AiFi Summit 2024 Devcon, co-hosted by GAIB, Codatta, and Kite AI (formerly ZettaBlock) at Park Hyatt Bangkok, was successfully concluded on November 12. The number of registered people for this AiFi Summit reached 1,300, and more than 500 people attended. 27 projects and investment institutions including Paypal, BNB Chain, Base, NEAR Protocol, Story Protocol, 0G, Aethir, io.net, Exabits, Plume, Space and Time, Hyperbolic, Faction, Hashed and Coinbase Ventures made wonderful speeches.
Sarah, the Asia Pacific director of BNB Chain, delivered the first keynote speech. She mainly introduced the construction of the entire BNB Chain ecosystem, various developer support policies, and updated the audience on the current progress of BNB Chain in AI applications.
In the second keynote speech, Kony, CEO of the organizer GAIB, expressed his views on the potential opportunities in the current computing power market. He mentioned that AI is the most important era after mobile Internet, and computing power has captured a large proportion of the value of the entire chain in the AI boom. Compared with other financial assets, investing in GPU computing power assets can bring unmatched returns to other targets, but the current problem of the GPU market is that it is impossible to efficiently connect the two parties. On the one hand, operators have to pay huge financing costs when increasing the scale of GPU external financing; on the other hand, investors find it difficult to invest directly in computing power assets and usually can only choose to invest in semiconductor stocks such as Nvidia. GAIB provides investors with more decentralized, transparent and AI cash flow-based on-chain assets by tokenizing computing power assets and their income and providing liquidity.
The theme of the first roundtable discussion at the AiFi Summit was: "AiFi: Financialization of AI & Compute Assets". Core members from GAIB, Exabits, io.net, Aethir, WitnessChain and Plume teams discussed AiFi's current opportunities, challenges and industry regulation.
Exabits CIO Jonathan mentioned: Currently, if users want to use GPUs, they can only turn to major cloud service providers such as AWS or Azure, but these platforms tend to serve large enterprises. This preference will limit the development of start-ups. We need more democratic and open GPU resources to support small and medium-sized enterprises. In the Web3 world, everyone can become a GPU investor to break AWS's computing power monopoly, which is a huge industry opportunity.
Asa, the head of io.net's Asia-Pacific region, mentioned that 50% of GPUs in independent data centers outside the three major cloud vendors are still not fully utilized, and these data centers lack the opportunity to reach users. However, GPUs need to ensure continuous operation and also face maintenance issues. How to build an incentive mechanism to ensure the interests of investors and other participants is a major challenge in the AiFi track.
Kartik, the head of Aethir's ecosystem, mentioned that the entire system includes computing power demanders, computing power operators, and investors. How to convince them to participate in a market that relies on on-chain mechanisms and how to ensure the needs of all parties are full of challenges. The regulatory risk lies in the fact that in some countries and regions, incentivizing data center services through tokens may cause certain troubles, so it is necessary to determine the compliance boundaries in the customer agreement.
Ranvir, co-founder and CEO of WitnessChain, said: Computing power as a new asset requires a new pricing mechanism. There is no unified formula to calculate the commodity price of computing power. Different platforms and GPUs have different costs and performance. At the same time, GPUs with different performance will make different contributions to the same task, which creates opportunities for the design of new financial mechanisms.
Plume's CBO Teddy also mentioned that we need to be cautious in facing regulation when new assets emerge. There is already a certain compliance framework for AI-related assets to make asset transactions formal and feasible, which is also what Plume is doing to help ecological projects.
In the following keynote speech, Codatta's CEO Yi explained how decentralized data trading can drive AI towards AGI, and Codatta's position and mission in this process. He mentioned that only vertical data can improve the reasoning and planning capabilities of the basic model in a specific field, and only by collecting a large amount of data from different verticals can AGI be achieved. Each piece of data we provide as data contributors can actually be applied to multiple different scenarios, each of which will be commercialized by different companies, which means that the vertical data we provide will bring revenue over time, which is why we regard data as an asset. For this reason, we need to make data asset transactions easier and get relatively fair pricing in the market.
The second roundtable discussion focused on the Open Data Economy. Core members from projects such as Spheron, Theoriq, Space and Time, Hyperbolic, Base and Nevermined discussed the current status of the AI data ecosystem, infrastructure support and future ecosystem needs.
Ron, co-founder and CEO of Theoriq, mentioned that we are currently seeing many applications that go beyond simple conversational robots and governance robots on DAOs. These applications combine the cooperation of multiple agents. In addition to the crypto field, these applications are increasingly appearing in marketing, analysis and other scenarios. Many people believe that the greatest use of data is in training models, but we see that data plays an increasingly important role in the decision-making process. Different agents can obtain different data and work together to create the greatest value.
Scott, co-founder and CTO of Space and Time, said that Space and Time is currently using smart contracts to build a rule engine for the agent system, which allows the agent to use your funds in a trustless environment and achieve the most ideal agent on-chain form. Space and Time's products allow users to query the historical behavior of the agent and formulate strict execution policies for the agent.
Don, CEO of Nevermined, believes that there are two conditions for winning in the data market: one is to form a monopoly on data transactions, and the second is to restrict data assets to prevent data contributors from uploading meaningless assets. A feasible way is to build analysis tools for corresponding scenarios around data assets, so as to maximize the mining of data value and profit.
As one of the organizers, Kite AI's CEO Chi announced in her keynote speech that she will upgrade her brand and launch a new artificial intelligence platform, Kite AI, during the summit. She discussed the difficulties in the current development of centralized AI and how KiteAI can expand the boundaries of AI through its own solutions. She mentioned that due to the lack of data distribution channels and data ownership confirmation mechanisms, a large amount of personal data and even corporate data is difficult to be used for large model training. In the past year, the proportion of data sets with open source licenses on the Internet has dropped from 95% to 75%. It is difficult for companies doing model training to get the best quality data to provide to the model, and it is also difficult to make breakthroughs in model effects. The industry needs decentralized AI solutions to obtain more valuable data.
In the third roundtable discussion, team members from GM Network, Mind Network, 0G Labs, NEAR Protocol and Chainbase discussed topics such as how Web3 companies can participate in AI competition, data privacy, and application implementation.
Max, a founding team member of GM Network, mentioned that users have been generating a lot of data, but this data has not been well used, which will make the data lose its value. We need to combine the collected data with AI to make smart devices smarter.
Leon, the Asia Pacific director of Mind Network, mentioned that although there is no perfect data privacy protection measure in reality, different methods may be combined to explore feasible solutions. In order to protect user privacy, Mind Network currently encrypts data at three different levels: one is to encrypt data in distributed storage, one is to encrypt through full homomorphism during GPU computing, and the other is to encrypt at the application level.
Chris, an AI researcher at 0G Labs, mentioned that in traditional AI models, even for open source models, it is difficult to know what data was used in training and how they will perform in new scenarios, which makes it difficult to trust the model results. 0G has a good data storage infrastructure, and data can be loaded directly from the cloud to the training process. In the future, it will be possible to build more secure and reliable models through personal verification data.
Chainbase's COO Chris mentioned that there are currently two narratives in the market, one is crypto for AI, and the other is AI for Crypto. There have been many references to using crypto to solve the problem of large companies controlling data, computing power, and models. However, many use cases of AI for Crypto have emerged recently, such as truth terminal and AI payment, and more and more projects have begun to cooperate to support the AI ecosystem. Users are very concerned about whether data can make money, and the key task of the platform is to solve how to distribute benefits between data contributors and consumers. Developers are not a vision-driven group. The most important thing is to help them save time and make money.
In the subsequent keynote, Head of IPFi Bu Fan from Story Protocol and Prakarsh, Ecosystem Head of Spheron, expressed their views on decentralized AI assetization and how their organizations can adapt to this trend.
Bu Fan mentioned that there are already many scenarios for the combination of AI and Crypto in the market. The first is user-oriented chatbots, where creators create AI characters and issue commercial licenses on the chain; the second is AI meme coins, where creators can legally connect to source IP assets on the chain and issue tokens to the outside world; the third is to provide model training data (such as pictures), and continue to earn income by collecting royalties on the chain. However, these are only some very early applications, and the models have not yet taken shape. Creators can continue to explore scenarios where AI+Crypto is combined. The Story protocol focuses on standardizing IP activities through tokens and disseminating IP in different forms. He believes that most AI is also a kind of IP. If IP can be assetized, then AI can also be assetized. For example, the pictures used to train AI models can be IPs, and the AI model itself can also be IPs. When the AI model generates new content, IP distribution transactions can be carried out on the chain to achieve assetization.
Prakarsh mentioned that in the AI era, computing power will become the underlying anchor asset for most agents and most AI applications. Distributed computing power will have many application scenarios. They currently see more potential scenarios, including knowledge sharing between hospitals under the premise of protecting data privacy, and AI dialogue systems based on local computing power and model support, which will eventually form personal AI systems.
The fourth roundtable focused on how to connect the worlds of Crypto and AI. Investors discussed the problems currently encountered by centralized AI systems and in what areas Crypto+AI can break through.
Hiroki, head of research at Lemniscap, pointed out that there are two difficulties in building a decentralized AI network. One is that the scalability of decentralized computing networks is difficult to compare with centralized competitors, and the other is that the quality of data contributed by individuals is difficult to control.
Will, an investment partner at Faction, said that currently you can let AI plan your entire vacation, but the plan cannot be implemented because AI cannot currently help you pay. Will believes that AI Agents need to have crypto wallets, and crypto wallets will act as bank accounts. There will be huge opportunities in the payment technology stack because all financial transactions must flow through these Agents.
Ryan, an investment partner at Coinbase Ventures, believes that most current models can only access public data, and cannot access sensitive private data such as financial and medical data. Crypto can promote models to access private data pools and improve AI performance in specific areas. Agent systems are currently unable to complete very complex tasks. They actually do not know how to understand the content of smart contracts and take action. We need a large model that can obtain, understand and make human-readable analysis of smart contracts.
Dan, an investor at Hashed, pointed out that the incentive system for distributed AI is not very perfect at present. In the entire AI value chain, only a few people have made significant positive contributions, but their contributions are not reflected in incentives. The lack of a good distribution mechanism has led to unfair distribution. In addition, the models owned by the community must be safe and controllable, and the ownership of the parameters must be returned to the community for research, rather than providing a black box like a centralized company. If the model involves scenarios such as emotional companionship, it should be governed in an open environment.
Sylvia, director of Bullish Capital, mentioned that the incentive model design process must fully consider what the needs are. For example, if edge devices are needed, it is necessary to consider how to find them among many decentralized computing devices. Therefore, before the model architecture optimization problem is clarified, there is no way to design a truly effective incentive model.
The above is a complete review of the AiFi Summit 2024 Devcon. Even in the face of challenges such as regulation and incentive mechanisms, the AiFi track is also full of opportunities. With the new highs of the market and the all-round popularity of the AI track, the industry is generally positive, with a continuous influx of talents and more and more innovations emerging.
For more content, please follow
GAIB: https://x.com/gaib_ai
Codatta: https://x.com/codatta_io
KITE AI: https://x.com/GoKiteAI
