After the "Hundred Regiments War" of the Chinese large model, users experienced aesthetic fatigue.
In particular, the performance of the large model "Internet celebrity" chatGPT has dropped significantly, the chat response speed has slowed down, the content has become verbose, repetitive and illogical, and the chat topic has become single and uninteresting, which has been complained by users.
The crazy involution of domestic large models has made some enterprises lacking technology accumulation and R&D investment dissatisfied and began to fall behind. After the capital operation, there was no grounded application scenario, and gradually began to lose its voice.
However, on August 15, iFLYTEK Spark Model 2.0 arrived as scheduled. After the live real-world demonstration, the code capability and multimodal interaction ability exceeded the phenomenon. Liu Qingfeng, chairman of iFLYTEK, said that iFLYTEK Spark Model V2.0 has a high degree of code writing capabilities based on Python and C++, and will fully surpass ChatGPT on October 24 this year, and will officially benchmark GPT4 in the first half of next year.
As the only technology company in China that accurately gives the timetable of large model development plans and leads the direction of technology, where does iFLYTEK's confidence come from?
01 Why does iFLYTEK Xinghuo cognitive big model stand out?
iFLYTEK Spark big model can benchmark and surpass ChatGPT, unlike those hype ChatGPT concepts to seek "breaking the circle", iFLYTEK has let the large model combine with business cases from the beginning, so that the big model can be touched and seen. Any entrepreneur, business owner, or individual can get involved.
First, mature product plan and commercialization ecology.
On August 15, the Spark 2.0 model was launched, and compared with the Spark 1.0 and 1.5 models released in the first half of the year, the code capabilities and multi-modal interaction capabilities of the large model will be significantly upgraded. It is more powerful in text generation, language understanding, knowledge question and answer, logical reasoning, mathematical ability, code ability, and multimodal ability, and is more mature in the four major industries of education, office, automotive, and digital employees.
iFLYTEK demonstrated that Spark 2.0 can realize development capabilities such as code generation, code completion, code correction, code interpretation, and unit test generation. Based on this, iFLYTEK announced an intelligent programming assistant based on Spark 2.0: iFlyCode 1.0, which has all the code generation related capabilities of Spark Model and can be seamlessly integrated into the development environment.
Liu Qingfeng said that the iFLYTEK Spark model has been well received by the majority of users and developers, and in the past 100 days, 700,000+ new developers have been added to the iFLYTEK open platform, and 7800+ assistant users have been created.
The Spark model empowers C-end hardware and creates a new experience. iFLYTEK intelligent notebooks, learning machines and voice recorders and other terminals won the sales champion of 7 categories of JD.com and Tmall dual platforms on 6.18 this year.
Over the years, HKUST has been a leader in the AI and voice industry, building a high moat in intelligent education hardware and education informatization. Previously, iFLYTEK launched a special research project on "1+N cognitive intelligence large model technology and application", of which "1" refers to the general cognitive intelligence large model platform, and "N" refers to the application of cognitive intelligence large model technology in education, medical care, human-computer interaction, office, translation and other industries. For example, iFLYTEK Spark model + education, + medical, + finance, + automobile, opened up the market imagination.
For example, iFLYTEK Spark Model 2.0 has improved its multimodal interaction capabilities, which can realize multimodal functions such as image description, understanding, reasoning, creation, text and image generation, and virtual human synthesis. Multimodal capabilities can not only empower education and healthcare, but also apply to industrial applications such as industry, automotive autonomous driving, and robotics, which is the only way to realize a universal large model.
At the Spark 2.0 press conference, iFLYTEK announced the launch of iFLYTEK Spark all-in-one machine with Huawei. "We not only have full confidence in the algorithm, but also taken an important step in computing power." Liu Qingfeng, chairman of iFLYTEK, said at the press conference that the iFLYTEK Spark all-in-one machine will make the privatization deployment of large model capabilities more convenient, more autonomous, safer and controllable for all enterprises, and carry out privatization deployment on a completely domestic independent innovation platform to protect their own data security.
Previously, iFLYTEK and Huawei jointly created an AI storage base to ignite the spark of AI models. Zhou Yuefeng, President of Huawei's Data Storage Product Line, said that in the era of large models, data quality determines the height of AI intelligence, and as a carrier of data, data storage is the key infrastructure of AI large models. Facing the AI era, Huawei and iFLYTEK are working with iFLYTEK to build a smarter digital world.
iFLYTEK and Huawei have sought breakthroughs in data storage service reliability, computing power and storage power collaboration system research, and data lifecycle management based on joint innovation. Through the innovative practice of AI storage power in the era of AI large model, it provides high-value reference experience for the industry.
Second, strong technical precipitation and continuous R&D investment in AI.
Similar to the start of 1 billion cores, the large model is also a very money-burning tool.
The cost of training and debugging large models is very high, which may be difficult for a company with no foundation, but iFLYTEK has strong R&D support behind it.
In 2022, iFLYTEK's total R&D investment will be 3.36 billion yuan, a year-on-year increase of 14.28%. The number of personnel accounted for about 60% of the total number of people in the company. Such a huge investment is enough to see iFLYTEK's determination to build a large model of "aircraft carrier". In 2022, the company won a total of 16 international artificial intelligence competitions in the field of key core technologies of artificial intelligence, including 13 championships in the field of cognitive intelligence.
Liu Qingfeng said that because each key technical module of the Xinghuo model is completely independently developed, iFLYTEK is controllable for each stage of technology, and the software and hardware platforms are all domestically independent and controllable.
Third, China's large model has taken the lead in technology and application development.
Recently, Xinhua News Agency Research Institute released the domestic large model report (hereinafter referred to as the "Report") shows that iFLYTEK Spark is the first, with the highest IQ and the strongest tool efficiency improvement ability, ranking first in the domestic mainstream large model evaluation list with a total score of 1013 points, and winning the first place in the two dimensions of IQ index and tool efficiency improvement index among the four major evaluation dimensions.
iFLYTEK Big Model broke through open-ended Q&A on June 9, and iteratively upgraded its mathematical ability and multi-round dialogue ability; On August 15, a major breakthrough was made in code capabilities and upgraded in multimodal interaction capabilities; On October 24, ChatGPT was benchmarked against ChatGPT, achieving a level of Chinese surpassing and equivalent English.
After October 24 this year, it will fully surpass ChatGPT, allowing China to achieve an all-round lead in the development capabilities and scenarios of large models. iFLYTEK has been working hard in AI for 24 years, and Liu Qing previously proposed a five-year goal: to reach one billion users, achieve hundreds of billions of revenue, and drive a trillion industrial ecology, which is the real value of iFLYTEK's AI in model.
02, group model dance, how to develop healthily?
As more and more large model companies and AI applications surface, how can the domestic large model industry be healthy and sustainable under the "group model dance"? It has become an issue that more and more people pay attention to and think about.
Up to now, more than 100 large-model products have been launched in China, of which there are as many as 26 enterprises that only position general large models and "general + industry". In addition, there are 7 "basic + industry large models"; There are 67 large models in other financial, medical, biological, transportation, data analysis and other industries.
At present, there are three main modes of large models:
(1) Startups that have a big model for the sake of a big model, without any technical precipitation and business foundation.
ChatGPT is such a company without technology accumulation and cultural precipitation, but there are giants including Google to support with financing, so instant noodles are getting bigger and bigger. More startups around the world aren't so lucky. For example, after experiencing a vigorous financing, Lightyear Company was "cleaned up by Meituan".
(2) There are Internet genes, there is no technical precipitation in large models, temporary formation of teams or acquisition of some companies, labeled as large models, on the one hand, to increase the stock price, on the other hand, to promote the business, but can not see the real liquidity ability.
There are many such AI concept stocks in A-shares, and the wind has been chicken feathers in the past. Just like the Internet + that blossomed everywhere, and in the Internet era, those Internet + only moved items from offline channels to online, without any changes.
(3) Such companies have deep cultivation in the AI industry and a strong technical team.
For example, iFLYTEK and Huawei in the communications industry. The purpose of entering the big model is very simple, that is, to combine the existing business and empower the front lever industry, talent, not for the sake of the big model.
03, the future of big models: the era of deep empowerment from entertainment to scene practicality has arrived
ChatGPT was born from the United States and then took the world by storm.
The first thing that comes to mind is to let the robot serve me, so I constantly ask the big model all kinds of questions that humans can think of, trying to "examine" it, so as to have a sense of gain. Answering all kinds of boring questions asked by humans and taking all kinds of exams is almost a machine to have children instead of humans.
Entertainment is one of the driving forces of scientific and technological progress, but it is definitely not the spiritual opium of Internet games.
In fact, humans are so difficult and testing a machine, not only curiosity, but more importantly, let robots create wealth for society, which is the most important value of the big model.
Therefore, any large model that cannot be commercialized is laborious and has no vitality. From a small company to a large country. Only high-tech that can make the country strong has long-term vitality. Looking at the world, only China can become the second pole of "wisdom emergence" outside the United States, and China must have its own national team of general artificial intelligence, which is crucial to the country's livelihood security and industrial chain security.
5G and the Internet of Everything have brought the era to come, and the big model has ushered in a 5-10 year period. According to the "Research Report on the Map of Chinese Artificial Intelligence Big Model of Intelligence" released by the New Generation Artificial Intelligence Development Research Center of the Ministry of Science and Technology, the number of large models developed in China and the United States accounts for more than 80% of the global total. Among them, 79 large models with a scale of more than 1 billion parameters in China have been released. In addition to large models with a scale of more than 1 billion parameters, many smaller and lighter models are also emerging, bringing more application possibilities to various fields.
We know that the general large model has always been the leading position in the United States, how Chinese companies break the "stuck neck" problem in the United States in terms of computing power, and iFLYTEK large model has taken a substantial step in terms of safety and controllability. As Liu Qingfeng said, we have taken the first step in independent innovation.
It is worth pointing out that in order for China's big model to be in the leading position in the new era, it cannot be completed by just one company or a certain group, but the whole society keeps pace with the times, actively embraces the AI era, and actively participates in order to keep up with the trend of the times.