After reviewing several popular projects on the Crypto+AI track over the past month, three significant trend changes were identified, along with a brief introduction and review of the projects: 1) The project's technical path is more pragmatic, focusing on performance data rather than pure conceptual packaging; 2) Vertical segmentation scenarios become the focus of expansion, with general AI giving way to specialized AI; 3) Capital places greater emphasis on business model validation, and projects with cash flow are clearly more favored; Attachment: Project Introduction, Highlight Analysis, Personal Comment: 1、 @yupp_ai Project Introduction: A decentralized AI model evaluation platform, completed a $33 million seed round in June, led by a16z and participated by Jeff Dean. Highlight analysis: Applying the subjective judgment advantages of humans to the evaluation shortcomings of AI. By manually crowdsourcing 500+large models, users have provided feedback that they can be exchanged for cash (1000 points=1 US dollar), which has attracted companies such as OpenAI to purchase data and has real cash flow. Personal review: Projects with clear business models are not purely money burning models. But preventing fake orders is a big challenge, and the anti witch attack algorithm needs to be continuously optimized. But judging from the $33 million financing scale, capital clearly values projects with cash flow verification more. 2、 @Gradient_HQ Project Introduction: Decentralized AI Computing Network, completed a $10 million seed round in June, led by Pantera Capital and Multicoin Capital. Highlight analysis: Thanks to the Sentry Nodes browser plug-in, there has been a certain market consensus in the Solana DePIN field. Team members from Helium, etc., have newly launched Lattica data transmission protocol and Parallax inference engine. They have made substantive exploration in edge computing and data verifiability, which can reduce latency by 40% and support access to heterogeneous devices. Personal review: The direction is very right, it happens to be stuck in the trend of AI localization "sinking". However, handling complex tasks requires more efficiency compared to centralized platforms, and the stability of edge nodes is still a problem. However, edge computing is a new demand rolled out in web2AI and also the advantage of web3AI's distributed framework. We are optimistic about using specific products with actual performance to promote the implementation. 3、 @PublicAI_ Project Introduction: A decentralized AI data infrastructure platform that incentivizes global users to contribute data from multiple fields (healthcare, autonomous driving, voice, etc.) through tokens, with a cumulative revenue of over 14 million US dollars, establishing a network of millions of data contributors. Highlight analysis: Technologically integrating ZK verification and BFT consensus algorithm to ensure data quality, and also using Amazon Nitro Enclaves privacy computing technology to meet compliance requirements. Interestingly, the HeadCap brainwave acquisition device has been launched, which can be considered as an extension from software to hardware. The economic model is also well-designed, where users can earn $16+500000 points for 10 hours of voice annotation, and the cost of subscribing to data services for enterprises can be reduced by 45%. Personal review: I feel that the greatest value of this project lies in the real demand for AI data annotation, especially in fields such as healthcare and autonomous driving that require extremely high data quality and compliance. However, a 20% error rate is still slightly higher than the traditional platform's 10%, and data quality fluctuations are a problem that needs to be continuously addressed. The direction of brain computer interface is quite imaginative, but the execution difficulty is also not small. 4、 @sparkchainai Project Introduction: Solana Chain Distributed Computing Network, raised $10.8 million in June, led by OakStone Ventures. Highlight analysis: By using dynamic sharding technology to aggregate idle GPU resources and supporting large model inference such as Llama3-405B, the cost is 40% lower than AWS. The design of tokenized data trading is quite interesting, directly turning computing power contributors into stakeholders and motivating more people to participate in the network. Personal review: The typical "aggregating idle resources" model logically makes sense. But a cross chain verification error rate of 15% is indeed a bit high, and the technical stability still needs to be further polished. However, in 3D rendering, which does not require high real-time performance, there are indeed advantages. The key is whether the error rate can be reduced, otherwise even the best business model will be dragged down by technical issues. 5、 @olaxbt_terminal Project Introduction: An AI driven high-frequency cryptocurrency trading platform, completed a $3.38 million seed round in June, led by @ ambergroup_io. Highlight analysis: MCP technology can dynamically optimize trading paths, reduce slippage, and improve actual efficiency by 30%. Catering to the AgentFi trend, it can be considered as finding a breakthrough point in the relatively blank sub field of DeFi quantitative trading, filling the market demand. Personal review: There is no problem with the direction, DeFi does need more intelligent trading tools. However, high-frequency trading requires extremely high latency and accuracy, and the real-time collaboration between AI prediction and on chain execution still needs to be verified. In addition, MEV attacks are a major risk, and technical protection measures need to keep up. Note: For more new projects on the AI+Crypto track, you can add them in the comments section. I will screen for projects with investment research value to follow up and share. Thank you.