you'll start thinking about decentralized AI and you are important, your data is important. So think about it.
Jeremy Gilbertson Jihau, thanks for joining us. Really appreciate it. Keep us posted on the progress and hopefully we'll talk again really soon.
Jiahao Sun Yeah, thank you. Thank you, Jeremy Mark. Thanks for having me. Thank you. Cheers. Take care. Bye.
Jeremy Gilbertson Take care.
Lessons from Thinking on Paper
Mark Fielding I like, I'm gonna do these end of show lessons from thinking on page.
Jeremy Gilbertson I love it, Mark. I love that. These reminders to be human cautionary tales, like...
Mark Fielding We should write our own book of fables. We could do like a technology fables book.
Jeremy Gilbertson I love that idea. Little short, short stories with some morals.
Mark Fielding Yeah, a fable, exactly. I think the old one of the oldest books in existence is the Asobs fables. Let's rewrite that for the technological age.
Jeremy Gilbertson That's actually a great idea.
Mark Fielding Publishers, you're there.
Jeremy Gilbertson gosh. Alright, so we can edit this little bit out, but let's run into backstage, kind of tee that stuff up. These notes were great. I thought we got to be a little bit more mindful of time, I think. I was trying to kind of move it along a little bit. I don't want to move over anything, but I think that 40 to 50 minute window is kind of where we want to land with everything, but I could be wrong.
Mark Fielding I think that, so for example, last week you were definitely right. Who was before David Bianchi? can't remember. So for example, with Philip Johnson, it's probably not. I so I think, yes, generally speaking, but it was really interesting and I didn't really want him to stop talking about the hallucinations. No, today with Ji Hao I thought...
Jeremy Gilbertson Who, Philip?
Jeremy Gilbertson I think we lost Kylie on the last two bits if I'm being completely honest.
Mark Fielding Yeah, maybe we're going to edit it out, but like he seemed to loosen up and he seemed to enjoy it.
Jeremy Gilbertson Yep. I agree with you, hot buttons early. You can actually edit this, the hot buttons bit out and push it up front too.
Jeremy Gilbertson I can look at it too, I haven't done any of editing yet, so I can look at
Mark Fielding Once I get my new computer, I can. So what did you learn today, Jeremy? Quick take home for our listener. We've just spent 45 minutes in the presence of Jiahao Sun from FLock, and we've been talking about decentralized AI and the blockchain. What's your big take home? What should the listener be thinking about?
Jeremy Gilbertson think it's interesting to think about how you could you could actually like with your personal information how that could be harnessed and monetized right that's that's kind of an interesting thing we've been talking about this stuff since I've been riffing on open mustard seed if we're nerd out on all of that stuff I got to sprinkle that out there every now and again but also the I what's that
Mark Fielding We have to do a show on that. We have to do a show on open mustard seed at some point.
Jeremy Gilbertson We need to find the people that put it together. But no, think it's interesting, but I also fall into this whole trap of, decentralization is awesome. It's the idea of it's great. But I think most humans, at least right now, are going to opt for convenience and not worry about the personal data until down the road when that 17 year old wants to become the, I don't know, the prime minister of England.
but somehow sent all of its texts, all of their text information out. And now as a point of manipulation in the future, I don't know. I just don't see the convenience and security seesaw changing.
Mark Fielding No, not unless that incentive is a lot bigger. Everyone, so there's always been people who...
Rebells the wrong word, but question centralized authorities and people who don't care. And right now with technology, most people don't care. Like you say, the shortcut is worth, worth ever think and check my thinking in at the door. I don't need it. And then you walk into your house and you do what you need to do. Those same people would stop using this centralized tech if they were going to get paid a hundred.
dollars a day, $50 a day, whatever that incentive is. But until the incentive matches, until the incentive outweighs the advantage of the shortcut, I think you're right on math.
Google Docs, and the cost of switching
Jeremy Gilbertson It's got a long way to go, think. So I look at Google Docs, right? Google Drive where I use this reference. I write all of my shit on there. I literally use that for planning, for writing, for developing ideas, for writing pieces that I'm writing. And the idea that it costs very little if nothing. But I'm basically feeding all of my ideas into a language model for someone else. So back to our...
Mark Fielding And not just do, every business I've worked for in the last five years does the same.
Jeremy Gilbertson Yeah, but the flip side of that would be, know there's private, off the grid, similar tools to that, but they cost more. They're a little bit clunkier. You can't collaborate with other people as easy.
Mark Fielding yeah, people don't use them, do mind if we use this decentralized offline version of GDocs?
Jeremy Gilbertson Yeah, maybe there's something, maybe I could just get an approach where I, I, I land all the stuff that I want to keep as me in a spot. And I don't know. It's, it's a lot to think about. Like we have so much other stuff that we have to think about, you know, raising our kids the right way, working to get food on the table, like all of these different things. And now we're like, man, we want to add something to make it a little bit harder, a little bit more difficult, just to touch more. but I'm going to get like three.
three tokens for doing so, right? So the idea is there, I think it's going to take a mind shift, large mind shift.
Mark Fielding I think you're right. I've been thinking a little bit about trying to balance the ease and difficulty of life. So people make shortcuts because it makes it easier. But maybe they instead should force themselves to make things difficult for themselves. Maybe if I can make this more difficult for myself, not only will it be more enjoyable, I'll have more gratification because what things which are difficult are things which are hard to enjoy, endure.
are great to remember, whatever that famous quote is, which, you you earn your beer at the end of the day. So maybe you force yourself to be, to embrace the difficulty and then that will be a reinforce, like a self realising, that's not the right word, but it will become a habit to enjoy the difficulty and then you'll stop just using these shortcuts because you've trained yourself to embrace the difficulty.
Jeremy Gilbertson I think one of the most critical things to think about in life as a life skill is knowing what to offload to a shortcut and what to spend intentional time, hold space for. I think knowing when to say yes to this, what bucket to put things into, I think that's super important, but it's funny you mentioned that Mark in my huddle yesterday, as I wrapped up practice with my high school lacrosse players, we've, one team's doing really well. One team is kind of, kind of getting their butts kicked a little bit because they're new players. we talked about the idea of being in difficult situations and what difficult situations can do to, to, to help your perseverance, help your tenacity, help your grit. And there's actually science behind this mark. I'm going to nerd out real quick. There's a section of your brain, according to a few studies called the anterior mid cingulate cortex. And it has been shown under in conditions with adversity that that area of your brain actually gets better and you're more willing to lean into more difficult situations because that muscle, so to speak, has been encouraged.
Mark Fielding I believe it, I believe it. You should download everything you have on Google Docs though and get a copy of that because, know, I think Google search is maybe below 90 % now and I think thinking on paper exclusive, you're gonna wake up in a year and Google's gonna be down 20 % of Google search, of internet search, people aren't using it, they're just using these AI alternatives.
And all it takes is one disgruntled Google employee to close access to G-Docs or take all of your information off there, Jamie. All your life's work will be gone.
Jeremy Gilbertson great advice. I'm on it. All right, well there you have it. There's our overindulgent riff in exploration, but I hope you guys enjoyed the episode. We got a bunch more coming up. Kevin Kelly is down the road, which is certainly going to help us navigate the future of work, family, life, all of that fun stuff. But that's it for today. Be curious, stay disruptive.
Mark Fielding Keep thinking on paper.




