Author:Haotian
After reviewing several popular projects in the Crypto+AI track over the past month, we found three significant trend changes, with brief introductions and comments on the projects:
1) The project technology path is more pragmatic, and begins to focus on performance data rather than pure conceptual packaging;
2) Vertical segmentation scenarios have become the focus of expansion, and general AI has given way to specialized AI;
3) Capital places more emphasis on business model verification, and projects with cash flow are obviously more favored;
Attachment: Project introduction, highlights analysis, personal comments:
1. @yupp_ai
Project Introduction: Decentralized AI model evaluation platform, completed a $33 million seed round in June, led by a16z and participated by Jeff Dean.
Highlights: Applying the advantages of human subjective judgment to the shortcomings of AI evaluation. Scoring 500+ large models through crowdsourcing, user feedback can be exchanged for cash (1000 points = 1 USD), which has attracted companies such as OpenAI to purchase data and has real cash flow.
Personal comments: Projects with relatively clear business models are not pure money-burning models. However, preventing fraudulent orders is a big challenge, and the anti-Witch attack algorithm needs to be continuously optimized. However, judging from the financing scale of US$33 million, capital clearly values projects with monetization verification.
2. @Gradient_HQ
Project Introduction: Decentralized AI computing network, completed a $10 million seed round in June, led by Pantera Capital and Multicoin Capital.
Highlights: Relying on the Sentry Nodes browser plug-in, there has been a certain market consensus in the Solana DePIN field. The team members come from Helium and others. The newly launched Lattica data transmission protocol and Parallax inference engine have made substantial explorations in edge computing and data verifiability, which can reduce latency by 40% and support heterogeneous device access.
Personal comments: The direction is right, and it is stuck in the trend of AI localization. However, the efficiency of handling complex tasks is better than that of centralized platforms, and the stability of edge nodes is still a problem. However, edge computing is a new demand generated by web2AI and is also the advantage of web3AI's distributed framework. We are optimistic about the implementation of specific products with actual performance.
3. @PublicAI_
Project Introduction: A decentralized AI data infrastructure platform that incentivizes global users to contribute data in multiple fields (medical, autonomous driving, voice, etc.) through tokens. It has accumulated revenue of over 14 million US dollars and established a network of millions of data contributors.
Highlights: Technically, it integrates ZK verification and BFT consensus algorithms to ensure data quality, and also uses Amazon Nitro Enclaves privacy computing technology to meet compliance requirements. More interesting is the launch of the HeadCap brainwave acquisition device, which can be regarded as an expansion from software to hardware. The economic model is also well designed. Users can earn $16 + 500,000 points for 10 hours of voice annotation, and the cost of enterprise subscription data services can be reduced by 45%.
Personal comments: I feel that the greatest value of this project is that it meets the real needs of AI data annotation, especially in fields such as medical care and autonomous driving, which have extremely high requirements for data quality and compliance. However, the 20% error rate is still a bit higher than the 10% of traditional platforms, and data quality fluctuations are a problem that needs to be continuously solved. There is a lot of room for imagination in the direction of brain-computer interface, but the implementation is also difficult.
4. @sparkchainai
Project Introduction: Solana is a distributed computing network on the chain. It completed a financing of US$10.8 million in June, led by OakStone Ventures.
Highlights: It aggregates idle GPU resources through dynamic sharding technology, supports reasoning of large models such as Llama3-405B, and costs 40% lower than AWS. The design of tokenized data transactions is quite interesting, directly turning computing power contributors into stakeholders, and can also encourage more people to participate in the network.
Personal comment: This is a typical “aggregate idle resources” model, which makes sense logically. However, the 15% cross-chain verification error rate is indeed a bit high, and the technical stability needs to be further polished. However, it does have advantages in scenarios such as 3D rendering that do not require high real-time performance. The key is whether the error rate can be reduced, otherwise even the best business model will be dragged down by technical problems.
5. @olaxbt_terminal
Project profile: AI-driven high-frequency cryptocurrency trading platform, completed $3.38 million seed round in June, @ambergroup_io
Lead investor.
Highlights: MCP technology can dynamically optimize transaction paths, reduce slippage, and improve efficiency by 30%. Catering to the #AgentFi trend, it can be regarded as finding an entry point in the relatively blank segment of DeFi quantitative trading, which can be regarded as filling the market demand.
Personal comments: The direction is fine, and DeFi does need smarter trading tools. However, high-frequency trading has extremely high requirements for latency and accuracy, and the real-time coordination between AI prediction and on-chain execution needs to be verified. In addition, MEV attacks are a big risk, and technical protection measures must keep up.
