Seemingly low-key, but actually robust: What will Byte encounter on the road to AI?

Standing at the forefront of large-scale models, Internet giants are continuing to promote the accelerated implementation of AI products.

Major manufacturers such as Baidu, Huawei and Alibaba have launched their own AI products, and are not only working hard to continuously increase their popularity, but also actively building their own AI product systems. In comparison, ByteDance (hereinafter referred to as "Byte") seems to be much more low-key.

Compared with the AI ​​products of other Internet giants on the industry side and To B side, Byte seemed to prefer to develop its own AI products on the consumer side and To C side. For example, Byte has launched painting and special effects AIGC functions on its Douyin, Toutiao and other products to explore the perceived experience of C-end users.

Just recently, another media broke the news that Byte is internally testing an AI dialogue product, currently called "Grace", and the product is in the testing and improvement stage. And this exactly reflects that Byte seems to be low-key in the field of AI, but in fact it is holding back its ambitions to expand.

Low-key and forbearing before launching a critical attack

Byte’s approach to the field of AI is very different from other giants.

Other Internet companies often want to take advantage of the momentum of the big model and establish their position in the AI ​​field and vertical subdivision industries in one fell swoop. They particularly value the opportunity to open up a second growth curve and are sparing no effort in the field of AI.

In comparison, ByteDance pays more attention to the C-side experience. According to Byte's view, the ultimate goal of large models and AI products is to provide services for its traffic-based apps, such as Douyin and Toutiao.

And Byte, which has long become a pole of China’s Internet, is also regarded as an “APP factory.” The mysterious algorithm mechanism and unlimited traffic with Douyin as the core have been widely spread. Driven by traffic, Byte has shown an all-powerful trend and has entered many fields such as live broadcast sales, local life, games, advertising and marketing.

For Bytes, traffic is everything. The ultimate destiny of AI products may be to serve traffic. Take the popular AI painting on Douyin as an example, it has become an excellent means of generating traffic. Just provide a picture, and AI can generate an anime-style picture based on the picture content.

Since the generated picture effects have certain "blind box" attributes, and the painting style is exquisite and beautiful, but also a little silly and nonsensical, it has inspired the enthusiasm of the majority of users to participate as soon as it went online. The number of submissions in a single day on Douyin The maximum number is 724w, and discussions and sharings such as "How to tame AI" and "Who will speak for me" have also been derived.

According to the homepage of AI Painting Special Effects, more than 27.583 million users used this special effect in early January this year alone. AI painting peaks at Douyin’s SSS-level hot-selling special effects, with a request volume of up to 1.4w per second.

According to the Douyin special effects and Bytedance intelligent creation teams, AI painting uses multi-modal generation technology. This is a cross-modal generation technology that generates pictures/videos/3D from text. Through large-scale data training, it is guided by only text or a small amount of other low-cost information to controllably generate pictures/videos/3D, etc. of any scene. The content has great potential application value in AIGC and other directions.

Byte's products, including Douyin and Toutiao, have continuously added AIGC functions in addition to AI painting, triggering huge traffic. It seems that Byte's approach to the AI ​​field is very purposeful. Of course, Byte's style of play seems to have many highlight moments, but they are mainly reflected in the To C side. Judging from the fierce competition in the entire AI field, it can be considered low-key.

But playing low-key does not mean that you will always work quietly. On the contrary, after Byte's AI products have been fully verified on the To C side, it has conducted in-depth cultivation of its own large-scale model system and already has a fairly mature and unique innovation path. "Critical attack" naturally follows.

It is understood that Byte has established an AI ecosystem, including visual language model, Volcano high-speed training engine, Volcano machine learning platform and industry applications. At the model level, Byte's accumulation in AI is mainly concentrated in the four fields of machine translation, intelligent speech, video images and multi-modality. Byte's research results include the non-autoregressive model DA-Transformer, the end-to-end speech-to-text translation model ConST, the multi-granularity visual language model X-VLM, the unified image and text generation model DaVinci, etc.

At the computing power level, Volcano Engine officially released a series of cloud products such as self-developed DPU at the Volcano Engine FORCE Motive Power Conference held in April this year, and officially launched an intelligent recommendation-high-speed training engine, supporting ultra-large models from 100GB to 10TB or more. efficient training.

At the platform layer, Volcano Engine has released an upgraded version of its machine learning platform, which has core advantages in the three areas of data management, resource management, process and model.

At the application layer, the AI ​​scene products of Volcano Engine mainly include intelligent creation, intelligent experience and intelligent marketing. Currently, it has served hundreds of customers and promoted customer business growth in video creation, interactive creation and live short video construction respectively.

In addition, Byte has continuously launched related products based on AI technology, such as Sponge Band and Feishu My AI. It is worth mentioning that the AI ​​dialogue product "Grace" tested internally by Byte this time is also one of the concentrated expressions of "Crit". According to the response from the relevant person in charge of ByteDance, "Grace" is the internal code name of an AI dialogue test project. It is still in its early stages and is only used for internal experience testing.

Although the response remained low-key, it also officially revealed the mystery of the progress of the Byte AI project. According to the analysis of industry experts and based on the general product testing cycle, the project is expected to begin external testing in 2 months.

In addition, Volcano Engine has launched intelligent drawing products for enterprise customers, which saves enterprises the time and cost of collecting data, training models and optimizing performance, and can quickly connect to the business. It can be seen that Byte seems low-key, but in fact it is stable and is galloping on the road of AI.

"Shidunzi" poses a huge threat

There is no doubt that Byte will definitely fully apply AI products based on large models to its apps in the future, and achieve great success and gain more traffic.

But if you think that Byte will only indulge in "enclosure and cuteness", it is obviously too small to underestimate its ambition. It is foreseeable that Byte will continue to expand its development of AI products and cooperate with more enterprises to achieve continuous progress in both vertical and horizontal directions. There is no doubt that Byte is working hard to become a leader in the AI ​​field and is fully prepared for future competition.

However, this also implies that Byte will have a direct conflict with many Internet giants and large enterprises in the field of AI. The latter have their own advantages, characteristics and powerful capabilities, and they are fully capable of becoming Byte's "stone block" on the road to AI.

Looking at the overall environment, many companies have made efforts to train large models and actively launch AI products. You must know that large models enhance the versatility of AI technology, allowing developers to develop better AI models for scenarios at lower costs and lower thresholds, helping to realize inclusive AI products.

According to statistics from China Mobile Think Tank, as of April 18, 2023, at least 19 domestic companies and scientific research institutes have participated in artificial intelligence large model training. These companies can be mainly divided into three categories: Internet giants and large enterprises, scientific research institutions and start-up technology teams.

Among them, Internet giants have almost adopted a consistent strategy in the field of AI - a four-in-one layout targeting "models, computing power, platforms and applications". For example, Alibaba is the "Tongyi Large Model Feitian Intelligent Computing Platform Magic Tower Community Industry Application"; Baidu is the "Wenxin Large Model Kunlun Core Flying Paddle Platform Industry Application"; Tencent is Tencent's Hunyuan Large Model HCC Computing Cluster Tai Chi Machine Learning Platform Industry Application”…

From the To B side, many Internet giants or technology companies are already in a leading position in large-scale application of AI products or directions. At present, Baidu Wenxin Large Model, Huawei Pangu Large Model, and Chinese Academy of Sciences Zidong Taichu Large Model have all accumulated benchmark application cases in the To B end vertical market.

For Byte, it is not easy to seize food from the tiger's mouth.

Looking from another perspective, domestic large models are currently in a state of blooming, but the industry generally believes that the first echelon is Baidu, Alibaba, Tencent, and Huawei. In terms of comprehensive strength, the four companies have strong investment in large model R&D, technical capabilities and talent teams; in terms of commercial promotion, the four companies rely on existing business fields to more easily form large model application scale effects.

If Byte wants to break the inherent perceptions of the industry and customers, it must come up with competitive advantages to achieve overtaking in corners.

In addition to Internet giants and technology companies, some vertical segment companies have also accelerated their pace to compete for the benefits of the AI ​​wave. For example, there is already fierce competition in the education industry.

The official video account of NetEase Youdao released a demonstration video on May 5, showing the AI ​​spoken teacher developed based on the "Ziyue" large model. On the same day, Xueersi officially announced that it is developing a large mathematical model, named "MathGPT", for global mathematics enthusiasts and scientific research institutions, and will launch product-level applications in 2023.

iFlytek released the iFlytek Spark cognitive model one day later, claiming to have seven core capabilities including text generation, language understanding, and knowledge question and answer. iFlytek took the lead in launching the T20 series of AI learning machines based on the Spark model, and successfully applied it to actual scenarios. On June 5, the Internet education platform’s Homework Helper was in internal testing of a large-scale education model based on the Chinese market, including multi-disciplinary problem solving, Chinese and English composition correction and other functions, involving tool apps, smart hardware, books and other businesses.

What you need to know is that Byte is also investing heavily in the education industry. Just before that, Byte acquired a Suzhou K12 education service provider. In addition, Qingbei Online School, an education product owned by Byte, has also issued a recruitment notice for online teachers in primary and secondary schools, claiming to provide outstanding teachers with an "annual salary of two million, with no upper limit" salary package. But in recent times, Byte’s education business has stagnated. While facing more intense competition, it is of great significance for Byte to promote the recovery of its education business through AI products.

Internet giants, large technology companies, vertical segmented companies, and even start-ups... Byte has encountered many "stones" on its AI journey. But it is this kind of positive and healthy competition that allows us not only to see AI products that are expected to be implemented and full of possibilities, but also to see the vigorous growth trend of the domestic AI field based on the overall forward-looking architecture. .

The above is the detailed content of Seemingly low-key, but actually robust: What will Byte encounter on the road to AI?. For more information, please follow other related articles on the PHP Chinese website!