It is an honor to complete this research under the guidance of Meteorite Labs, based on the experience exchange of hundreds of Web2 AI applications.
PIN AI is a selected project of the a16z Crypto Startup Accelerator Autumn Program. It has raised US$10 million in seed round financing. In addition to a16z Crypto, well-known VCs include Stanford Blockchain Accelerator, Hack VC, and Foresight Ventures. Angel investors include NEAR Protocol founder Illia Polosukhin, Gitcoin co-founder Scott Moore, Solana Foundation Chairman Lily Liu, SUI/Mysten Labs CEO Evan Cheng, Worldcoin research engineer DCBuilder, etc.
I just finished reading the article co-written by the three co-founders of PIN AI, and found it to be the most attractive Web3 AI project to me recently besides Sahara. The application scenarios are very interesting.
PIN AI is an open AI network where developers can build useful AI applications. "Useful AI Applications" is the focus of its products. It is somewhat similar to AI Agents such as Web2's MultiOn and Jace.ai. It is committed to providing users with applications that are useful in daily life and realizing the user's intentions, such as purchasing goods online, planning travel, and planning investment behaviors.
Source: Weixin
Let’s briefly introduce Jace.ai, which is an AI Agent that can complete browser tasks autonomously. It is based on LLM and its proprietary model AWA-1 (Autonomous Web Agent-1), which can support AI to perform operations on web pages.
Jace's greatest ability is to plan tasks independently and perform operations in the browser on behalf of the user.
Take an example to understand the application scenario and tell Jace, "I am going to travel to Beijing for a week on September 20th with a budget of 5,000 yuan. Please help me plan it." Jace will automatically plan a travel plan, including the attractions to visit, the hotels to stay in, and the food to eat. If I agree to this travel plan, they will help me book all the scenic spots, find the most cost-effective hotel on Meituan and place an order. All I had to do was enter my personal information and pay with one click.
In fact, what PIN AI does is very similar. The biggest difference from generative AI is that this type of AI project mainly focuses on the daily life of users, rather than work.
1. Deconstructing PIN AI design ideas
Simply put, PIN AI = AI + DePIN
The PIN AI network is composed of two types of AI:
Personal AI: Personalized AI Agent that can instantly adapt to user preferences. It is the connection point between users and proxy services, a bit like a coordinator. Users can download it to their mobile phones or computer devices for use.
Agent Services: AI Agent built on the chain for the Web2 platform. It can perform tasks on some top Web2 platforms. The execution process and completion status are recorded on the blockchain.
The official also mentioned External AI, which may support interaction with other LLM or Web2 AI Agents in the future.
PIN AI architecture core:
PIN Protocol, a DePIN decentralized data storage network, allows anyone to connect their devices to the network and share data. Integrated BERT-based model to anonymize user data at all stages of processing, ensuring privacy and compliance with data protection regulations.
Personal AI is built into it. On the one hand, personal AI is provided with personalized data, and on the other hand, the agent service is provided with the most relevant data.
PIN Protocol is built from three components:
Private storage and computing layer: Store data in a decentralized manner, securely store device data shared by users (including photos, videos, etc.), and make the most relevant data available to personal AI and agent services at any time. Users can connect their devices to the network, provide device data, and be rewarded with native tokens $PIN.
Data connector: Use zk technology to track and verify user data connected to the network. I think it is equivalent to the nodes of the PIN network. Node operators need to stake $PIN tokens for verification, and some token holders can stake tokens to the node, and both can get token staking rewards.
Agent Connect: Designed to match personal AI with agent services. Consisting of an agent registry and a transaction mechanism, the former is used to track performance indicators, while the latter "thinks" how to match personal AI to agent services (based on the cost, performance and completion quality of each agent service)
User usage pattern/business logic:
When the user puts forward a specific requirement during use, PIN AI will follow the following steps:
Step 1: Personal AI - Collect user needs
The user makes a request to the personal AI, and the personal AI forwards the request to the PIN Protocol
Step 2: PIN Protocol - Prepare for task execution
Break down the user's intention into specific steps, find the most suitable and cost-effective agency service, retrieve the most relevant data, and provide it to the agency service. (If multiple Web2 platforms are involved, the intent needs to be split into different proxy services)
Step 3: Proxy service - perform specific steps
Step 4: PIN Protocol - Feed back the results to the user
After all, daily life needs mostly require money transactions. In PIN AI, the flow of funds should be:
The user pays Gas fees to PIN Protocol (personal guess is to initiate this intended transaction). Since PIN Protocol first disassembles the user's intention, and then indexes and sends the data most relevant to the intention to the proxy service, after the proxy service completes the task execution, part of the service fee will be fed back to PIN Protocol as a tip.
Therefore, both PIN Protocol and proxy services can draw commissions from the service fees given by users.
Source: Weixin
For example:
Users can download personal AI to their computer or mobile phone, make requests to the personal AI, such as "Buy the cheapest GTX 3080 graphics card on Amazon", and pay the fee (purchase fee + service fee + PIN Protocol Gas fee).
Personal AI communicates this need to PIN Protocol.
After understanding and decomposing the user's intent, PIN Protocol will decompose the user's intent into specific task steps and send them to the proxy service together with the most relevant data. There may be dozens of proxy services dedicated to Amazon shopping at the same time, so PIN Protocol needs to comprehensively consider their cost, performance and past performance to select the most suitable one.
Find the most cost-effective GTX 3080 graphics card on Amazon through an agency service and place an order. Upon completion, the intent disassembly fee and data call fee will be paid to PIN Protocol. PIN Protocol and Personal AI report results back to users, sending them PIN tokens as rewards.
network participants
Personal AI users: Install personal AI on your computer or mobile phone, connect your personal data to the PIN Protocol, and receive PIN token rewards.
Source: Weixin
Value transfer users: Like the above usage model, users who conduct valuable transactions will also receive PIN token rewards.
PIN Protocol Node: Tracks and authenticates user data connected to the network. Operators need to stake, token holders can stake tokens to nodes, and both can receive staking rewards.
Agency services: Developers can earn service fees.
2. Core development team
Davide Crapis - Co-Founder
Blockchain protocol design background, some AI background
Served as a senior data scientist at Lyft, where he designed and implemented an incentive distribution algorithm that distributed $xx in annual growth incentives to riders and drivers. Later, after resigning, he worked as an independent researcher for a period of time, studying incentive programs and token distribution. Before founding PIN AI, he served as a research scientist in the field of "Robust Incentives" at the Ethereum Foundation.
He once developed a machine learning model of "consumer sensitivity to investment/credit product interest rates" and also served as a researcher and mentor in the field of machine learning at Columbia Business School for four years. Joined the Web2 developer community "South Park" to explore the intersection of large language models and blockchain industries.
Ben Wu - Co-Founder
Operational background, may provide strategic direction and AI product ideas
MIT graduate and Y Combinator alumnus. Before founding PIN AI, he served as the director of database and operations in Yahoo's strategic data solutions department, responsible for the operation and management of large-scale data projects.
Bill Sun - Co-founder, Chief Scientist
Quantitative Trading and AI Background
PhD in mathematics from Stanford University and worked on AI research at Google DeepMind. I once worked as an artificial intelligence/quantitative trading stock investment manager in a Wall Street asset management company. Created the AI research organization AI+Club and the AI technology community AGI House. Angel investor in a16z scout fund. He is also the founder of Generative Alpha, which develops enterprise-level AI solutions.
3. Thoughts and Summary
In the first industrial revolution, machinery freed up hands;
In the second industrial revolution, electricity broke the boundaries between day and night;
In the third industrial revolution, the Internet merges the boundaries of virtuality and reality.
The emergence of AI is generally considered to be a symbol of the fourth industrial revolution, and AI Agent is the ticket on this journey of exploration. Each of us can board this ship to the future of "human-computer interaction".
In the past few decades, a large number of activities have occurred on the Internet every day and a large amount of data has been generated. However, users do not have ownership of these data.
iPhone16 has just been released, bringing Apple Intelligence, but PIN AI has the opportunity to build a more open AI Agent ecosystem than Apple Intelligence.
Among them, developers can be rewarded for developing innovative Web2 platform agent services, which will lead to AI Agents with increasingly higher performance and quality, triggering a wave of innovation.
And billions of mobile phone users can not only use personalized personal AI, but also share device data to earn rewards.
User data supports the entire PIN AI ecosystem. This is the power of users and the starting point of Web3 - decentralization and ownership.
We hope to see the implementation of the PIN AI network as soon as possible and whether the incentive mechanism it brings can work effectively, so that an army of open source contributors can flow into it and create a larger wave of innovation. The testnet may be launched in October, and the mainnet and TGE will be launched in January next year, which is worth looking forward to.
[Disclaimer] There are risks in the market, so investment needs to be cautious. This article does not constitute investment advice, and users should consider whether any opinions, views or conclusions contained in this article are appropriate for their particular circumstances. Invest accordingly and do so at your own risk.
This article is reprinted with permission from: "Foresight News"
Original author: Heimi, Baize Research Institute
