Space Summary
The Twitter Space Intuition: The Core Primitives hosted by 0xIntuition. Exploring the core primitives of intuition and trust within the Alpha Group space delves into the fundamental aspects that drive human interactions and decision-making processes. Intuition serves as a guiding force in navigating complex choices, while trust forms the bedrock of relationships and communication. Through a deep dive into the interconnection of intuition and trust, this space sheds light on how honing intuitive abilities and establishing trust protocols can enhance creativity, problem-solving skills, and overall understanding of human behavior.
For more spaces, visit the Alpha Group page.
Questions
Q: How does intuition influence decision-making?
A: Intuition guides and informs our choices based on subconscious patterns and past experiences.
Q: Why is trust crucial in relationships?
A: Trust forms the foundation of strong connections, fostering cooperation and mutual respect.
Q: Can intuition be developed over time?
A: Yes, with practice and self-reflection, individuals can enhance their intuitive abilities.
Q: What role do core primitives play in human interactions?
A: Core primitives offer insights into the underlying dynamics of social behaviors and connections.
Q: How can trust protocols enhance communication?
A: Establishing trust protocols ensures reliable and secure interactions in various settings.
Q: Why is balancing intuition and logic important?
A: Combining intuitive insights with logical reasoning results in more comprehensive decision-making processes.
Q: In what ways do core primitives influence understanding human behavior?
A: Studying core primitives provides a framework for interpreting and analyzing complex human interactions.
Q: How does intuition contribute to creativity?
A: Intuition sparks creative thinking by offering unconventional perspectives and ideas.
Q: What is the connection between trust and intuition?
A: Trust and intuition are intertwined elements that shape our perceptions, decisions, and relationships.
Q: How can intuitive communication enhance relationships?
A: Intuitive communication fosters deeper connections and empathy, strengthening relationships.
Q: What benefits come from developing intuition?
A: Enhanced problem-solving skills, creativity, and better decision-making are among the benefits of honing intuition.
Highlights
Time: 00:15:45
The Power of Intuition in Decision-Making Exploring how intuition influences choices and judgments in various scenarios.
Time: 00:28:20
Trust Building in Relationships Discussing the importance of trust in forming strong and lasting connections.
Time: 00:42:10
Balancing Intuition and Logic Understanding the dynamic interplay between intuitive insights and logical reasoning.
Time: 00:56:05
Core Primitives and Human Behavior Analyzing how core primitives shape social interactions and communication.
Time: 01:10:30
Developing Intuition for Creativity Exploring how intuition contributes to innovative thinking and problem-solving.
Time: 01:25:15
Trust Protocols for Effective Communication Ensuring secure and trustworthy interactions through established communication protocols.
Time: 01:40:50
Intuitive Understanding in Relationships Exploring how intuitive communication enhances empathy and strengthens relationships.
Key Takeaways
- Intuition serves as a fundamental aspect of decision-making processes.
- Trust plays a pivotal role in establishing connections and relationships.
- Understanding the core primitives helps navigate complex human interactions.
- Intuition can be honed and cultivated through practice and self-awareness.
- Trust protocols are essential for secure and reliable communication.
- Developing intuition leads to improved problem-solving skills and creativity.
- Effective communication relies on trust and intuitive understanding.
- Balancing intuition with logic aids in making well-rounded decisions.
- Core primitives contribute to a deeper comprehension of human behavior.
- Trust and intuition are interconnected elements in human psychology.
Behind the Mic
Initial Greetings and Introductions
Yeah, sounds good. I can. A bit of an echo, but I've got you. Hello? Hello? Yeah, hey, everyone. Hey, Aliyah. Yep, we've got you. Sounds good. How you doing? I'm good. So it's Monday, so lots of things from all directions, as usual. As usual in crypto. Absolutely. Hey, Anna, how are you? Hey, good morning. Doing well. Good morning. Yeah, I forgot you guys are all. We're all over the place. It's 01:00 a.m. my time right now. Okay. It's 09:00 a.m. it's a pretty reasonable time for me. There we go. I think we're just going to give everyone a few more minutes to join in and then we can kick things off.
Preparing to Start the Discussion
Okay, guys, just waiting a couple more minutes for more people to join. Make sure we don't have any surprise technical issues. I mean, it's spacious. I think that's the definition of surprise. Technical issues a given, really. All right, guys, let's get started. Welcome. Really excited to be hosting this panel on user owned AI. Today we're joined by whit from ring fence, Ilya, who I don't think needs an introduction. But anyway, you guys are welcome to introduce yourself and also Anna from Vana. We'll be covering pretty much everything, just exploring user owned AI, diving into past and current challenges of AI's development, closed source, open source, user owned AI, and also getting some insight into how we imagine the future of AI development and available data and what that might look like.
Community Interaction and Introduction Requests
There's also time for some community and user questions at the end, so if any of you have questions you want answered, just head to our Twitter and then a reply to the latest tweet with your question. And yeah, as long as it's relevant, we'll be able to answer it for you. Maybe if we start with introductions. Ilya, would you like to say a bit about yourself, about, you know, about Nir and what you're building and then how we can. We can get started as well after we've had introductions, for sure. Yeah, great to be here and kind of dive into username AI. Amelia.
Discussing AI and Blockchain
I'm co founder of near, which is a whole ecosystem of variety of tools that promote user ownership across blockchain itself, chain abstraction and AI and other things. And on specifically AI side, we're working on kind of how do we bring all these different efforts that are happening right now in web three and really help them through kind of accelerators, incubation, as well as through some of the kind of fabric of integration that we're working on, we're calling Neo AI hub. My background as well is in AI and machine learning. I was a Google research, led a team working on question answering, machine translation, and I'm a co author of the paper called attention is all you need, which introduced transformers core technology, powering chat, GPT midjourney, and all the other modern AI tech.
Introducing Vana and the Importance of Data
Yeah, good to see you. Wait, and Hannah here. Yeah. Really grateful to have you. Go ahead. Thanks. Yeah, I'm Anna here from Vana, one of the co founders. We're focused on building user owned AI through user owned data. So one of the big kind of bottlenecks in AI development today is lack of data. Ultimately, AI models are only as good as their data. And one of the bottlenecks today is called kind of this data wall where researchers have actually run out of data to train AI models on from just the public Internet. And so our approach at Bauna is to really unlock the walled gardens and the private parts of the Internet that right now are just owned by primarily big tech companies.
The Role of Data Ownership in AI Development
But we have users directly export their data, contribute them to different data daos built on the platform in order to really create better AI models than what centralized AI has. I come from kind of both a crypto and AI background, did research back at CSAIL, which is MIT's AI lab, and then first got into crypto through mining Ethereum with free electricity in my dorm room back in 2015. That story never gets old. That's great. I love, I love the thought of mining with free electricity. It's always a good thing. My name is Wood. I'm the co founder of Ring Fence. We're focused on building the data monetization layer of AI.
Collaboration with Other Teams and Initial Release
We work closely with Ilia near AI and the near foundation. We recently ring fence did complete the near horizon AI cohort. And we're focused now on our first release, which is Escher. We opened account registrations seven days ago now, and we're fortunate to have had over 80,000 people register their Escher accounts. And here pretty soon, we're going to open up our closed beta and give creators the ability to choose how they collaborate with AI on their own terms, which we're really excited about. Great. Thank you.
Exploring Interest in Crypto
I think you may be curious to hear about how maybe Anna and Ilya, and maybe you could speak about this yourself as well, is how you guys got interested in crypto in the first place. I think we've already heard an idea of what you're building. But, yeah, how did you get into crypto? And then maybe if you could mention why you're focused on this advancement of user and Aihdenhe. Whit do you want to kick off there? Maybe tell us a bit about your background in crypto? Sure. Yeah. So I got into the space in 2017 as a trader, realized I was really bad at that, and then I really focused my attention on what I had always enjoyed throughout life, and I was really just tinkering with things.
Transitioning from Trading to Mining and Development
And that led me to mining. Started as a GPU miner, was one of the miners of the Ravencoin Genesis block in 2018, hosted a podcast for about four years, and that evolved into one of the larger bitcoin mining podcasts in the space. And then in 2020, I launched compass mining with a couple of friends, and that took off like a rocket. And were very fortunate to be able to help tens of thousands of smaller miners be able to get access to bitcoin mining for the first time. And then from there, after my exit with Compass, I focused a lot of time and attention on this intersection of AI and web three, which led me to work with a firm in the UAE which was actually developing big data systems for the government in Dubai.
Challenges in AI Development and Collaboration
We worked on a specific project, building an app for asthmatics in a UAE, and I realized that the process of collecting data for these niche products was just a huge pain. And in the development of that app, we spent the better part of a year collecting the data to then just have to go through clinical trials and extended the development process for this app, which was badly needed by over a million people in the UAE, to well over two and a half years. And that's what led me to connecting with one of the ring fence co founders, Doctor Dmitry Mikhailov, at which point we started to plan how we could better source data that could be used for the training and fine tuning of AI models that could lead to the faster development of needed products.
Anna's Journey into Crypto
Amazing. Anna, could you say a little bit about what pushed you to get more involved in crypto specifically? And then, yeah, we'll get into this a lot later. But why? You're focused on this advancement of user and AI as well. Yeah, so I was really interested in currency, just centralized currency as well pretty early on. I had worked at the Federal Reserve back in high school and had a picture of Janet Yellen, chair of the Fed in my high school bedroom. It was just obsessed with currency, because to me, it was central to the whole economy, really.
Exploring Decentralization and the Power of Crypto
Like whoever controlled currency controlled a huge part of the economy. And so when I got to MIT and learned about these decentralized currencies and that I could run like, a small version of a kind of decentralized central bank, creating currency in just my room, I was super interested and just fascinated by, like, how do you build technologies that are decentralized and take something that usually would have to be owned by one big central institution and make it owned by many people, controlled by many people, and really allow for innovation to foster that way, too. Right. The sorts of applications that are built on Ethereum, really, they haven't been possible on the US dollar.
Decentralized AI and Data Models
Right. And so, like, how do you uniquely enable those sorts of applications and then the context of kind of AI and data? And Vana actually comes from a similar experience to wit, where, like, building AI models. So I started a machine learning company in 2017, dropped out of school, went through YC, and really what I saw was that the only thing that mattered for AI models was data. You just needed better data. That was really the thing that made the difference. So then I started thinking through, like, hey, where is most of the data today?
Data Accessibility in AI Development
Like, if you want to build really frontier, kind of cutting edge AI models, how do you get more data using a decentralized system? And that's how we eventually found our way towards Vana. Amazing. Thank you. I love the detail that you had a picture of Janet Yellen. That's amazing. Ilya. Yeah. If you'd like to say a bit as well about how you got into crypto, obviously a lot of people here know your background, but it would be interesting to hear a bit more about that.
Ilya's Journey into Crypto
Yeah, I'll triple down on the same kind of journey. We started the near AI in 2017.
Teaching Machines to Code
We were teaching machines to code, pretty much taking what was yet to be published paper on transformers and trying to leverage it to be able to write code from natural language descriptions. It sounded like a science fiction back in 2017. Now, obviously, everybody's raving about it on twitter, but we kind of thought that we will be able to acquire a lot of interesting data for this, for coding. And we had a lot of students around the world to pretty much do small tasks for us. They would be writing description for code, they would be writing small functions from descriptions. And we faced the challenge of paying them. Right. This is a lot of kind of gig, micro gig around the world, and we definitely did not want to set up entities to pay them and do all the kind of, like, traditional operations that need to come with this.
The Role of Cryptocurrency
And crypto seemed like a really powerful tool to have, you know, this global payment network that anybody can receive money for the work immediately. And, you know, they don't also need to trust us either. But, you know, this 2018, there was nothing that would actually allow us to really build that scale, cost efficiency and user experience. That kind of building product, you imagine, and being from kind of background of distributed systems and algorithms, blockchains to us seemed like, hey, this is something that we can just build the underlying platform to serve our use case, but also all the other use cases. And so we ended up kind of going from near AI to building near protocol, building that, launching that in 2020, obviously expanding the ecosystem, but also bringing back near AI now this year to really pretty much revise on that AI developer kind of functionality and research, as well as working with all of the folks in the space to really develop next generation of user owned AI.
Challenges in User Owned AI
I wonder if, as well you could maybe, especially with your experience, a lot of experience here on the panel, if you could maybe dive into what kind of current challenges you're seeing with the development of user owned AI. What challenges, you think? Maybe not just in terms of blockchain projects, but anyone trying to contribute to this advancement of user owned AI. What challenges are projects and people facing at the moment? Yeah, that's a great question. There's probably kind of a whole spectrum of challenges and, you know, us try, all of us trying to address them from different directions. But I think the fundamentally is this. There's kind of this split between people who are kind of thinking through, you know, who should control the AI and people who don't. Right. That that's kind of at the core and that is the ridge between, you know, most people just don't care about this. And I think that's the biggest challenge.
Fundamental Challenge of Coordination
And so, as, you know, you talk to researchers or other startups and they kind of don't care. They just find, you know, plugging in whatever open AI API into their endpoint, you know, funneling all of their data there and, you know, figuring out later if something needs to be changed. I think the kind of fundamental challenge is that we are yet to coordinate kind of a compatible solution that somebody can go and use and in kind of open source, user owned area. I think that's a real challenge. This is something that we're trying to address by bringing a number of partners together is like, how do we actually get to a solution that is ideally better than what you get from an open AI for a variety of reasons, because using open source and you can modify it better because you're using maybe extra data sources at OpenAI cannot because maybe you can monetize it differently.
Technical Challenges Ahead
But we need kind of a better product first and then we can then bring the values forward with that better product. And so a lot of things then span from there. And there's some technical challenges like, hey, how do we do actually like private computation? You know, if we're pitching that, this is actually something that is actually belongs to you. Ideally, nobody else should see what's in the media. But like technically it's extremely challenging, kind of there's like set of methods that are available and they are either too slow or not very efficient. And then you have kind of, you know, as you build a bigger systems, you know that complexity is growing. Or you can run everything on your side, right on edge on my device, on my cloud, but then yourself, like either limiting capabilities or limited capacity. And also software is not there yet. So those are kind of more practical technical challenges.
Philosophical and Product Challenges
I think there's a philosophy value challenge which limits how many people are looking at the space. There's a product challenge that I think we should address first, and then there's the technical challenges that we're going to be all working through as we go. It's interesting that you obviously bring up OpenAI because I think r1 challenge with these current AI systems is there's a real lack of transparency in how our data is handled and how data is scraped from the Internet. Anna, maybe if you'd like to speak a bit on that and talk about this lack of transparency in how our data is handled and maybe how you see this evolving in the future. Yeah, I mean, I think that a lot of people don't think of their data as their own. Right? Like if you think about your Twitter data or your email data, you might just be like, well, that's Google or that's exes or that's Reddit.
Ownership of Data
But in terms of kind of the law and the actual ownership of that data, it's sort of like when you park your car in a parking lot, like the parking lot doesn't own your car, you retain kind of full ownership. You're just leaving it there for a little while. And so it's the same thing for any platform. That data is kind of fully yours. And that's actually a really important legal point because it actually holds back AI development. Today, AI researchers can't train on a lot of the data sets that they might need access to because they don't have the legal right to. I think there is one angle which is, hey, it's so valuable to build AI right now that people are willing to take risks and it's okay if it's illegal. I think what we've seen to date from the data collection and the data sales industry is that a lot of the big players, it's not worth them taking that risk.
Decentralization and Data Privacy
Like Facebook paid a several billion dollars fine in the EU and they actually buy kind of data that they already have in their servers because they want to avoid that sort of situation. So, yeah, I see decentralization as really unlocking these new private data sets and also allowing for better attribution and kind of a fuller picture of a user. If you think about what a single platform has, Facebook, they have Instagram, Facebook, WhatsApp, they have a slice of my data, but they're never going to be able to get access to Apple's imessages or notes of me. If you want that full picture of the user, you have to have the user bring it. That's where we see decentralization really coming in.
The Importance of User Adoption
And I think that's so important to ask. Right. Because often, I mean, rarely does a user use a Product or a product C mass adoption for ideological reasons. Right. When you use a new product, you're asking, what can I do that I couldn't do before? Not like, oh, this is extra private, so I'm going to use it. And so I think making sure that there really is a step function in terms of what decentralization enables to ensure that it's actually adopted is so important in today's Kind of AI race. The last thing I would say is that because AI models Today are only trained on public data, you still do have this advantage of Private Data because it hasn't yet been trained on. But that sort of advantage it doesn't hold. I'd say we have about two years to do it.
The Urgency of Data Control
So now we're in this phase in AI where people are buying off the Private Data, right? So Reddit, selling your data for $200 million, like photo bucket, all these other platforms are kind of selling Data. So I think there's a lot of urgency to this question because once the data is out there in model weights that are owned by centralized companies instead of end users, it's sort of like the cat's out of the bag. Like it's pretty hard to go back. But we're still in the phase where the private data hasn't been used to train models yet. And so if we can get ahead of it, then you really can build user owned AI that beats the performance of centralized AI.
The Challenges of Attribution and Compensation
I mean, you know, you mentioned these like, kind of private deals around Reddit. A more recent one I saw was academic publishing giant Taylor and Francis. They just sold all of their research to Microsoft AI for, I think it was around like eight to 10 million. And that kind of bypassed a lot of the consent of, you know, or at least explicit consent of academic authors and publishers on the platform. And I think we're starting to kind of lose that attribution, and that's one thing that decentralization brings. And there's been a lot of legal battles around attribution and compensation, especially with respect for generative AI. Whitney, you've known this quite intimately. Original artists maybe having their works kind of scraped en masse from the Internet.
Future of Content Creation
Do you feel like this problem is maybe too far gone to be fixed, or do you feel like something can be done to bring attribution and compensation to these original artists? So I think that, you know, reparations in this situation are unlikely to happen, right? I don't think we're going to see anyone who's had their content or their original art scraped at this point, see any real benefit from it. I mean, outside of like, the New York Times and these major publications like Getty, you know, these big companies that have the ability to fight back. But I do think that there is a path forward where we have a world where creators are able to collaborate effectively with AI.
Empowering Creators through AI
It's very important to, you know, if you're an artist or if you are a content creator, you know, which is a continually growing industry, there's more and more people each year that want to become content creators. I think that's the number one dream job of most people that are teenagers right now, oddly enough. But they're continuing to create content, they're continuing to post online, and with everything that's getting posted, they're opening themselves up to the risk of having their content, their original art taken and not seeing compensation for it. So I do think that we're entering a world now where we're going to see these content creators and we're already seeing them use AI to enhance their work.
A New Wave of Content Creation
But we're going to see this grow exponentially in the coming years where not only will people start to use AI as a tool in the way that they are now, but they will actually be able to have their existing artwork that they've made fine tune, smaller models that they can then use to produce derivatives of their own work without feeding that original work into these larger models, which will then be able to, you know, have other people produce derivatives of their work. So I don't think we're too far gone. I think we're actually right at the right place for this, right at the right time. And we're going to see, I think, a new wave of content creators who maybe they lack some of the creative talents and they're going to be able to use AI to enhance what they're able to produce, but all of them will be able to effectively protect their work and collaborate with AI in a way that makes sense for them, which ultimately benefits them more than it does the underlying model.
The Debate on AI Models
So I'd also like to move on to maybe more of a broader discussion as we are maybe marching towards AGI, that scary phrase. I think there's growing discussion around proponents of open source AI, closed source AI and now user owned AI. I think it would be good to hear each of your thoughts on maybe, you know, pros and cons of each. I think we've already kind of scraped the surface of these especially. It's kind of more open source model or closed source model with OpenAI. Yeah. If you guys want to maybe elaborate on what you think the pros and cons of each of those are, that would be great to hear. Ilya, maybe if we could start with you.
Classifications of AI Models
Yeah, so kind of the classification I use is kind of at the root level is what are these models are optimized for? Because we need to understand that in models or agents or systems, they're optimized for so specific goal. Right. There's a benchmark behind the scene or set of benchmarks, and people are kind of looking at results and going back and changing the data set, changing the model, changing the training procedure, changing the filtering post pre whatever. Right. So, and that's where kind of I classified as between kind of corporate owned AI and user owned AI.
Corporate vs User-Owned AI
This is when it's developed by some company and corporation. It's designed to make more money for this company. Naturally, it doesn't matter if it's open source or closed source like they will, you know, continue using it to generate more revenue. That is exactly what their goal is. And as a company is already gigantic and have fully saturated their addressable market, the only thing they could do is to actually continue increasing how much money they make per individual user, which naturally means that these models are going to be leading you to use their product more, to spend more time on it, to pretty much spend more money on it one way or another. The user on the eye is actually models and systems and agents that are designed for your individual success, for your community success.
Optimizing Well-Being through AI
And so this is really about kind of the overall loss function of the system is really how to kind of optimize for your well being both, you know, not just financial, but also kind of holistic, right? Instead of showing you engaging content, instead of showing you empty calories content, how do we show you education? How do we enable you to learn more, to become better yourself over time? So that's kind of, you know, it's a bit philosophical, but it actually goes down to a very specific, like, you know, there's a feedback loop that is in every of the systems and it's really important to understand kind of what that feedback loop is based on. Just to give an example, right. You know, when I worked with Google, you launch a product, you look in at a b test difference and you see what model you trained and plugged into the system. If it improves revenue generated as a result of this change, that's just how this products are launched. That means you go back and you optimize and you change, maybe filter some data out to maybe make some decisions.
User Owned AI and Data Ethics
So that's really what from my perspective, the most important component kind of difference with what we're doing is user in user owned AI. Now one level down is indeed like, are you trained on, you know, open data? Are you trained on data that is kind of public domain, a training on data that belongs to people, the creators, and you rewarding them for this? Right. You have then is the model parameters are open or closed, or maybe they're mixed, meaning anybody can use them but you cannot download them effectively. So this is actually something that I think a lot of people are exploring now in web three space can we have a swarm of nodes that contain parameters, but no single party can actually use it. And so that gives a lot more control to the system, but has that control in the hands of the community. So that's why there's closed source, there's open source and there's kind of in the middle and same, there's open parameters, closed parameters and something in the middle.
Centralized vs Decentralized AI
So those kind of different levels of classification. I see. And again, then if it's open source, are you running on kind of centralized provider, inference provider that potentially can switch what happens under the hood. Are you running on decentralized provider that gives you different types of guarantees on that? The result you're getting are correct. And importantly, per use case here, you actually may have different requirements. That is all depends on what you're doing. And same like I using your public data, your private data that nobody else can see, maybe you have mixed data that you want to sell at the inference time, not the training time anymore. So there's always, I would say like a spectrum on these things, but that is kind of, to me, trapped by this meta level optimization process that I'm trying to at least get us to change.
Pros and Cons of Open versus Closed Source AI
Yeah, I really appreciate that insight, Anna. Maybe you have something to say as well on these pros and cons of each, because I think often we can quickly maybe bash on closed source AI, but each of them seem to have their merits and maybe even their downfalls. What do you think about that in terms of user and AI? Open source? Closed source. Yeah, I think that to date a lot of the conversation has been kind of just between two options. Like either you have model weights that are open and it's fully open source, or you keep them closed and it's closed source. And one of the good things about being closed source is that you have the ability to monetize the models, which is quite important because it can be really expensive to try trained models, right? It could cost $10 million, even $100 million in just compute for some of these leading AI models.
Privacy Concerns in AI Data Ownership
And then you have to think about data, which is one of the big costs, the research, etcetera. And so you have that kind of closed option. I mean, as an example too. Like if you have an AI model of yourself, right, which is trained on all of your private messages and all of your experiences, journal entries, recordings, I personally would not want that to be open source. I actually would want that to be private and I would want to earn as it's used and kind of have the little AI ANA agent ability to charge for its work. And if I made it open source, I would lose that ability. But one of the great things about open source is that it is really kind of community owned in a way that you see a lot of iteration.
Community Collaboration in AI Development
I think we saw this in especially the image space with stable diffusion, so many different fine tunes of the model that really made it better. Same with llama 70 B and Mixturel. Really kind of like improvements on the models because they're open source and the community can build on them however they'd like. I think user owned AI really gets you the best of both worlds, right, where you're able to maintain privacy of the weights of and the underlying data so that you can charge for the AI models, but you're also able to have kind of a large group of people collectively govern access. I think one example of this in the Vana ecosystem is the Reddit data down the AI model that they trained on their data, which is to date just good at shitposting, right? It's not a frontier model that's beating chat DBT, but it is kind of an end to end example, which shows, hey, we all collectively contributed our private Reddit data, which includes both kind of like posts and comments, but also messages, and now trained a better AI model at a particular niche task ship posting, and we get to decide how it's used, how we charge as it's used.
Governance and User Control in AI
They actually went through the full governance process of understanding that, choosing how to monetize it and all those different pieces. I think that's really the power of user owned AI. The last thing I would add is that I think a lot of the fears about AGI really come from losing control of AI. And this idea of, I don't want an AI model that takes my job and is all intelligent, but if you have proper attribution and the people who have helped to create the AI through their data are ultimately the ones that earn as it's used and decide how it's used, then you really, I think, can be a lot more comfortable with very advanced AI systems because they're more about scaling the power of humanity rather than replacing it.
Future Predictions for User Control in AI Development
So user and AI kind of bringing the best of both worlds and then maybe solving AGI, what do you think? Actually, I have a follow up question. So I think Ana and Ilya have done a good job covering the pros and cons of each of those. But just a question to each of you. I'm curious if we are going to move towards user owned AI, but we are battling this dynamic of closed datasets versus open datasets. Do we run into an issue with user owned AIH where we don't have enough available? And I know this, you know, not enough available data is a problem everywhere. But specifically, if we have everyone trying to compete for these specific closed data sets, are we going to be able to get to user owned AI? And how do we get there if everyone is clutching their data so tightly?
Inefficiencies in Current AI Model Training
Well, I think. I mean, there's few thoughts here that I have. First one is, I mean, the current approach of training these models is completely inefficient and it requires, you know, showing the same fact in kind of thousand different contexts for a model to actually remember it. And at the same time, we don't really like, there's no clear understanding where some of the capabilities, like actual reasoning capabilities come from, right? Beyond just, okay, it memorized that x is y, but, you know, it being able to reason now that, like, on a completely novel data, right. I think that is kind of, there needs to be advancement on how we're training these models in general so that they not, you know, required to read everything that humanitarian wrote to be able to do, you know, to respond to do a basic shit.
Shifts in Data Utilization for AI Training
Pause on Twitter. There is some research around training, for example, on the first five years of data collected from child, like an individual child or a few sets of children, there's a data set like that. And it shows that model is way more efficient at learning grammar and language than just learning from random articles on the Internet. Which kind of makes sense, right? You're not throwing just completely random gibberish, but actually having like a nice curriculum of similarly how human children are trained. So I think there's some advancements that are coming from this. We also see just some of the more exploration. We see synthetic data now being generated by bigger models. And I actually expect that we'll probably see a little bit of a change routine, like training routine, the pre training on next token generation, maybe not the absolutely best way to train those models, given we now have a synthetic data generation from already pre trained model.
Private Expertise Data in AI Training
So I'm expecting that general intelligence training will be happening kind of either from more synthetic data and in general, more curated data sets. And where the private data comes in is the expertise data, right? It's the data that, you know, how to do medical reasoning, how to respond about your individual things that you do, how to, you know, shitpost like you and not like somebody else. Like those is where the private data comes in and you don't need to train on it, you know, a global model. You just specialize it one way or another to leverage that. So I think, like, I'm just, you know, my at least believe is that we'll see kind of, we'll see reduction of how much data is needed of just like random crap on Internet, and we'll see a lot more kind of rigor on what data we feed to these models.
The Future of AI Training Approaches
And that in turn will, you know, it's more like, okay, now you have a grad student level like intelligence, you know, high intelligence, but not specialized. Right. You still need to train them the job you want them to do. You still need to give them context of what exactly you want them to do. Right. So I think that's kind of how it's going to proceed, but it'll take a few years to get there. Thanks, Ilya. Yeah, I think to add onto that one of the kind of the current data we train on today, it's very low quality, just as an industry. It's like, if you imagine taking a child and the only thing, the only way they can learn is by browsing the public Internet, just sort of randomly what comes at them.
Challenges of Low-Quality Data
That's like a really bad education source for a kid, right? You would just, like, have a very skewed, like, you're just like, in four chan, like, weird parts of the Internet. Like, you would have a very skewed understanding of the world. And I think that the more you can get high quality data, the better you can really create from kind of a performance perspective, and the less data you need as well. I think that textbooks is all you need. Paper was kind of interesting where they trained on much higher quality data and therefore needed way less data involved. And some of that data was actually synthetic, right? So you can kind of take some high quality real data and then scale it up through synthetic data and create really powerful data sets that then you train these AI models on.
The Need for Better Data Attribution
I think today, the world we live in today comes from this place where, like, we don't live in a user owned Internet. And so it's hard to have attribution, and generally people are like, hey, don't train on my data. Like, stay away from my data. So because of that, you end up with kind of like, low quality data sources that are available for training. Generally, it's through, like, scraping and kind of whatever's available. Even, like, if you take the academic journal articles as an example, that's like the end product. But having the AI model have access to, say, like a professor or their research team's kind of full thought process all throughout, all the notes, all the sort of breakthroughs that they had, and those different steps of the process would be really valuable for that kind of chain of thought like, process of reasoning.
Improving Data Quality through Attribution
So I think as we start to have better attribution of like, hey, how much is your data actually contribute to an AI model? In Vana's context, we use proof of contribution for that, just measuring, hey, how valuable actually is your data for improving this AI model? Then we're going to start to see higher and higher quality data, because if you're just scaling quantity but not scaling quality, then you're probably not going to be improving models. Amazing. Thanks, guys. I'd like to go back to something Anna mentioned earlier about these kind of private datasets being brought up. For example, the Reddit deal or the one I mentioned, Taylor and Francis.
Concerns over Data Consent and Ethics
And in a lot of cases, there's been a little bit of outrage to these, especially from, for example, academics and writers who kind of haven't given explicit consent and had their works taken and sold. And I wonder if this is the kind of trend at the moment. Every few months we seem to see one of these big deals go through. What kind of predictions do we have about how AI is going to evolve in terms of user control? What's AI development going to look like in the future, and how is user control going to be a bigger part of that wit? Do you have something to say about that? Well, I think that, you know, as were, you know, my personal belief is that, you know, as Anna mentioned, you can't build, you know, something like this on an ethos, right.
The Future of User Control in AI
There needs to be something tangible for people to sink their teeth into. And I think that we're coming into an era where people are hyper aware of ways that they can monetize their existing activities. It's kind of the something for nothing mentality. And, you know, as we come into this new era, I think that we're going to have the ability to speak to that desire in people to where they are going to want to monetize that which they are already doing for free.
Monetizing Data Control
And I know this is a big thesis for us at ring fence, and we're talking about how we can help people make money off of their data. And we're not really looking at it from the stance that it's the right thing to do, although we do believe it's the right thing for people to do, or more focus on the fact that there is money to be made. And people now more than ever, I think, are desirous of getting control of every possible revenue stream that they can. So I do think that this is going to be a big factor. I know a lot of people, especially in the crypto community, you know, we may love the tech, but most people get into crypto for the money at first, and then we stick around to see all the great things that come from the benefits of web three or bitcoin or whatever the case may be. So, you know, we're leaning heavily into this and I do think that this is the next wave.
Shifting Perception of Data Ownership
I think that this next wave of people that are going to be turning 18 in the next three to five years are going to be hyper aware of the money that they're leaving on the table in different areas of their life, and they're going to be looking to recapture that in every way that they can. Thanks, Whit. Anna, what do you think about how AI is evolving in terms of user control, especially in light of these kind of private sort of deals that are going through? Yeah, I mean, I think we've already seen a pretty big shift in public perception over the past, like year or two. Like if you think about kind of maybe ten years ago or even five years ago, I think the only sorts of movements that cared about owning your data and kind of data privacy were pretty radical, like sort of like cypherpunk.
Evolution of Data Privacy Awareness
I think, like, a lot of the crypto ethos really valued it, but I, it was not a mainstream conversation at all, and I think it also was very abstract for people, right. Of kind of like, what does it mean to own my data? What would I do with an export of my email data for the past ten years? I have no idea what to do with that. But then you have AI, which becomes this really visceral way to experience your data. I think one thing, if you hear a clone of your voice, you're like, wow, I want to make sure that I'm the only one who has control over that. Or kind of some of the image models that can generate images of you, or an LLM that has kind of your history, et cetera. People, just everyday people, actually care a lot about owning that in a way that I think we haven't seen in past waves of data ownership or privacy.
User Interest in Data Ownership
And so I think that's driving a lot of it. Like, you even have companies asking sort of, hey, does OpenAI train on my data? Or even looking at the terms of service when products like web, two products ship new features like notion ship something, and people are like, oh, do you train on my data? And they actually want to go and find that out. And so I think we've already seen a really big shift in public perception that's helped to accelerate user owned AI, and we're going to continue to see it shift as people learn more and more about the economic impact of these AI models. Yeah, I couldn't agree more on that public perception part. Ilya, anything else to add about whether you think AI is going to evolve towards this kind of user owned AI do you think now with this public perception kind of helping, it's just that's naturally going to take course? Or do you think maybe there's additional challenges that might be kind of not yet predicted?
Challenges in User Behavior
I mean, yeah, I think there's going to be challenges because although people care more, they don't actually change behavior because of that. Companies maybe more. But consumers don't actually, like, assign value to privacy, even if they like. During the survey, we'll say that they are concerned about data privacy and concern about impact of Instagram, et cetera. The number of people who are like, hey, I'm spending too much time on Instagram, and then they still have Instagram is tremendous. So I think a lot of it will because the product needs to better. Because of GDPR, it's actually way more expensive to handle user data and private data. And the products that are designed kind of differently can provide it at a way cheaper, lower cost, lower operational cost, etcetera.
Transformation of User Experience
So I think those are going to be important aspects. I think looking a little bit beyond, though, I'm expecting that we're going to see transformation of the user experience, like how you experience Internet, how you experience computing. We've got to start changing again. This is kind of where were investing back in 2017 with near AI, seeing that if you can empower everyone on earth to be able to build software, then the whole swaths of industry pretty much is not needed. Software as a service is not needed anymore because you can just build it for yourself. But then on the other side, you also can transform how you interacting with your friends, with social, with content, with creators, all of that is going to start transforming because you can create kind of on a fly experiences instead of getting a rigid app built by somebody else that doesn't support your workflow but doesn't have the intent, doesn't support the intents you want to execute.
Innovative User-Centric Applications
And so, I mean, we have like, you know, pretty cool prototypes and I mean, obviously the, like broader space as well where you can build front ends of applications already right on the fly without writing even a single line of code. And it, you know, plugs into blockchain data. It plugs it into other things and allows you to kind of interact, transact, but it's all like built from scratch kind of as user wants. And so you can imagine that in a predictive loop where it's, you know, predict that you want to do something, you know, suggest you to do it already with generated front end. And so I think, like, there will be a lot of user experience transformation that will go through. And all of that is enriched by your own data, by some of the prior data, by data that is sold at the moment that you kind of paid with crypto to access, et cetera.
Future Innovations and Anticipations
So lots of exciting work, you know, definitely time to build and build really great products for people. Yeah, for sure. I mean, that does partly lead into, you know, my next question. I just open this up as we approach at the end, you know, as you kind of think about how maybe the Internet, the way that we use the Internet, is changing, do we think that AI development is going to change quite radically as well? I mean, from whether that's from a project perspective, you know, people who are actually developing AI or even from the user perspective, people that maybe might be more incentivized and more willing to contribute data.
State of AI Development
Do we think, you know, I guess it's a two part question. Number one is AI development already changing in light of the public perceptions around user owned AI? And do we think, number two, is it maybe going to change quite a bit more radically from there, if that's open to any of you, if you have points there? Well, just quickly mention, I mean, I think AI development is changing every year right now, right? This is, we're in this kind of hockey stick growth stage right now. And so like, I look at any benchmark, right, it was like zero last year, it's like 50% this year, like ten months later. So I think, you know, there's a lot more attention, a lot more people trying to do things, which is exciting.
Innovative Collaboration in AI
Again, this is like, you know, going through and seeing multiple kind of winters, or at least autumns of AI, it's exciting to see the, I think as we build more interesting things where we have federated learning, where we have other components, I'm also excited to see new opportunities. But I think a lot of what I'm excited is to see better open source collaboration. A lot of this centralized labs, the reason why they can actually execute effectively is because all of the researchers and engineers are sitting in one codebase. They build it on top of each other's features. You come up with an idea, you add it to codebase. If I'm tomorrow, when I run something, I already have your feature and your idea in my code base, I can just plug it in, turn it on.
Challenges in Open Source
This is not happening in open source right now. Everybody builds their own environment and so there's not much, there's cross learning, cross kind of collaboration. Even though we are kind of in open source and we're like supposed to collaborate more. So I think that's one of the things that we can actually fix pretty effectively in this user owned AI and kind of open source and decentralization incentive space. So I'm excited about that. Yeah, sure. Appreciate that insight, Anna. Any final points from you on that or.
The Future of Varna and AI Development
I mean, otherwise I'd be really excited to hear about what's next for Varna, what you guys have going on, what's in the pipeline? Yeah, I think AI progress is accelerating so quickly. I mean, it's just honestly amazing to get to live during this period of time where there is an insane amount of innovation coming out like every single week. It's sort of like maybe six months ago or maybe a year ago, agents were like, oh, we're going to have agents and they're going to work. And I think sometimes there's an initial wave of hype, like, oh, it's going to happen. And then none of the products really deliver, but then they do like replica agent in the past week or so really can kind of autonomously do work and in a really powerful way.
Plans for the Future
I think for us at Vona, we've got kind of our testnet right now in the works, which is very developer focused for people who are building data daos. We're starting to bring users on board as we progressively decentralize that infrastructure and then some exciting announcements kind of later this year as we reach some broader kind of decentralization milestones. Amazing. Thanks for sharing, whit. Anything exciting coming from ring fence? We've got quite a bit planned, but maybe you want to speak on specifics.
Rapid Evolution in AI Development
Yeah, I mean, just to echo on the kind of this trajectory of AI development, you know, it really does feel like we're drinking from a fire hose right now. As Ana mentioned, the replic posts that I've seen over the past week or so just been amazing and it's almost too much to process all at once, but it's really exciting to see. And I think we're, you know, naturally, I think everyone views this as a bit of a bubble, but I think that once the bubble pops, and inevitably will, we're going to see the real companies that have been building meaningful products come out the other side much stronger.
Anticipating Future Advancements
And by that time there are going to be so many advancements in so many different areas. Some that I'm really excited about are in the health and wellness space. I think there's a ton of advancements that are coming there with AI and then also obviously with content creation, that's something that is really exciting to. I know the entire ring fence team with Escher coming in the next couple of days, we're going to be opening up our closed beta to 10,000 users and it'll be an incentivized beta program where people will get to help us troubleshoot, find bugs, earn some good rewards, and then we'll be opening up Escher to the public here in the next month or so.
Exciting Developments Ahead
So we've got quite a bit in the pipeline at ring fence and Escher and some really big announcements that'll be coming towards the end of this month and through the month of October. Very excited indeed. Ilya, maybe you guys have something to share as well. I know recently, you know, people thought maybe the near protocol account had been hacked. There was some crazy stuff going on. So maybe. What have you guys planned coming up? Maybe around redacted or anything else planned?
Unveiling Plans for Upcoming Innovations
Yeah, of course. I mean, we had a marketing campaign pretty much to, I would say raise awareness of kind of user ownership. That's what everyone obviously saw on Twitter and, you know, including the fact that, you know, Twitter itself is not user own platform. Right. Everything belongs to Twitter or xnow. But yeah, the idea was to kind of surface that we have a redacted hackathon that launched on the fifth. So this is an opportunity for anyone who's building in the web three and AI space to really come together and innovate on this user owned space.
Inviting Participation in Innovation
We have a whole host of partners as well, giving bounties there. And this hackathon will run through October into a conference we have at Bangkok right before Defcon called redacted again, kind of surfacing this idea that all your data should be redacted from centralized servers and centralized providers. So all of this is really kind of, we've been building this kind of awareness as much as possible around user ownership, kind of broadly both of your data, of your assets, of your power of choice and kind of AI is part of that.
Encouraging Collaboration and Engagement
So yeah, I invite everyone to participate who's developing or wants to develop something, find teammates. If you don't know how to code, I'm sure there's people looking for help across the board, and if you do, you should definitely join and build something amazing. Obviously a bunch of things going to be coming down the line in November as we're going to be talking, introductory.
Gratitude and Closing Remarks
Yeah, looking forward to it. So I'm aware it's now about two in the morning for wait, so I'm going to let him get some sleep if he can manage. I want to say super grateful for you guys for joining and speaking. Really appreciate it. Everyone listening, go follow Ilya Anna Witte, the projects near protocol at Withervana, at Rainfence AI. And yeah, really appreciate everyone for joining.
Recording and Farewell
Space has been recorded, so if you have anyone who's missed out, they'll be able to listen to the recording as well. Yeah. Thanks, guys. Appreciate it. Have a good one. Thanks, everyone.
