Chainfeeds Introduction:
Artificial intelligence is one of the hottest and most promising narratives in the crypto market recently. Ed Roman, managing partner of Hack.VC, wrote an article analyzing the real-life use cases of combining cryptocurrency with artificial intelligence and analyzing its challenges and opportunities.
Article Source:
https://x.com/ed_roman/status/1803922033820193226
Article author:
Ed Roman
View:
Ed Roman: Use cases where cryptocurrency combined with AI can bring significant value: 1) McKinsey estimates that generative AI can bring $2.6 to $4.4 trillion in value to various industries each year, equivalent to 1.5 times the UK GDP in 2021; 2) Reduce the cost of GPU use through GPU DePINs. GPU DePINs aggregates underutilized GPU computing power for AI reasoning. This is similar to "Airbnb for GPUs", reducing the cost of AI reasoning by providing lower-cost GPU computing resources; 3) Open source models can avoid self-censorship of AI content by companies like OpenAI, thereby supporting more types of applications. 4) Large companies are worried about handing over internal data to centralized third parties. Web3 can protect corporate data by enhancing privacy technologies such as trusted execution environment (TEE) and fully homomorphic encryption (FHE). 5) Open source software (OSS) models continue to innovate and can replace proprietary software. Through Web3 AI, leveraging these open source models can bring more innovation and economic benefits. 6) Web3 AI reasoning requires verification to prevent validators from cheating. ZK proofs and random sampling combined with high penalty cost methods can effectively prevent cheating and improve the reliability of consensus. 7) Save costs through composable OSS stacks. Web3 can also save costs by using open source models that do not need to make profits like proprietary software. 8) Obtaining data through a decentralized network can improve the data quality and timeliness of AI model training. Startups like Grass are exploring this approach to improve the efficiency and coverage of data acquisition through a decentralized data collection network. Challenges of combining Web3 with AI: 1) Decentralized AI training: The main problem with AI training on the chain is that high-speed communication between GPUs is required, and decentralized networks increase latency and bandwidth costs; 2) Decentralized AI data iteration: AI training requires processing of large amounts of data, which are usually stored in centralized and secure systems. Data processing and iteration in a decentralized environment is very difficult, especially in the absence of existing best tools and frameworks. 3) Redundant computing consensus for AI reasoning: To ensure the accuracy of AI reasoning results, the idea of repeated computing was proposed, but the shortage of high-end AI chips makes this method costly and difficult to promote. 4) Web3 specific AI use cases: Currently, the Web3 specific AI use case market is still in its infancy, with low demand and unstable customers, which increases the difficulty of business expansion; 5) Consumer GPU DePINs: Decentralized AI computing networks rely on consumer-grade GPUs, which are suitable for low-end AI reasoning tasks, but for enterprise use cases that require high reliability and high-end GPUs, data centers are more suitable. [Original text in English]
Content Sources 
