In recent days, Beijing took the lead in issuing policies to support the development of “ChatGPT-like industry”, which can be regarded as the official recognition of this technical direction. So the question is, who is the strongest among the domestic enterprises with technologies similar to ChatGPT?
In fact, the answer to this question is easy to find. Just ask me, which restaurant in Beijing is the best for boiled fish?
The answer I gave you must be that the best boiled fish is the one with the longest queue and the most parked cars. At present, the front-line players who can afford ChatGPT are mainly Internet giants such as BAT and ByteDance.
A pure AI enterprise may still be at the stage of powerlessness at present. This is mainly due to the characteristics of ChatGPT – big model, big computing power, big data, which determines that it must be a very small number of head enterprises to burn, and other AI enterprises simply do not have enough resources to follow up.
At present, domestic enterprises are striving for the service of the first ChatGPT in the slot, but the vast majority of enterprises have obviously insufficient talent reserves in this area.
Therefore, as long as we look at where these enterprises collectively go to dig people, we can roughly infer who has a higher level. The accurate information that the author has received is that at present, all enterprises qualified to play ChatGPT, including A and T, have targeted at one enterprise, which is Baidu.
Yes, Baidu.
Moreover, according to the author’s knowledge, this round of poaching basically has nothing to do with the headhunter company, or the headhunter company basically cannot play a role.
Because it is not a few algorithm engineers that they want to poach from Baidu. The starting point of digging people is at least targeted at high-level technical experts and department heads. Some enterprises even hope to poach Baidu’s team in big model. Therefore, the main force of this round of poaching is mostly senior investors.
After all, the identity of investors is suitable for wandering among enterprises, and investors also have the habit of reserving senior talents in their own pockets at any time.
Baidu, of course, is also ready for this. After all, it is conceivable that the conditions for AT and other enterprises to open up must be extremely rich. Therefore, Baidu also needs to use another method to retain talents, which is called “hope” or “dream”.
You can go, you can start again in another team… But if you want to make the first ChatGPT application in China?
Then don’t go. It’s in Baidu.
This is the most favorable condition for Baidu to retain talents. I am the closest to success. I have technical faith. In addition, one can prove that Baidu is far ahead in China in terms of GPT-like big model from the side, as well as the information released by Baidu.
At present, Baidu is the only company that has given the expected launch time of a similar product (ERNIE Bot), and you have detailed the information released by other companies. Although some companies have roadmaps, they are quite vague on the timeline, which has explained the problem – due to lack of confidence.
Point-to-point technology tree
Of course, digging people and official propaganda are both side evidence. The author dares to say that Baidu is the most advanced in this field in China, and there must be some more direct testimony. Here are some information that has not been disclosed in China at present. According to the information mastered by the author, in the middle of last year, Baidu conducted in-depth special benchmarking evaluation.
The core content of this evaluation is the comparison between Baidu Big Model and OpenAI’s GPT-3.0 (there was no ChatGPT at that time).
According to the materials mastered by the author, Baidu is ahead of OpenAI in some key indicators. Baidu is quite confident in the competitiveness of “Wenxin”. Although it is affirmed that GPT is still the forefront of the industry, it also makes an objective comparison of the comparative advantages of “Wenxin”. In short, this report and evaluation is quite inspiring to Baidu’s decision-making level.
It can be seen that Baidu has actually invested and accumulated in this field for a long time.
We can understand that Baidu “Wenxin” is the farthest path in the development of domestic AI big model, and has the most landing practice and proven technical path.
The last time Baidu said so, it was many years ago. When Baidu just started AI research for a short time, Robin Lee once gave a speech. He said that we used (deep learning) in image, voice and other searches, and found that the progress made once was more than the sum of years in the past.
As we all know, ChatGPT is the advanced version of GPT-3, or, according to some people, the 3.5 version or branch of GPT.
Then, the information revealed from Baidu’s internal evaluation and comparison is that Baidu is right on the technology tree this time. At least, it is the right choice to focus on research and development on the route of generative artificial intelligence. Therefore, Baidu should be the closest to the world-class level in this field among domestic enterprises, and almost none of them.
For Baidu, which is heavily betting on AI, this is good news at a strategic level.
But Baidu is not in a hurry to release these details. My guess is that Baidu’s external products of this kind still need polishing. Another version also comes from inside information. It says that even without ChatGPT, Baidu will probably upgrade its products this year, and it is likely to launch a model like Microsoft Bing, turning the entire search system into a retrieval+generation dual-mode system.
The author also needs to add a message – in fact, Baidu’s generative search has long been online, but there is no big official announcement. However, you can find that Baidu’s search results have changed from providing search results to providing answers to questions.
If you don’t believe it, you can search for questions such as “Tesla and BYD have higher profits” and “Beijing and Guangzhou have higher GDP” to see how Baidu answers.
On the other hand, the author believes that readers who care about this matter will also find that with the combination of Microsoft search bing and ChatGPT, more and more problems will be exposed after the surge of users of ChatGPT.
This shows that ChatGPT has shown its extraordinary talent, but there are still many shortcomings in terms of fidelity, credibility and accuracy. One of the most important reasons is the lack of vertical domain knowledge, and the data volume of the whole training is not large enough, which makes it very interesting, but it is difficult to spread widely in the real application layer in the short term.
Under the public beta of so many users, ChatGPT has continuously exposed low-level errors, which has instead returned to the essence of technological development – because as a super technology created by human beings, it also has to follow a process of continuous self-improvement. Under the circumstance that Benzun has continuously exposed problems, Baidu must also be highly cautious about launching similar applications. After all, Google has been the first to face up, and the lesson of 200 billion yuan is too profound.
Who rules the ups and downs?
If you ask me who is the most likely winner of ChatGPT in the Chinese Internet world, I will tell you frankly that I still think it is Baidu. Some people may ask, isn’t Google struggling to catch up with openAI?
If Google can counter – surpass Baidu, why can’t domestic giants counter – surpass Baidu in the same track.
In fact, the answer is already there. The day when Google and OpenAI have the smallest gap in ChatGPT is the day when ChatGPT is released.
According to a research report of UBS Group, ChatGPT achieved 100 million monthly active users only two months after its launch, becoming the fastest growing consumer application in history. Before that, this record was maintained by TikTok. It took about 9 months to reach 100 million live users.
After two months and 100 million users, Google has no chance to catch up with OpenAI.
Perhaps you have noticed that after Google became the global search leader, no one has won over Google in search, including Microsoft and Apple—— After WeChat has become the leader of instant messaging in China, many enterprises (including Byte and Alibaba) are catching up, but they have not surpassed WeChat—— Baidu has occupied the top position of Chinese search for nearly 20 years, and has encountered numerous challenges, but why no one has been able to challenge Baidu’s position so far; You can cite many theories, such as user habits, migration costs, latecomer disadvantages, etc., but I think they are not as accurate as two concepts, that is, Self-reinforcement effect and flywheel effect.
Self-reinforcement effect has its own corresponding characteristics in psychology, corporate strategy and deep learning technology, which is a very magical word.
The flywheel effect may be more familiar to you.
Anyone who has ridden a bicycle knows that in order to make the stationary wheel turn, it needs to make great efforts at the beginning. When the flywheel reaches the critical point, it will turn faster and faster, and it will be lighter and lighter to step on.
Therefore, when these two effects are linked, there is a trend: according to the self-reinforcing effect, Baidu will always set a standard for itself before doing this business, and use it to judge and evaluate the gap between its own behavior results and the established standard. If the result exceeds the standard, it will produce positive self-evaluation. If the continuous positive feedback is received, from strategic confidence to implementation determination, Will become stronger and stronger.
In response to this growing confidence and action, if Baidu takes the lead in producing products above the qualification line, it will gain a large number of users at the first time. In addition, Baidu itself is the absolute leader of Chinese search, so these users will grow rapidly to reach a high level, and then generate a large number of user data and feedback.
Baidu can further use these feedback and data to continuously optimize its own models and services, and finally make the flywheel rotate at high speed, thus becoming the absolute leader of the new generation of search again.
I remember that more than ten years ago, I asked Robin Lee a question. I said, is it Microsoft that is most likely to catch up with Google (Baidu)?
Robin Lee’s answer is that search engines are not the best for those who have the most money. Otherwise, Microsoft should be the best search engine in the world?
When I asked this question, Apple had not released the iPhone, and Microsoft was the richest IT company. Facts have also proved that if Microsoft did not buy some shares of OpenAI with “money capability”, it would always be a second-rate product in the search market.
Robin Lee later explained clearly from his own point of view why the phenomenon of “once you are ahead, you will always be ahead”. He said: “Good applications must be related to the size of users, because more users, more problems will be exposed.
If the leader solves these problems before other enterprises, he will always be at the forefront of innovation. ”
This is also an enlarged version of the self-strengthening effect+flywheel effect. I remember that Robin Lee still said that “the development of the Internet in China and the United States is the same. China has many Internet users and many problems. If Chinese enterprises can solve some problems before American enterprises, we may surpass them in Internet innovation.
“This statement was at least partially confirmed later. For example, the Tik-tok we mentioned earlier is a very obvious example. Due to the existence of self-reinforcing effect+flywheel effect, it is difficult for latecomer enterprises to solve the problem of catching up with” brute force “.
I can give another example. At that time, 360 coveted Baidu’s share in the search engine market. Zhou Hongyi came up with a plan to seize 20% of Baidu’s market share in a very short time.
This was something that Google could not do in China at that time.
But Zhou Hongyi almost did it. 360 Search once soared to nearly 20%, making Baidu nervous. At this time, Baidu held an internal meeting, and they had to calculate an account. Why did 360 catch up so fast?
After a calculation, they found that the reason why 360 is catching up with and surpassing others is that 360 has many external traffic resources, such as navigation stations, and all kinds of evil pop-up windows… Zhou Hongyi replaced all the search sites above with 360 search. So for a while, 360’s search share rose sharply, approaching Baidu’s warning line.
But at this time, Baidu is not nervous, because according to their calculation results, there are two judgments: First, Zhou Hongyi will spend all his resources and probably increase to 22% of the market share at most, which will not be higher; Second, because these search volumes are generated by using 360 search to forcibly replace Baidu search, their commercial value is very low, user experience is poor, conversion efficiency is also very low, but the cost is quite high.
Therefore, 360 will not persist for more than 3 months. As expected, three months later, the 360 offensive disappeared.
The reason for telling this story here is to show how amazing the first-mover advantage can be in the killer application of the Internet with a large scale and large user population. Basically, once you take the lead, you will take the lead.
If you want to surpass, you need to make a revolution of intergenerational replacement. This time ChatGPT perfectly deduces this theory. At present, although Baidu has encountered various difficulties, its position in the Chinese search field is still unbreakable, and the first application scenario of ChatGPT is search.
That is to say, there must be a gap between Baidu’s ChatGPT (that is, ERNIE Bot) and the original version. Maybe 8 points, maybe 7 points or even 6 points. But it doesn’t matter. As long as it is above the passing line and quickly combined with Baidu’s search business to launch a dual-mode form, Baidu’s first business curve will be reborn.
But in fact, in addition to search, if ERNIE Bot can be combined with Baidu’s intelligent cloud services, it will also subvert the way of cloud services, and will find more landing businesses in content and information related industries and scenarios, which has been publicly confirmed.
This is not only the second growth curve Baidu is building, but also a market with a higher ceiling. You may not like Baidu, but in China, only Baidu has achieved the determination and action of investing 100 billion yuan in AI in the past 10 years. Besides, it is true that people have applied strategic foresight to the right technology tree.