Interviewer: Grapefruit, ChainCatcher
Guests: Sean Ren, Tyler Zhou, Sahara Co-founder
Editor: Marco, ChainCatcher
Since OpenAI released the large-scale AI model Sora, AI has once again become the most popular track in the market. Investment waves have continued, and innovative projects integrating AI and Web3 have sprung up like mushrooms after rain. According to the encrypted data platform Rootdata, the number of projects included in the "AI and Web3" section has reached nearly 240, which has obviously become an independent track. The decentralized AI network Sahara is one of the star projects in this section.
Founded by Sean Ren and Tyler Zhou in May last year, Sahara is a decentralized AI network infrastructure that helps AI assetization and aims to help users deploy or build customized, personalized AI products.
Sean Ren is a tenured professor in the Department of Computer Science at the University of Southern California and has 15 years of industry research experience in the field of AI; Tyler Zhou served as investment director at Binance Labs and participated in the investment and incubation of multiple projects.
In March this year, Sahara announced that it had completed a US$6 million financing round led by Polychain Capital as early as August last year, with participating investors including Sequoia Capital, Samsung Next, Nomad Capital and other investment institutions.
The two founders told ChainCatcher that Sahara has provided data services to more than 30 corporate clients, including well-known companies such as Microsoft, Amazon, MIT, Snapchat, Character AI, and has earned millions of dollars in revenue.
In an exclusive interview with ChainCatcher, Tyler Zhou revealed that Sahara will launch C-end user products in April-May; the Sahara testnet will be launched in Q3, and the mainnet will be launched in Q4.
On April 4, Sahara launched its first points event, Sahara Social, on the task platform Galxe for early users. Users can earn early points rewards by connecting to the Sahara network, registering for the waiting list, and other tasks.
The story behind the creation of Sahara
1. ChainCatcher: What is Sean Ren’s personal experience? How did he get involved in Web3? What is his job at Sahara?
Sean Ren: My personal study and work experience are more inclined towards engineering background.
Before founding Sahara, I had been a tenured professor in the Department of Computer Science at the University of Southern California (USC) for seven years, mainly doing academic research in AI and NLP.
When I was studying for a PhD in Computer Science at the University of Illinois at Urbana-Champaign, I created my first entrepreneurial project, StylePuzzle (an e-commerce platform for clothing recommendations for fashion experts), and received investment from Plug & Play Ventures, all the way from the Angel Round to the C Round.
Tyler and I have been friends for 6 years. The opportunity for us to start a business should be in 2022. At that time, we discussed many shortcomings of Web2 AI products themselves, especially the economic model problems.
Only a small number of professionals benefit from the current AI economic model. Other AI ecosystem participants, such as data owners, collectors, providers, and model feedback recipients, do not receive reasonable economic compensation, and users' data privacy issues are not resolved, which is not conducive to long-term development.
The first principle of Sahara products is to solve the pain points of the current traditional AI industry and ensure that all participants in the AI ecosystem can obtain appropriate or reasonable benefits based on their work contributions, and are no longer limited to the computing power of large models and landing scenarios.
Currently, I am mainly responsible for product development and BD at Sahara.
ChainCatcher: Tyler Zhou previously served as investment director at Binance Labs. Why did he leave Binance and choose to start a business in the AI field?
Tyler Zhou: After graduating from UC Berkeley, I worked in investment banking and private equity investment, mainly investing in infrastructure, information technology (IT), and real estate.
I joined Binance Labs in early 2022 and was responsible for investment work in the US market, mainly incubating and investing in projects. The first phase of MVB was launched under my leadership.
There are several main considerations for leaving Binance and Professor Ren to establish Sahara in early 2023: First, there are many problems with the economic model of the entire AI ecosystem, and blockchain's encryption technology and token economy may be used to solve these problems.
I personally feel that Professor Ren's background and expertise make him the most suitable person to develop related products. There is no team in the market that understands the entire closed loop of the AI system's ecology, technology, and economy as well as Professor Ren.
In addition, Professor Ren is not a researcher who only does research in the traditional sense, he also has a very strong business sensitivity and sense of smell.
2. ChainCatcher: What is Sahara's product positioning and goal? In addition to the commercialization of AI, what other problems do you want to solve?
Sean Ren: Currently, Sahara's main product is a decentralized network infrastructure that supports anyone to build or deploy their personalized AI products.
Sahara can be regarded as a decentralized network consisting of Execution Layer, Transaction Layer and Application Layer.
At the application layer, Sahara provides a native built-in decentralized data market (Decentralized Data Marketplace also known as Sahara Data), and provides supporting facilities such as data processing-related toolkits (such as collection, labeling, QA, etc.) for users to use and access to help train their AI models.
Most users come to Sahara to build their own AI products. Sahara Data can first help solve the problems of data collection, labeling and conversion.
In addition, as a data market, Sahara is an important linking platform that attracts data suppliers and demanders. It can not only provide high-value data services for AI model training, but also help users with data needs find more data providers, so as to build autonomous AI more smoothly.
Sahara Data's decentralized data market is a major advantage of Sahara products and is also the key to distinguishing it from other Web3 AI projects on the market. Launched in October last year, it has been online for about 6 months. Initially, it only served some corporate customers, such as Microsoft, Snapchat, MIT, Motherson Group, Amazon, etc., providing them with relevant data services and handling some of the most difficult data demand issues in the industry.
The Sahara execution layer supports data encryption and attribution, i.e. proof of ownership, which is achieved by using innovative digital watermarking technology and public key facilities. It is similar to a proof of ownership. When a user creates a data point, data set, or model, he can embed his Did into the data or model to generate a watermark to prove his ownership of the data. This watermark will always exist as the user's data and model flow, so that the data and model can be attributed.
By using the proof of ownership mechanism, when users train their own AI or make inferences, if they need to rely on a basic model built by a certain person or a group of people, the income generated by the AI product in the future can be distributed to the holders of the underlying model.
3. ChainCatcher: On March 5, Sahara announced that it had completed a $6 million seed round of financing led by Polychain Capital, with participation from Sequoia Capital and others. How did Sahara approach these investors? Why do you think they are optimistic about Sahara? What kind of help did the investment institutions provide?
Tyler Zhou: In fact, the seed round of financing is not a recent financing, it was completed as early as August last year. Unlike other teams, Sahara did not make it public immediately after the financing was completed, but waited for the right time and the product was launched before promoting it.
The reason for the announcement now is that Sahara product development has entered a new stage, and a series of new products will be launched next, including ecosystem-related applications.
Sahara's seed round is oversubscribed, and there are many options to choose from, but our choice of investors is very strategic. These investors can help Sahara see the development of AI companies around the world, what AI startups at different stages are doing, the global AI economic trends, the differences in AI economies in different countries, and what leading AI companies are doing.
Sahara's advantages and development progress
4. ChainCatcher: How do you think Sahara is different from traditional AI products such as ChatGPT?
Sahara co-founder Sean Ren: It can be viewed from two levels.
First of all, Sahara is not an application developer, and the final product it delivers is not a GPT product. Sahara is a decentralized network infrastructure provider, and its application layer does not define the appearance of the AI Agent product that developers want to build, but provides related supporting APIs, SDKs and other toolkits, allowing anyone to easily build their own AI Agent.
Second, ChatGPT is a conversational robot in the form of question and answer, while Sahara Knowledge Agents (KA knowledge agents) on Sahara are custom artificial intelligence programs. They are very different from traditional conversational robots. They can analyze data autonomously, make reliable decisions based on specific needs, and act or perform tasks based on certain instructions to achieve a certain purpose.
For example, the purpose of a KOL's KA is to help it screen the advertising invitation information in his Twitter DM, generate a concise report every day, reply to other people's DM, etc. KA can automatically execute these commands at any time.
Sahara is an infrastructure builder and provides tools and platforms for building custom Knowledge Agents (KA).
Tyler Zhou: Compared with ChatGPT and other AI projects on the market, Sahara focuses on Personalized Agent, a customized artificial intelligence program whose capabilities are not limited to chatting, but can also help users perform many things.
There are two prerequisites for building a "Personalized Agent". The first is that you need to have your own database, and then train the AI Agent based on your own data to enable it to achieve some desired capabilities; the second is that you need to have the relevant infrastructure and tools to build the Agent.
Sahara not only provides data-related processing tools such as data markets, but its execution layer can also ensure that user data can train its own AI while protecting privacy, and provide relevant infrastructure tools to help users better build KA.
The Sahara network allows users to customize their own AI Agents without sacrificing their data privacy, and also provides a no-code platform for building Agents.
5. ChainCatcher: What are the data of Sahara since its establishment? What is the next focus of work?
Sean Ren: First of all, regarding the product, Sahara has been building a decentralized data market, Sahara Data, and launched the product in October last year.
To date, Sahara Data has worked with 31 corporate clients, honing its technology while increasing revenue.
As of the end of Q1 this year, Sahara Data has accumulated 200,000 users.
Next, Sahara will focus on three aspects, which are also important actions in the short term:
First, the decentralized data market is open to the public, and can be used by both individuals and businesses;
Second, based on Sahara Data, we will build or launch some C-end user products related to the execution layer, such as Knowledge Vault, Knowledge Marketplace, etc.
Third, the Sahara testnet will be launched in Q3.
6. ChainCatcher: What is the biggest challenge the company is facing now?
Sean Ren: The company's team has expanded too fast, from a dozen people at the beginning to more than 40 people. According to the recent plan, the team may increase by another 30-40 people in the next one or two months.
Tyler Zhou: From the perspective of the market and ecology, I think the biggest challenge is that in the initial stage, Sahara is positioned as a "Crypto for AI" product (that is, using Crypto to empower the entire AI), and the market and ecology it reaches will be larger than "AI for Crypto".
Regarding the difference between "Crypto for AI" and "AI for Crypto", the former "Crypto for AI" refers to the use of Crypto and blockchain technology and AI to make a better combination to help enhance and improve issues related to AI products, etc. This is a larger global market; the latter "AI for Crypto" means that AI is used to improve encrypted products, such as in smart contracts, or some uses of blockchain, etc. At this stage, this is a relatively small market and is more of a narrative.
However, the current mainstream hype in the market is the "AI for Crypto" product, which ignores the economy of the entire AI ecosystem and the entire AI system, as well as the trends of the global economy. This has led to a very noisy market, especially in the past six months.
It is a big challenge for Sahara to stick to her original intention of doing what she wants to do, what she should do and what is right in such a noisy environment.
Sahara has served more than 30 companies including Microsoft and Amazon
7. ChainCatcher: Since its establishment, Sahara has attracted more than 30 corporate clients, including Microsoft, Snapchat, and MIT. What services does Sahara provide for them, and why do these well-known companies choose to cooperate with the company?
Sean Ren: Sahara's first product was the decentralized data market Sahara Data. There are similar competitors in Web2, such as Skill AI centralized data service providers. Compared with centralized data service providers, Sahara Data has more advantages.
First of all, Sahara can reach a large number of AI data collection and annotation workers in various regions of the world through a variety of rewards and economic incentive mechanisms. There are currently about 200,000 AI-related workers on the Sahara network, and most of them are natives of the Web2 AI industry. The reason why they are attracted is that Sahara allows them to gain benefits from data contributions, such as using Crypto as payment.
Regarding the cooperating companies, they have various data needs. For example, Snapchat needs to collect conversation data, Microsoft collects multimodal data, and MIT needs various video data.
As a data supplier, Sahara has great advantages in data diversification. It has a very diverse candidate database for customers to choose from and can adapt to different data needs.
By cooperating with more than 30 companies, Sahara continues to refine its products, making them more and more mature, so as to better adapt to the data needs of various companies and merchants around the world, forming a positive cycle.
8. ChainCatcher: Currently, users need to be on the waiting list on the official website to participate in Sahara. How is the product development progress? What are the ways for users to participate in Sahara? What rewards will be given to early participants?
Tyler Zhou: From April to May, Sahara will launch its first C-end product, where C-end users can contribute their knowledge and skills to the platform.
Sahara will also have different reward mechanisms for early users. More information will be released after it is made public in April or May. The Sahara Testnet will be launched in Q3 or Q4, and the mainnet will be launched in Q4.
9. ChainCatcher: What do you think of the popular decentralized GPU, Agent and other AI encryption projects in the market? How to judge the reliability of an AI encryption project?
Sean Ren: Currently, the crypto AI products on the market can be divided into two factions: "AI for Crypto" and "Crypto for AI".
"Crypto for AI" is a larger market. For AI industry natives like us, we are more concerned about how to use blockchain and Web3 technology to solve some criticisms of Web2 AI products, especially economic models, data ownership and other issues.
Currently, there are many projects that use the economic model of blockchain to incentivize certain behaviors in AI, but I personally think that they are too superficial. They only look at the economic model and do not consider the entire ecosystem behind AI, such as the privacy and encryption of training data.
If we look at the entire AI ecosystem, the upstream should be the data and data processing sectors. There are also data-related projects in the market, but most projects only have the most attractive crypto-economic models (such as Label to earn), and do not touch on issues related to the data itself, such as ownership, model attribution, etc. They just build an application.
Regarding the use of distributed GPUs to train large models, I personally think this direction is very challenging and depends on the degree of decentralization of the project. If you just symbolically tie together the data in the same computer room or several computer rooms not far away, this is an artificial decentralization; if you want to tie together idle GPUs around the world for decentralized training, it is a bit difficult to achieve due to the huge difference in speed between different networks.
There are also some popular ones such as machine learning and ZK, which are still in the long term. Therefore, when judging a project, we should first see which projects are achievable and commercializable in the short term, and which are research projects that require long-term exploration.