DePIN打猎之旅：AI算力作饵、道阻且长

The author's Hong Kong Carnival has come to an end, but the pulse of freedom is still beating and infiltrating into other industries. Compared with the last cycle, the logic of this bull market has changed from the original innovative narrative to the mainstream approved fund-driven model. The development stage observed by the author has also evolved from a closed and niche absolute freedom to a truly inclusive relative freedom stage. Under this logic, it is no longer possible to jump out of the box to analyze the original innovative narrative. Embracing compliance has begun to refocus the financial sector under the constant promotion of the Hong Kong Government, and mainstream financial institutions have also accelerated their participation in the conference through spot. In addition to the entry of mainstream financial institutions, we have also seen an opportunity to connect and harmonize the track, especially the promotion of the development of large models, which has made the redistribution of computing power in the sub-track once again a hot source of concern. However, the training of large models is not the best landing scene. Blockchain is to build trust through technology and is an extremely needed one. The managing partner of the industry said at the conference that it was not a new track. It was proposed several years ago that it was the big model that broke out. The industry had a lot of discussions on computing power and data. According to estimates, the cost calculated by the big model doubled every year, and there was a global shortage. Companies like NVIDIA were at the top of the food chain in the current market demand, and there was a great monopoly on pricing power or decentralization. The debate about the cost became the reason for the hot discussion again. Although the big model training was the cause, Rome was not built in a day. Model training is not the best landing scene at present. The demand for computing power in large-scale model production mainly revolves around three aspects: reasoning and training. In the training process, a complex neural network model is trained by feeding a large amount of data. In the reasoning process, a large amount of data is used to infer the combination of decentralization and computing power. The difficulty coefficient decreases step by step from training to fine-tuning training to reasoning. It can be seen that more projects in the industry focus on reasoning rather than training. The main reason why most enterprises adopt NVIDIA cluster for training is that it has its own powerful parallel computing ability and memory bandwidth's requirements for parallel computing ability and bandwidth are much lower than those of reasoning links, and large-scale model training pays more attention to stability, because once the training is interrupted, it needs to be retrained. If a decentralized computing application is built on the Ethereum for use, only a single matrix multiplication operation will cost hundreds of millions of dollars and it will take another month. This paper analyzes the current situation of several popular projects in this conference, showing a trend that the supply side exceeds the demand side, that is, the supply of computing power distributed around the world exceeds the demand of developers for model training or reasoning tasks. It does not mean that the founders with no demand put forward to raise trillions of dollars to build an advanced chip factory that is twice as large as TSMC and used for chip production and model training. Stanford University research also shows that no matter which language model has training parameters that exceed that scale. After the critical value, its performance, such as accuracy, will be greatly improved, which is contrary to the law of making great efforts to make miracles. It also means that there are still many problems to solve the idea of centralized computing power in reality. The historical traceability of the track can be traced back to the sharing economy itself. The concept is not difficult to understand and even can be traced back to the review of the Internet industry. At least for a year, players are immersed in aggregating personal tangible assets to create a sharing economy. If intangible assets such as idle servers are directly passed through, The form or form of redistribution to the demand side to decentralize the technology blockchain can optimize the production relationship through the incentive mechanism, which is what we have to do. Therefore, the enthusiasm of everyone on the supply side below the track is high. In fact, it has been accumulated for a long time for redistribution, but this time, the intermediary is directly removed. At present, there are nearly 1,000 projects, especially ecology. According to statistics, ecology is in a leading position in infrastructure. This is because the public chain has high infrastructure integration and performance. In terms of regional distribution, it is predicted that many top names will come from Asian sources from to and there will be a lot of cross-point computing power as the common currency in the future digital world. However, the most reasonable landing scene of decentralized computing power is not the easiest to land. At the cross-point of and, in addition to technical difficulties, there are many other branches, such as giving the creator ownership of agents and small-scale model computing power scenes, which are worth exploring and will be more grounded. The success of business models and technological breakthroughs will always be balanced, and the hunting trip that is accelerating this process will return home with full load. 比特币今日价格行情网_okx交易所app_永续合约_比特币怎么买卖交易_虚拟币交易所平台

注册有任何问题请添加 微信：MVIP619 拉你进入群

文章转载注明 网址：https://netpsp.com/?id=54989