从OpenAi到Bittensor：去中心化AI网络的范式转移

One thing to get used to when you live in Asia is that you often wake up and see heavy news. For example, you were fired last Friday. When I saw this news, I almost choked on milk. Why did the board dismiss an extremely clever success model? And this person just gave a wonderful keynote speech at the conference a few days ago, thinking that he was fired for political reasons because he would deploy the latest breakthrough. Promoting the development of artificial intelligence too fast, some people don't like this kind of company structure, which is very strange and almost dysfunctional, because it was originally a non-profit organization and later decided to transform into a for-profit enterprise. Now non-profit organizations control the development direction of for-profit entities and provide limited upside space for investors. As the truth emerges, the next few weeks will be very exciting. Will this be another Steve Jobs moment? Will you start another company and compete? But it is certain that it will be internal transportation. Although it has become an ubiquitous tool and is used by hundreds of millions of people all over the world, there is still an obvious disconnect. As ordinary daily users, we find ourselves standing outside trying to peep in through the mysterious veil around these giants. As we continue to integrate into all aspects of our society, this lack of transparency is worrying. Recently, I have been thinking about such a problem, what is the intersection between encryption and artificial intelligence? However, most people agree that this convergence will release great potential. When we think of the convergence of encryption and encryption, we usually think that these are decentralized networks, which can provide necessary calculations for model training. The logic is very simple. With the rapid development of artificial intelligence, the demand for computing resources will also increase dramatically. In this case, peer-to-peer networks may have a significant growth, so they are engaged in pickaxe and shovel business, but I think this only touches the surface of potential, which is like saying monkeys. The child is the same as the peak, and then I met with the upstream or different support model training, focusing on reasoning, and the downstream using the training model to generate output is a decentralized network incentive model, especially a large language model to deal with various tasks such as text generation, image creation and music production. At present, the network has a subnet, and each subnet is dedicated to specific tasks. Simply put, it can be imagined as a decentralized miner value producer. Miners develop and host models on the network, and they will get tokens as rewards according to the model performance related to specific tasks, which encourages the developers and verifiers of better and more efficient models to reach a consensus. Producers and verifiers evaluate miners' output and rank their performance on specific tasks. They also interact with users who submit tasks to verifiers and send them to the appropriate miners. I may have oversimplified the technical complexity, but there are several things that are obvious to me. Verifiers exchange knowledge and share parameters, which can be self-optimized over time. The network aims to make use of the advantages of multiple independent models to produce the best possible output. The expert set is the effectiveness of the network. The economic structure of tokens is similar to that of bitcoin. There is no hard upper limit and fair release of 10,000 tokens. There is even a halving cycle, and the first halving will take place in. Today, there are 10,000 in circulation, all of which are fairly distributed through mining and verification on the network. The market value is slightly higher than US$ 100 million, and the number of new releases to miners and verifiers every day is pieces. My thoughts are still in the initial stage. The network has a devout community, but the participants are still small. There are only about 10,000 active accounts, and the busiest subnet is dedicated to text generation. There are about 10 active verifiers and many miners. What is really attractive is the concept of decentralized network, which not only reduces the risk of decentralization, but also raises a question whether these unique economic incentives can be cultivated. Models developed by capital-rich entities such as Beyond and Google have become mainstream with the emergence of tools. Deep-tech startups usually focus on obtaining proprietary data and developing special artificial intelligence models based on machine learning for specific tasks, such as using real clinical data of cancer patients to develop artificial intelligence models and integrating them into tools to support cancer researchers and health care providers. The goal of startups is to productize and monetize these proprietary models. Perhaps it is more appropriate to say that this paradigm shift is a technology-driven business model innovation rather than a technological breakthrough. For example, it provides a way for proprietary data and models to be jointly developed for a wider audience without opening them up. I can imagine such a future with thousands of specialized subnets that can cope with a series of challenges, whether it is environmental health care issues or energy issues. Honestly, I found that if a team can cooperate with Bitcoin, It is very attractive to design their token economics in the same way, which shows their motivation. They are different from today's team, which often optimizes its token economics according to the model of venture capital funding, providing a large number of token allocation for founders and investors. I am not sure where to go. It may be a hundredfold success or a complete failure, but its potential and the ideas behind it are too attractive for me to be indifferent. 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=63056