Imagine the combination of AI computing and computing power mining model

Recently, there was a sensational news in the field of AI. A team called OpenAI created the GPT-3 AI model, using 175 billion parameters, 45 terabytes of data, and spending $12 million to create a large "toy". Of course, this GPT-3 is not as smart as everyone imagines. There are reports that some English articles generated by the GPT-3 model based on keywords have some grammatical errors and repeated use of multiple words, but compared with previous AI models, it is already much more advanced.

AI has been a relatively hot industry in recent years. The explosion of AI began with Google's AlphaGo defeating Lee Sedol in Go. Of course, as a landmark game, it is of far-reaching significance. After AlphaGo, there are many more very powerful AI Go players, and many science fiction films with AI as the theme have been staged, and the AI ​​craze has officially begun.

The success of AlphaGo is mainly due to the maturity of artificial neural network technology. Simply put, this is a model that uses existing data for training, extracts features from the original data, and then performs various linear calculations or other matrix calculations, etc., to finally generate new data. By feeding and calculating data multiple times, "great effort can make miracles happen." Of course, the amount of calculations required here is still very large.

There are roughly two main directions of AI: one is algorithm and the other is hardware. The development of algorithm in recent years is relatively good, but due to the limitation of hardware, the development of AI is still relatively slow. When used in business, it can only simply replace the previous telephone customer service. There may still be a long way to go to reach a deeper level.

At present, AI training is mainly carried out using hardware such as GPU. Computing power is a very critical factor in the development of AI. As for computing chips, they are still relatively weak. Training a model often costs a lot of money and takes several months or even more than a year. Even after a bug occurs, the team dare not recalculate. Therefore, the market demand for AI computing power is still very strong, which has also led to a large number of chip design companies starting to study AI chips.

We can understand AI chips as a special chip, just like the CPU, GPU, FPGA, ASIC and other chips used in digital currency mining. They all need to perform specific tasks and then return results. In the proprietary computing chip industry, we all know that Bitmain is very well-known. For example, the mining chips for digital currencies such as Bitcoin are dozens of times more efficient than ordinary general-purpose chips. Therefore, for mining chip research and development companies, if the computing power of mining chips can be applied to the field of AI chips, then this will have a greater impact.

Now, not only Bitmain, Canaan Technology, etc. have begun to expand part of their energy into AI chips. In fact, they are also taking advantage of their own advantages. Once ASIC chips are used in a certain link of AI calculations to replace existing general-purpose chips (artificial neural networks use a lot of mathematical calculations, such as image processing, which theoretically can be ASICized), then this will also bring about rapid development in the field of AI. Its prospects are naturally very impressive. You must know that the current computing power of the entire Bitcoin network has actually exceeded the total computing power of the world's supercomputers, and the efficiency is also very high.

By comparing the AI ​​market and the mining market, we can see that it is basically difficult for Bitcoin's dedicated computing chips to significantly reduce power consumption. In other words, the competition among mining machines in the future may still focus on process costs. The room for improvement of advanced low-power computing chips will gradually decrease in the next few decades, so the market will definitely become saturated in the future.

If there are no new excellent mining coins, it will be a bad thing for mining machine manufacturers. In the field of AI, the algorithms are now relatively mature and data acquisition is relatively easy. At present, it is mainly stuck on the computing power chip, and computing power has become the biggest difficulty in the field of AI.

For the mining field, if AI chips are used in the future, a mining farm computing model similar to the current one can be born. We can boldly imagine that enterprises need to use AI algorithms to perform relevant simulations, and then assign tasks to mining pools using a model similar to digital currency mining. The mining pools then decompose the tasks, allowing a large number of mining farms to run AI calculations while the mining farms make profits. This forms a perfect closed loop. At the same time, in this link, the cost of related calculations can be reduced and the calculation time can be reduced, thus winning an advantage for enterprises.

At the same time, we all know that although most of the current high-tech technologies, such as 5G, artificial intelligence, big data, cloud computing, AI and other industries are very promising, there are very few channels for the public to participate. In other words, most of us can basically only be consumers, not contributors to technology. The difference here is that the development dividends brought by these high-tech technologies have little to do with most of us. Even if there are ways to participate in the stock market, it still seems unfriendly.

Through this mining-like model, ordinary people can actually purchase devices with AI computing power, and then perform related computing tasks at home or in a mining farm. The final income is also proven based on the workload of the machine, which enables the public to truly participate in the AI ​​industry chain.

In fact, some teams have already experienced this model during the COVID-19 pandemic. According to media reports, earlier this year, Ethereum miners donated their computing power to conduct research related to COVID-19, which was a good start.

There is a well-known saying in the cryptocurrency circle: "computing is power". Computing power is power. In the future, whether in digital currency mining or AI, there will be a lot of computing. The distributed operation of computing has greatly increased the participation of the public and promoted the development of society in a more intelligent direction. At the same time, participants can also obtain computing rewards. In the final analysis, whoever controls the computing power controls the future. From this perspective, dedicated task computing plants such as mining farms will also have social value and significance in the future.