Thanks to the rapid development of AI today, the market demand for AI computing chips is extremely strong, making Nvidia's market value surpass Microsoft and Apple in one fell swoop, becoming the world's most valuable listed company.
With the shareholder trend of AI, NVIDIA has developed from a company in the semiconductor field to an important "basic equipment supplier" that is indispensable in today's AI era. It can be seen that AI will enter all industries in the future, and those companies that embrace AI may win and survive, and those that do not even disappear.
In China, technology companies such as Baidu, iFLYTEK, SenseTime, Huawei, Alibaba and other technology companies have deployed AI large models, and have successively launched new large model products and AIGC applications since March last year. ByteDance's main AI intelligent assistant "Doubao" and multi-modal large model BuboGPT were not launched until August. There is no doubt that compared with domestic competitors, ByteDance has fallen behind in the field of AI.
One of the big factors why ByteDance will fall behind is: "big company disease". Earlier, according to official information, ByteDance CEO Liang Rubo criticized at the company's annual meeting that many people within the company have shown signs of complacency, and ByteDance will not start discussing GPT until 2023. The second is the success of Douyin in terms of user scale and commercialization in the past few years, which has made ByteDance focus a large part of its energy on commercialization, and its observation of various cutting-edge technologies has been reduced.
However, in the first half of this year, Douyin began to tilt traffic and resource promotion for its AI application tool "Doubao". So in the era of AI, will ByteDance's miraculous play still work?
01. ByteDance accelerates its entry into the AI battlefield
In the era of mobile Internet, ByteDance is good at creating popular applications, so it has also gained the user scale and traffic that countless startups envy. Therefore, as long as there is a good product, the existing users will also easily convert into users of AI products. After entering the AI era, although ByteDance has launched a series of AI applications and products, the competitiveness of its products in some fields, such as image and video content generation, is relatively weak, and it is still at a relatively basic stage.
After Sora exploded this year, it brought a blockbuster "king bomb" to the Wensheng video field. Prior to this, ByteDance had launched the Wensheng video models Boximator and MagicVideo-V2, which were too far behind the Sora model and were not "comparable". Therefore, at the beginning of the year, Zhang Nan, CEO of Douyin Group, left his post and announced that he would shift the focus of his future work to Jianying CapCut.
Another signal is that since last year, Byte has successively shrunk its game, VR and local life businesses to varying degrees, and has given more job recruitment places to AI positions. So we see a large number of sales of business in ByteDance's gaming sector, and large-scale layoffs in PICO under the XR sector.
In November last year, ByteDance set up an AIFlow department specializing in it, and transferred internal senior management and business backbones to promote the AI process. Its core products mainly include AI dialogue assistant Doubao (Cici for the overseas version), Buckle for the AI bot development platform (Coze for the overseas version) that benchmarks GPTs, and the phone furnace for the two-dimensional group. At the Volcano Engine Conference on May 15 this year, ByteDance renamed the self-developed "Skylark" model to the "Doubao" model. According to media statistics, since August last year, ByteDance has launched a total of 8 large models, 2 agent development platforms, and nearly 20 AI applications native or based on existing products in the AI field.
Any technology company with significant global influence often has the phenomenon of a ship turning around. However, judging from the speed of ByteDance's transformation and the scale of existing products, it is indeed fast-tracking AI. This is inseparable from the wave of AI craze set off by ChatGPT, and global technology giants are competing in AI and large language models, and ByteDance is naturally not willing to miss the opportunities in the field of AI.
02. Can strong force still produce miracles?
As of the first half of this year, ByteDance's AI products cover multiple categories such as dialogue, tools, and interactive plots. It can be seen that ByteDance once again adopts a multi-directional trial-and-error strategy in the AI application track, actively training large models and looking for AI application scenarios. In addition, Zhu Wenjia, the former head of TikTok product technology, was appointed as the business head of the department, and personnel from other business lines such as Douyin and Feishu were also recruited. From these staffing, it can be seen that ByteDance also has expectations for AI internally.
It is worth mentioning that since the beginning of this year, Douyin has begun to tilt traffic and resource promotion for its AI application tool "Doubao", and we can often see bloggers on the Douyin platform begin to promote this product, and the data in recent months has also shown remarkable results. According to Quest Mobile's "2024 China Mobile Internet Spring Report", in May this year, Doubao's monthly active users reached 26 million, and as of April this year, Doubao has accumulated 137 million downloads in the Android market, leading AIGC applications by a large margin.
It is unclear how long ByteDance will be able to promote the resources of the Doubao tool in the future, but we refer to ByteDance's previous radical approach in the XR field. Once the launch effect is not good, the loss is serious, and ByteDance's resource tilt is canceled, PICO's headset will quickly lose its original sales and popularity. Of course, in addition to the abolition of resource tilt, a very important reason for the current situation of PICO is that the overall content ecology is not done well, and users will give up using it after using it several times after purchasing it because there is no high-quality game and content.
Returning to the large model, the Doubao large model was launched later than other domestic large models, so once released, it will face fierce competition with foreign OpenAI, Google Gemini, and domestic Wenxin Yiyan and Huawei Pangu large models. In the face of the "100-model war" in the field of AI, some industry insiders pointed out that according to the current computing power reserves and financial strength of technology companies themselves, the threshold for simply releasing a large model is not as high as the public imagines. However, only when a large model can continue to iterate with high-quality data scenarios, and a large model whose performance is gradually approaching ChatGPT is expected to be "rare" in the end.
In addition, compared with overseas technology giants that have entered the market earlier, there is still a considerable gap in the technological development of Chinese artificial intelligence, and due to well-known factors, domestic AI companies are limited by the tight computing power caused by chip control, which may directly affect the infrastructure of AI large models. In addition, the high cost of large model training and other factors also make the road for Chinese enterprises to catch up with a long way to go.
03. ByteDance's "ambition"
AI+ hardware is undoubtedly a hot track in the current AI field, and devices such as mobile phones and PCs that we are familiar with have begun to be equipped with AI large models. Previously, the Ray-Ban glasses launched by the industry giant Meta are smart glasses products that combine AI technology, and the Meta Ray-Ban glasses have a built-in large model developed by Meta and support multi-modal interaction. This means that users can interact with the glasses in a variety of ways such as voice, gestures, eye movements, etc., for a more natural and convenient operating experience. With the help of AI, it is also able to recognize objects and faces in the surrounding environment, and provide corresponding information and assistance based on the recognition results.
Since the beginning of this year, the large model of Doubao has begun to cooperate with a number of hardware manufacturers, and opened Doubao's AI capabilities to various hardware manufacturers. Previously, it was reported in the media that Byte's long-dormant PICO has been developing a number of wearable devices since the second half of last year, which will be equipped with AI. With the integration of AI + hardware, more and more AI applications will be integrated into XR hardware, and the number and variety of XR applications will also increase, which can promote the construction and expansion of the XR ecosystem to a certain extent.
Previously, it was rumored that ByteDance is working with American chipmaker Broadcom to develop advanced AI processors. Although it has been officially confirmed to be false news, to a certain extent, it can also see ByteDance's "ambitions" in the field of AI. However, on the hardware track, ByteDance obviously still has a lot of room for improvement. ByteDance's starting business is the Internet, and its core business is mainly focused on software and applications. However, hardware is a very competitive market, and even if you invest a lot of money in an aggressive way, you may not always get the expected returns. Secondly, if the market positioning of hardware products may not be clear enough, then it will be difficult to gain user recognition and acceptance in the market.
04. The status quo of the application of large models
According to the latest forecast of the "China AIGC Application Panorama Report", the size of China's AIGC application market will reach 20 billion yuan in 2024 and trillion yuan by 2030, becoming an important growth point in the global AI field. In this 10 billion market, some characteristics are beginning to appear. B-end products have become the main force, accounting for eighty percent of the revenue of the entire large model market, but half of the C-end products such as Wenxin Yiyan and Baichuan Intelligence, which we are familiar with, are still free or have free versions.
Thanks to the domestic market size and abundant data, we have reason to believe that AI application scenarios will be the key market for development in the future. This may be an opportunity for domestic AI companies to overtake in corners. And judging from the information summary of this year's large model company's press conference, most companies no longer compete with the parameter scale and computing power scale of large models. Then adjust to the large model to empower the industry, the combination of scenarios, and the combination of industries.
In terms of scenario combination, serving the B-end market has also become the preferred choice of large-scale model enterprises. For example, Microsoft is vigorously promoting the integration of its products with large models, and its product Office Family Bucket is connected to GPT-4, bringing a new office production model; HUAWEI CLOUD is also promoting the integration of the Pangu model with industry applications, and has released industry models such as the Pangu Mine Model, the Pangu Meteorological Model, and the Pangu Wave Model.
In addition, the landing of domestic large models in the B-end market is still facing the problems of cost and application scenarios. Although many large models have reduced prices at this stage, with the changes in the market and the increase in the operating costs of large models in the future, it is not excluded that there will be small and medium-sized enterprises in the future that they will give up because they cannot afford the cost. In terms of application scenarios, although the potential application range of large models is very wide, how to combine large models with the actual business of enterprises to create application scenarios with commercial value still needs to be explored.
05. Write at the end
At this stage, the public's demand for AI no longer needs an extra "bean bag" for entertainment and complaining. But partners across industries do need an industry model to lead productivity change. The popularity of large models has caused the industry to have a "hundred model war", but the market will eventually become rational.
For the success of any technology or product, the depth of the technology and the business value effect are crucial. Even if the development of a large model is rapid, if its business application is not deep enough and lacks actual business value effects, then this development may only be superficial and short-lived. Vigorously producing miracles may be able to gain customers quickly in the short term, but in the long run, it is not conducive to taking root downward, nor can it continue to gain the power of upward growth.