The construction of general artificial intelligence accelerates, and listed companies seize the opportunity period of industry empowerment

On May 8, many ChatGPT concept stocks rose. iFlytek recently released the Spark cognitive large model. Vertical industry companies such as Tianyu Digital, Hanwang Technology, and Vision China have disclosed the progress of AI applications, which has become the epitome of the acceleration of the "universal large model + application" route of artificial intelligence in China at this stage. .

WeChat picture_20230506141224

Industry insiders said that driven by general artificial intelligence, the trial cycle of a series of innovative technologies such as game engines, digital humans, and edge computing power will be shortened. The emergence of various open source models and vertical models will catalyze "quantitative change to qualitative change", allowing All walks of life will use and be able to use general artificial intelligence.

General large model file upgrade acceleration

On May 6, iFlytek released the Spark cognitive large model and announced the "1+N" product path based on the large model. Solutions for industries such as education, office, automobile, and digital employees were also unveiled. It is understood that iFlytek's large cognitive model has capabilities in seven dimensions, including text generation, language understanding, knowledge question and answer, logical reasoning, mathematical ability, programming ability, and multi-modality. "In 2011, iFlytek undertook the construction of the National Engineering Laboratory for Speech and Language Information Processing. We proposed to make machines listen and speak like humans. This large-scale cognitive model is essentially the emergence of conversational general intelligence. The core ability of speech and language is a basic condition." Liu Qingfeng said.

Domestic general-purpose artificial intelligence models are accelerating in the competition with each other. Before iFlytek, Baidu, Alibaba, and SenseTime released ChatGPT products and corresponding large models of "Wen Xin Yi Yan", "Tong Yi Qian Wen" and "Riri Xin SenseNova" respectively. Baidu Group founder Robin Li said in an internal event that "Wen Xin Yi Yan" has completed four major technical upgrades in more than a month of internal testing. Alibaba announced that in the future, all Alibaba products will be connected to "Tongyi Qianwen". In addition, Alibaba Cloud also stated that it will open the "Tongyi Qianwen" capability to the public to help companies create exclusive large models.

In addition to general large models, many companies with the ability to develop large models have also developed industry-level large models. On May 5, Xueersi announced that it has developed a large self-developed mathematical model called MathGPT. On the same day, NetEase Youdao showed an AI speaking teacher video developed based on the "Ziyue" large model. According to incomplete statistics from a reporter from China Securities Journal, more than 30 large models have emerged so far.

It has been half a year since OpenAI launched ChatGPT in December last year. During this half-year, the construction of domestic AI large models has gone out of the line of progressing hand in hand with general large models and industry large models.

The market is more concerned about when domestic general large model capabilities can catch up with ChatGPT’s large model level. At the launch of the iFlytek Spark Cognitive Model, iFlytek Chairman Liu Qingfeng said that the current Spark Cognitive Model has surpassed ChatGPT in three major capabilities: text generation, knowledge question and answer, and mathematical ability. The future goal is to surpass ChatGPT in Chinese and reach a comparable level in English. He further revealed that iFlytek's general-purpose large model will undergo three major upgrades this year, and will overall catch up with ChatGPT by the end of October.

Zhou Hongyi, founder of 360 Group, said in a live broadcast on May 7: "Domestic players initially adopted a strategy of imitation and following, but in the process of gradually investing in research and development, as the scenarios and users broadened, it may even be possible in the future Achieve overtaking in corners.”

More companies have a conservative attitude towards catching up with ChatGPT. Robin Li said at Baidu’s internal awards interaction that the challenges facing Baidu are unprecedented. Whether it can catch up with ChatGPT depends on Baidu's own efforts, ability to seize opportunities, and execution.

Accelerate industry empowerment

AI large model applications are blooming in all walks of life. Tianyu Digital announced on May 8 that the company has recently released the company's first zero-threshold 3D virtual digital human intelligent application platform "MetaSurfing New Version", which is connected to pre-trained large models and incubates "MetaSurfing" with AIGC technology as the core. "Interesting" content generation application, realizing Wensheng texts, Wensheng pictures, and Wensheng audio through TTS.

Visual China announced on May 8 that the company has recently launched a creative tool based on generative AI technology. Users can use the "model conversion" and "illustration conversion" functions in the tool to re-create copyrighted images online to improve the efficiency of content generation. At the same time, it can also increase content diversity.

Hanwang Technology stated in the investor relations activity records disclosed on May 4 that at this stage, whether it is software products or hardware products, the company will continue to strengthen the implantation of AI capabilities; a lot of the work done now is also preparing for the future. Service robot prepares. In the direction of "AI + intelligent hardware", the company has recently released products such as AI electronic paper study books, Korotkoff sound sphygmomanometers, and image recognition electronic guns PK Gun; the developed intelligent bionic robot bird has been commercially sold.

In addition, listed companies are also paying attention to the opportunities that AI brings to the design industries such as construction, automobiles, games, and financial services to reduce costs and increase efficiency. Youzu.com said in a research event that the biggest cost for game companies is labor cost. AI can help producers improve efficiency, which actually reduces costs. As a game company focusing on "card + strategy", the company's art costs such as 2D concept setting account for more than half of the art research and development cost structure, and AI is relatively mature in 2D art applications. In April, the company outsourced 2D art Costs have already been reduced by about 25%, and greater cost reductions are expected to be achieved by the end of the year.

Li Xunlei, chief economist of Zhongtai Securities, said at a media event on May 6 that we are currently in the stage of global technological development and explosion. AI may trigger a new round of technological revolution, and the global economy is expected to usher in new impetus. Or it will bring about a reduction in labor costs and an improvement in labor productivity. "AI+" will bring about a new profit model and form a large enterprise.

Guosheng Securities said that the opportunities in the implementation of general large-scale models should not be underestimated. When facing B-end users, it is not only the API interface of the large model that needs to be called, but also many issues such as data privacy, computing power deployment, response speed, usage cost, etc. need to be considered. This is the process of truly turning a general large model from a toy into a tool. More stitching and innovation in engineering implementation are needed, which may be the next stage of opportunity.

Pay attention to AI risks

Under the ChatGPT craze, while AI applications are accelerating, problems have also arisen.

The first thing to bear the brunt is the frequent cases of using ChatGPT to generate false information. Generative AI may become an entry point for the proliferation of false information. Fan Yuan, chairman of Anheng Information, recently said that AI and large models themselves will bring new challenges to network security. Some automated generation tools will significantly increase the probability of viruses, Trojans, backdoors, etc., perhaps even hundreds of times. , the challenges we face at this time are beyond doubt.

What’s even more noteworthy is that Hinton, the 2018 Turing Award winner and the “father of deep learning,” recently resigned from Google and expressed concerns about the problem of out-of-control AI.

All these factors have forced the industry to carefully consider whether to develop general large model technology at a faster speed and how to avoid risks in the implementation of AI applications.

Chen Jiachang, director of the High-tech Department of the Ministry of Science and Technology, said in February this year that the Ministry of Science and Technology established the Artificial Intelligence Planning and Promotion Office, the Strategic Advisory Committee and the Artificial Intelligence Governance Professional Committee. In the past two years, it also formulated the "New Generation Artificial Intelligence Governance Principles" and " "Ethical Code" and launched major scientific and technological projects on artificial intelligence.

The increasing emphasis on AI security has given rise to new development opportunities in the network security industry. Network security vendor NSFOCUS Technology stated that the company’s “AI + security” product is expected to be released in the third quarter.

"The network security industry is seeking the intelligent application of 'AI+security', which will bring greater opportunities and results for intelligent protection and response. I think the curtain of 'AI+security' has just begun, and there will be greater success in the future. possibility." Fan Yuan said.