Kung Chan and Xia Ri
Recently, there have been reports that Apple is in discussions with Google on a large-scale collaboration agreement that could redefine the rules of the artificial intelligence industry by integrating Google’s Gemini artificial intelligence (AI) engine into the iPhone. There are also reports suggesting that this exposes Apple’s progress in AI may not be as smooth as expected, while also potentially exacerbating the antitrust scrutiny faced by both companies. There are various opinions surrounding such an issue, with some industry giants expressing concerns; some of these opinions are sensational while others are imaginative.
However, this matter is actually corresponding to the terminal model for the future AI era of laptops previously established by ABOUND’s founder Mr. Kung Chan. This model includes features such as natural language replacing keyboards, intense competition between two major operating systems, hardware design templates trending towards the iPad, further intelligentization in software devices, and primary supporting devices being headphones and various wireless broadband interfaces. Based on this model, Mr. Kung Chan believes that Chinese smartphone brands such as Huawei and Xiaomi may fall behind a generation due to AI, becoming not unlike Nokia phones in the past.
Firstly, AI development in China is lagging behind the productization trend. During the initial stage, large-scale model enterprises were competing in terms of parameters and computational power, while the second stage focused more on commercial applications. Ultimately, AI applications need to be reflected in products, and the competitiveness of products relies on AI to bridge generational gaps. As of now, AI products have emerged in South Korea. On January 17 this year, Samsung held its annual flagship product launch event in San Jose, California, United States, and unveiled the world’s first AI smartphone, the Galaxy S24 series. Among them, the most astonishing feature of Galaxy AI is its real-time voice translation and text translation capabilities, which can almost render all simultaneous interpreters worldwide unemployed. According to Samsung, this feature currently supports 13 languages, including Chinese, Korean, English, French, German, Japanese, etc., and to some extent, it breaks through human language barriers. Therefore, when Apple saw Samsung’s products and realized that this sets the standard for generational divisions, it sought cooperation to catch up with Samsung.
Otherwise, it would fall behind the times completely. In comparison, China is still at the stage where it considers itself to have achieved technological superiority, yet nothing is further from the truth. In 2023, influenced by ChatGPT, large-scale AI models in China showed explosive growth for a while. Since large-scale models are not easy to measure intuitively and there is no objective evaluation method, major model enterprises claim that they are doing well. In fact, a number of multimodal large-scale models in some Chinese enterprises, such as Tencent’s Hunyuan, Baidu’s ERNIE Bot, Huawei’s Pangu, Alibaba’s Tongyi Qianwen, 360 Zhinao, and others, most solutions are limited to integrating multiple modal information at the input end. Hence, there are hardly any of them that can truly generate multimodal content at the output end. Even if they are being applied, they mainly focus on leisure and simple tasks with high fault tolerance. In more valuable and serious fields, as well as in work and professional scenes, such usage by Chinese AI models is still in lacking. More importantly, there has been a notable absence of substantial Chinese AI products, highlighting a significant lag in development.
Secondly, the technological monopoly abroad exposes the gap in China’s AI development. OpenAI, with its huge talent pool, massive financial support, years of sustained investment, and commitment, took over eight years to develop the breakthrough product GPT-4 and is continually improving on it. Meanwhile, China’s large-scale model development just started not long ago, which is mainly divided into three categories: original large-scale models, shell-out models from overseas open sources, and assembled models. The current situation in the country is that many of the players are ostentatiously competing in developing the so-called original large-scale models but are actually concerned about the high risks involved, leading to a large number of them producing shell-outs and assembled large-scale models instead. As long as foreign advanced open-source models shut down their interfaces and prevent some Chinese large-scale model enterprises from training their own models, the actual situation of China’s AI will naturally be revealed. The most typical example is ByteDance. In 2023, ByteDance, in its secretly developed large-scale model project, used OpenAI’s application programming interface (API) and data output by ChatGPT for model training. After the media reported about it, OpenAI suspended ByteDance’s account, stating that it would conduct further investigations and, if confirmed, would require changes or termination of the account. Eventually, ByteDance had to stop using the GPT API service for experimental project research. Even if foreign open large-scale models are available, China’s AI still faces the problem of computational power blockade. The mainstream market in China mostly adopts NVIDIA’s A100 and H100 chips as related training equipment for computational power. According to OpenAI’s public data on ChatGPT, its entire training computational power consumption, considering interconnection losses, requires ten thousand A100 chips as the computational power base. In August 2022, the United States issued a policy prohibiting NVIDIA from selling the A100 and H100 AI chips to China. As a result, only the H800 and A800 versions of chips were exported to China, with only about sixty percent of the original capabilities. To make matters worse, due to limited production capacity, such chips have quickly entered a state of having value but no market.
Thirdly, the unique information environment in China hinders the growth and advancement of AI in the country. The “technical level” of AI typically refers to the ability of large-scale models, which rely on long-term self-learning to achieve. This requires massive amounts of information for the maturity of AI to empower various industries. As it stands, large-scale models in China mainly rely on Chinese data. In the past, data collection was primarily done using web crawlers, but now open-source datasets are available. Although Chinese internet companies have a vast amount of data from e-commerce, social media, search engines, and other networks, the types of data are not comprehensive enough, and the credibility of online knowledge lacks strict assurance. Additionally, professional data services in China are still in their infancy, and high-quality datasets processed for AI model training are relatively scarce. Furthermore, the Chinese market lacks effective data protection measures. An AI leader at a major company stated, “In China, for the data you can get, others will be able to get the same thing as well. If you spend a lot of money to obtain high-quality data, others can obtain it at a much lower cost, and vice versa”. While these issues may be gradually addressed, the more critical challenge is that due to the firewall system set up on the Chinese internet, obtaining open, vast, and comprehensive information required for long-term learning and training of large-scale models is difficult. Consequently, large-scale model training and learning are confined to the restricted range of information within China. Once beyond the Chinese border, while information may be available, it is certainly restricted. Without specialized, standardized, and open information data for continuous optimization and debugging, achieving greater breakthroughs in AI in China is challenging. Therefore, the application of AI in the country remains highly questionable.
Final analysis conclusion:
In the new wave of artificial intelligence (AI) development driven by the U.S., the Chinese industry is striving to catch up. However, although the industry in China claims to possess technological prowess, the actual development situation shows that it is lagging in many aspects. Firstly, it lags in the trend of AI productization. Secondly, it is constrained by foreign technological monopolies, revealing the true state of AI in China. Thirdly, the unique information environment impedes the growth and advancement of AI in China. Considering the competition in terminal products, Chinese smartphone brands may fall behind a generation due to AI, becoming the Nokia phones of the new era, effectively rending them into non-smart phones.
No comments:
Post a Comment