Text | Zhao Jiaqi
Edit | Zhang Kun
Design|Li Zhengwei
After more than a month of momentum and anticipation, Wen Xin's words finally came. It's just that it seems to be "on the street" as soon as it starts.
On March 16, Baidu held a press conference at its Beijing headquarters to officially push Wen Xin's words to the front of the stage. Successfully ahead of Google, Meta and other technology giants, becoming the second Internet company in the world to release a large language model after Microsoft Bing.
At the press conference, Robin Li, founder, chairman and CEO of Baidu, demonstrated the comprehensive application ability of Wen Xin Yi from five usage scenarios: literary creation, commercial copywriting, mathematical calculation, Chinese understanding, and multimodal generation.
Disappointingly, as a large language model of real-time interaction, Baidu did not choose to live real-time demonstration, but more conservatively used prepared questions and answers to display. Such a performance is obviously not convincing, so many people ridicule, "Others are ChatGPT, and Wen Xin is ChatPPT".
But no matter how ridiculed by the public, Wen Xin's words seem to have become the most beautiful boy in the Chinese Internet field at present. The ChatGPT boom behind Wen Xin's words has actually been transmitted to the entire automotive industry.
1
Car companies and Baidu's "two-way rush"
In fact, since the date of the official announcement in early February, car companies have continued to buy "early bird tickets" in an attempt to enter Wen Xin's circle of friends first.
Up to now, a number of car companies such as Changan, Jidu, Geely, Lantu, Hongqi, Great Wall, Dongfeng Nissan, AIWAYS, Zerosport, and Haima have confirmed that they will prioritize the internal testing experience and access their relevant capabilities, and jointly create an AI interactive experience for automotive intelligent scenarios with Baidu. According to relevant sources, if the project can be successfully implemented, the model equipped with the results of the cooperation between the two parties may meet with Volkswagen within two years.
It can be seen that the key direction of cooperation between the two is still concentrated in the field of artificial intelligence communication and dialogue, and the object is the intelligent cockpit, which is also the core function of Wen Xin at this stage.
This is not surprising. In the face of an increasingly involuted automotive market, car companies urgently need a breakthrough and disruptive "selling point" to empower the competitiveness of their products. At the same time, the landing of Wen Xin's words also needs to rely on car companies to help them quickly roll out the application network and provide feedback to improve it. In other words, the large-scale cooperation between Wen Xin and car companies is not a unilateral "backward" by one party, but a "two-way rush" based on commercial considerations.
However, judging from the press conference, Wen Xin's performance seems to have failed the expectations of car companies, and the ability shown does not have a subversive effect in automotive applications.
Why do you say that? As far as the functions currently displayed are concerned, Wen Xin's words are not outstanding.
For example, literary creation and commercial copywriting, as we all know, this function is no longer a new word in the Internet field. As early as 2015, Xinhua News Agency launched the writing robot "Kuaibi Xiaoxin", which can achieve rapid production in 3 to 5 seconds, which has triggered a discussion about whether the profession of journalist editing will be replaced by AI. Nowadays, relying on the development of technology, AI writing poetry, writing novels and other high-demand and highly innovative text creation activities are no longer a problem.
Another example is Baidu's most eye-catching multimodal generation in this exhibition, which can generate pictures, audio and video based on a piece of text. If it had been earlier, it might have caused an exclamation, but unfortunately, just the day before Wen Xin's words were released, there was also a multimodal appearance in ChatGPT-4, and its display was more intuitive in comparison.
In addition, compared to the coherent language comprehension ability shown by ChatGPT-4, Wen Xin's words seem to be lacking. In the presentation of the press conference, we can also see that the test of Wen Xin's words adopts a question-and-answer factual inquiry, and the answer method is more patterned. In a set of comparative tests, the difference was even more pronounced: after giving Wen Xin a hidden task of saying "ask for a phone number", it forgot it in the second round, while ChatGPT remembered the hidden task of the previous round.
Moreover, in terms of automotive applications, as far as Baidu's current technology is concerned, Wen Xin's words have not played much role in the core competitiveness of automobiles, but only stay at the interactive level of experience. Among the five usage scenarios shown by Wen Xin, there is a high probability that only Chinese understand that this can be used appropriately to make the originally slightly clumsy AI model "smart" and improve the intelligent driving experience of drivers and passengers. In addition to this, do you do mathematical calculations in the car?
An obvious conclusion is that whether it is a full-scenario application or a single automotive field, Wen Xin is not prepared at this stage.
2
Baidu's "hard injury"
In fact, it is no wonder that Baidu is betting in such a hurry.
In recent years, Baidu can be said to be very anxious. From 2011 to 2021, Alibaba's revenue in the same BAT camp increased from 20 billion yuan to 853 billion yuan, Tencent also increased from 28.5 billion yuan to 560.2 billion yuan, while Baidu only increased from 14.5 billion yuan to 124.4 billion yuan in the same period. Up to now, the market capitalization of Baidu, Ali and Tencent is 46.264 billion US dollars, 245.011 billion US dollars, and 3.44 trillion Hong Kong dollars, respectively, Baidu is at the bottom.
Before this, people mentioned BAT and thought of Baidu, Ali and Tencent, and now many new generations will naturally understand the "B" in BAT as ByteDance, coupled with the rise of Internet companies such as Meituan, Pinduoduo, JD.com, Netease, etc., Baidu's situation is difficult to say well.
For a long time in the past, Baidu has frequently bet on major outlets, chips, smart cars, artificial intelligence, meta-universe, everything Baidu is willing to do. However, the money is spent, and the results are small.
Take the example of intelligent driving, which is most closely related to the car. Since 2013, Baidu has begun to comprehensively deploy vehicle-road collaborative autonomous driving, and has participated in the formulation of a number of relevant national standards; On March 1, 2017, Baidu announced through an internal email that it would integrate its existing business and resources and establish the Intelligent Driving Business Group (IDG), and a month later, Baidu launched Apollo, the world's first autonomous driving open platform.
Today, although Baidu Apollo has grown into the world's largest open platform for autonomous driving, it has created a high-precision map of intelligent driving with an automation rate of 96%. However, due to the long-term difficulty of landing and commercialization of high-end autonomous driving, this profit route has never been able to run through. Today, IDG's Intelligent Driving Division is still in the red.
In addition, with the explosion of the meta-universe, Baidu also immediately followed up and launched the meta-universe platform Xiyang. However, due to the lack of organizational structure, lack of games and social genes, and hasty launches, except for a few car companies that held a few press conferences in Xiyang, this platform seems to have no news.
At the root of the problem, some industry insiders believe that Baidu lacks system capabilities, which is the "hard injury" that it is difficult to support such a large-scale new exploration.
Why are car companies like Volkswagen strong in their ability to resist risks? The core lies in "system power". Baidu, which lacks strong system capabilities, has a very strong boundary effect, so the coordination within the organization and the control of specific affairs are relatively weak.
This time, Baidu once again stood on the cusp of the wind and pressed the treasure on Wen Xin Yiyan. What is different from the past is that due to the previous accumulation of search engines, modular capabilities and data engineering technologies, as well as the reserve of talent resources, Baidu has an incomparable comparative advantage.
Objectively speaking, Baidu is very suitable in the exploration of ChatGPT-like technology.
However, if this "hard injury" cannot be improved, it is difficult to say whether Baidu can lead Wen Xin to go head-to-head with ChatGPT, and truly penetrate into the hinterland of the automotive field and even full-scene applications.
3
Let the bullets fly a little longer
It can be seen that at this stage, Wen Xin's words are being criticized from many aspects, but the other side of criticism is also people's ardent expectations for domestic ChatGPT technology. This can be seen from the name of Wen Xin's word - "Chinese version of ChatGPT".
There is no doubt that Baidu is ahead, but Baidu is not the only one on this road. In terms of automotive applications, in addition to voice interaction, the industry is also interested in incorporating this model into autonomous driving.
At the China Electric Vehicle 100 Forum just held, Academician Ouyang Minggao said, "ChatGPT will trigger a new round of revolution in artificial intelligence, and it will also have a profound impact on intelligent driving." Although most of the current cooperation between car companies and ChatGPT is still in the stage of human-vehicle communication, it is foreseeable that with the continuous evolution of intelligent driving in the future, the model and technical logic of ChatGPT will gradually be applied in the context of autonomous driving. ”
In this regard, many companies have also begun to try. For example, it is deeply engaged in the field of autonomous driving.
In January this year, Gu Weihao, CEO of ChatGPT, highlighted ChatGPT on AI DAY, and said: "ChatGPT has launched research on the technology behind ChatGPT. ”
In Gu Weihao's view, the idea of ChatGPT-like technology and the idea of autonomous driving cognitive decision-making are actually consistent, and the automatic driving system learns the correct driving behavior through the model, and then stably outputs the optimal solution in the face of different driving scenarios. In this way, the pass rate is expected to increase by more than 30% in difficult scenarios such as U-turns and roundabouts.
However, it must be admitted that this is only just beginning. In order to truly realize the real landing of this technology in the automotive field, the technology companies represented by Baidu still have a long way to go, and car companies also have a long way to go.
To use a popular phrase these days, I know you're in a hurry, but don't rush yet. Letting the bullets fly a little longer is the best way to deal with sudden and dramatic changes.