Machine deep learning revolution
The machine deep learning revolution has roughly gone through three waves, each bigger than the last. The first wave, around 2010, saw breakthroughs in speech recognition, replacing Markov models with deep learning, with unprecedented results; The second wave is the breakthrough of convolutional neural network in image recognition in 2012, which is beyond people's imagination; The recent third wave of language models, close to or even beyond the human level, has overturned the past 40 to 50 years of computational linguistics, natural language processing research, is a revolutionary technological breakthrough.
The key to large models is that there is a pre-training step in the method, where previously machine learning was either human-directed or machine-crafted. Pre-training is to build a common sense system of the world, based on the rules of how the world works. This connection is constructed through common sense, and the effect is beyond imagination.
The Big model is a new methodology that gives AI human-like intelligence. In fact, the large model is to input the cross-domain and multi-industry human knowledge system into the mathematical model of the human-like brain neural network through super-scale computing power, and through the computing paradigm, it has content production and logical reasoning capabilities in many fields.
Our country is still chasing a stage
In the new round of AI competition, can China keep up quickly? Building a large model requires at least thousands of chips, large computing power requires interconnection between chips, and interconnection requires ultra-high-speed bandwidth. At present, China's large model in the overall comprehensive strength and innovation development, compared with the head enterprises still have a big gap.
Large models also have application and cost issues. Currently, the cost of the chip, especially the cost of training and inference, is high.
According to reports, AI chips can be divided into two categories: training chips and reasoning chips. For large models and AI algorithms, training is not the goal, inference and application are the ultimate goal. In terms of training chips, many enterprises in China are still in the catch-up stage due to the limited production and manufacturing equipment and processes. There are many opportunities in the field of reasoning, because of the fragmentation of the scene, there are no international standards and monopolistic enterprises.
Artificial intelligence is an industrial revolution, and everyone is very concerned about how artificial intelligence can be combined with industrial digitalization to achieve industrialization, enterprise, specialization, and vertical. Although it takes time to catch up on the original innovation, our country has a large population, many enterprises, and many scenes, which will produce many innovation opportunities.
The industry is expected to usher in an explosion period
Innovation requires the power of ecology, the power of curiosity. In the past two years, there have been many disruptive innovations, and more scientific and technological innovations will emerge in the future. China's application scenario data volume is large, but it lacks a set of industry standard system, and the differentiation advantage is not really played out through technical tools. It is necessary to use industry standards and technical standards to measure the goals and effects of opening various scenarios.
When each new generation of technology appears, the biggest beneficiaries are the people who are most sensitive to technology, and the key is to combine industry needs with technology to do the best. Artificial intelligence is an extremely efficient production tool that will soon have an impact on thousands of industries.
In the future, each person may have three robots: a service robot at home, a work robot at the office, and a travel robot - a driverless car or a low-altitude manned drone. These three robots may be driven by large models or new technological paths and AI capabilities evolved from large models, and we hope that they all have chips that we independently developed. In the future, robots will help us work, live and travel better. Around artificial intelligence, digital economy, next-generation chips, and application scenarios, the entire industry is expected to usher in an explosive period in the future.





