news

한어Русский языкEnglishFrançaisIndonesianSanskrit日本語DeutschPortuguêsΕλληνικάespañolItalianoSuomalainenLatina

About the incident where I was forcibly detained for ten minutes at ChinaJoy.

Nowadays, online and offline advertising for large-scale products have become very common, with various fancy ways of going viral emerging one after another, and the scale of marketing can be described as "volume".

Offline, billboards in areas with dense white-collar populations, such as subways, office buildings, and airports, have been replaced with advertisements of large model companies.

The advertisement shot by Hedgehog Commune at the subway station downstairs

And online, whether you are in school or working, in the past six months, if you open Bilibili or Xiaohongshu, from generating papers with one click, to how to report to leaders, to real-time understanding of hot Internet memes, the big model company is trying to tell you that AI can cover all copywriting from school to work.

According to AppGrowing statistics, in the first week of June, Bilibili’s AI advertising volume was 3 to 4 times that of the same period last year, and the advertising revenue of AI evaluation up-hosts increased instead of decreased.

Chen Rui, CEO of Bilibili, mentioned in his speech that technology is one of the most popular contents on Bilibili, and AI is the fastest growing technology content. In the past year, 200 million users have watched technology videos on Bilibili, and the number of views of popular science content has increased by nearly 200%. UP hosts have produced more than 1 million AI-related articles throughout the year, and 60% of the consumers of AI-related content are born after 2000.

What are the marketing ideas and strategies of domestic large-scale model companies behind the large-scale launch, and how do users feel about them? In addition to marketing, what is the way out for domestic large-scale model companies?

How do domestic large model companies place orders?

Bilibili is Kimi's main battlefield. On Bilibili, Kimi has associated almost all the terms that users may search for "AI" on Bilibili, such as ChatGPT series, AI series, efficiency series, assistant series, worker series, etc.

Kimi's success has obviously set off a trend, and other large model manufacturers have joined the marketing war. Tiangong AI (China's first search engine to incorporate a large language model) has the same idea of ​​investing in traffic. Although it has the disadvantage of falling into homogeneity, it has objectively diluted the label of competitors.

The effect of advertising is quite intuitive. Take Kimi for example. According to the statistics of the monthly report on AI applications by the Quantum Bit Think Tank, the number of new downloads in June exceeded 3 million, a 50% increase compared to May. However, this growth rate ranks only second among all large model companies.The first place is Doubao, owned by ByteDance, with nearly 17 million new users in June. Compared with May, the growth rate of new downloads was 100%.

This is thanks to ByteDance’s generous investment.

Slightly different from Kimi's traffic investment method, Doubao's main marketing battlefield is Douyin, which is also owned by ByteDance. Its marketing strategy is similar to that of traditional Internet products on Douyin - first, information flow advertising, recommending users to download and install in short videos; second, co-promotion with Jueliang Star Map, relying on KOL cooperation to increase popularity and influence; finally, explaining usage in the live broadcast room to guide users to download.

Doubao's advertisement on TikTok

According to AppGrowing statistics, in early June, Doubao's advertising amount reached 124 million yuan, mainly through the Pangolin Alliance under ByteDance. Pangolin's advertising effect is very good, but according to insiders in the big model industry, Pangolin has barriers and does not allow other big model companies to advertise.

Therefore, other manufacturers can only find other ways.For example, Kimi, in Douyin marketing, focuses more on emphasizing tags"Domestic AI", "Developed by Tsinghua University Team", "Tsinghua AI", "Tsinghua AI Artificial Intelligence" and other words are all high-frequency words used in Kimi's soft advertisements.

When everyone started using tag rolls to attract customers, Kimi turned his attention to Xiaohongshu. Searching for "Kimi Jingran" on Xiaohongshu, you can get the following content: Jingran can automatically write emotional hit titles, Jingran is an emotional companion AI, Jingran can automatically detect and replace all kinds of banned words... This marketing method actually further subdivides the use and user portraits, and before the technology has not yet achieved major innovations, it digs deep into product functions to make up for the shortcomings.

There are signs that even though each company has different strategies, domestic big model companies all pay more attention to the C-end and early marketing. The marketing is mainly aimed at the productivity improvement scenarios of knowledge workers, because these scenarios require a higher level of intelligence for big models, which is more in line with the mission of big model companies to explore general artificial intelligence (AGI).

In May this year, Deepseek was the first to cut prices, and ByteDance followed closely, launching a price war in the big model industry.Free and price reduction are more for marketing purposes and are the spillover of product homogeneity.

The price reduction methods of domestic large-scale model companies are different from those of mobile Internet companies. Whether it is online car-hailing, e-commerce or food delivery, the so-called price war is often achieved through huge subsidies. The price reduction logic of large-scale models is driven by technical means, reconstructing production functions, improving R&D efficiency, and thus reducing costs.

However, price wars in the mobile Internet industry often start in the middle and late stages of industry competition, when the middle and lower-tier companies are eliminated and only the top few companies are left. At that time, the technology is relatively mature, the market is clear, and the competition landscape is clear. However, the competition among large model companies is still in a white-hot stage, and the progress of product development and commercialization is not very clear. It is risky to rush into price wars.

Faced with marketing bombardment, users only care about whether it is easy to use

Insiders in the big model industry pointed out that these companies' marketing wars were forced upon them and were no different from the business logic of the mobile Internet era: "Mobile Internet commercialization has already explored a more efficient path, and I think this is China's advantage."

The reason for the volume marketing war is partly because there has not been a greater breakthrough in technology.“At this stage, unless you are the first to create a more powerful technical product, if you cannot let more people know about you, you will actually fall behind.”

In addition, with more users, we can understand their segmented needs, which helps to make the model better. Yang Zhilin, founder of Dark Side of the Moon, once said in an interview: "Today, users have helped us discover many scenarios that we have never considered. He used this to screen resumes, which we did not think of when designing the product, but it works naturally. User input in turn makes the model better."

The aggressive marketing has indeed attracted many users. Xiaoyi saw Kimi's promotion and found that it focuses on workplace communication, so he tried to use it to deal with the "nonsense literature" in the Chinese workplace.

"Some materials have never been written before, and I don't know how to express them. It is very convenient to use AI to produce the first draft and then refine it. I once encountered a very outrageous scenario. The main project leader of Party B of our project resigned. I had to send a letter to Party B's CEO in the name of the company's president, requesting reassignment of personnel and strengthening of the personnel stationed in our company. This kind of communication letter is a million miles away from my job, so AI is very suitable." Spoken language can be input into the language and converted into clear and easy-to-understand professional written language. It is also very clear to read long articles and make literature summaries. Domestic AI performs very well in text summarization and text generation.

But when it comes to processing complex data, the ability of domestic AI is not so good. Netizen @AIGCLINK pointed out that "the answers of domestic AI to programming questions are basically a bunch of correct nonsense, and no solution is given at all," and there is still a big gap compared with foreign large models.

Zhipu AI’s advertisement at the airport

Many users also said that they were initially attracted by the marketing of domestic AI. "The marketing is really great. I was attracted to try it several times, but I was disappointed every time." The main reason is the poor accuracy of the generated content. Take the large language model as an example. The problem of "nonsense" has not been overcome. This uncontrollable state makes it difficult for it to be truly implemented in production and play a big role. It still needs time to polish.

Users actually don’t care whether the model or algorithm is self-developed. The most direct perception of users is whether it is easy to use.

This is actually a major problem faced by domestic large-scale model companies, that is, the promotion customer acquisition rate is high, but the user retention is poor. The premise for customer acquisition to be useful is that the product is valuable and can retain customers. However, the value provided by large models to ordinary people on the C-end is scattered, random and unstable, making it difficult to make a profit, and the business model is still being explored.

At the same time, whether from the investment platform or the potential users, the users are the same group of users, the cake is so big, and there is no discontinuity in the model capabilities of each company. How far can the crazy marketing go?

Why don't large foreign model companies expand their marketing?

OpenAI also implemented a price reduction strategy this year, but looking at the foreign large-model ecosystem, there is no trend of marketing wars and price wars, and major manufacturers are still competing for capabilities.

Foreign large model companies have always been competing in R&D "hard innovation". For example, OpenAI recently released GPT-4o; Google launched an update to Gemini 1.5 Pro, showing off the super AI voice assistant Astra, and Veo, which is benchmarked against the current leading video generation model; Microsoft launched AI PC and updated Copilot; Apple launched the first M4 chip that focuses on AI functions...

This is also a trend that has existed since the mobile Internet era. "Foreign large model companies have strong product power. As long as they hold some press conferences and official information, people will naturally use and spread them." A senior Internet practitioner lamented.

The AI ​​market abroad is relatively more mature, and users have strong payment habits, so the generative AI application market is progressing faster. However, Chinese users have never developed the habit of paying, the application development ecosystem is also immature, and costs limit developers' enthusiasm, resulting in relatively slow development of the entire market.

Since the beginning of July, we can see that domestic manufacturers have begun to invest in the application and promotion of browser AI plug-ins. Prior to this, foreign AI manufacturers have begun to try to launch various AI plug-ins. This move is also to compete for user entrances and accumulate more user data to train more powerful models.

There are many types of plug-ins, ranging from browser derivative plug-ins based on the big model functions of the product itself, to "shell" products based on existing models (that is, not developing products but only doing product integration), to functional plug-ins such as translation and speech-to-text that only do vertical functions... There are currently a dazzling array of plug-ins. Although it looks more like the anxious behavior of big model companies who are afraid of not keeping up, this is also a reflection of AI manufacturers looking for more segmented landing scenarios.

A domestic large-scale model industry practitioner said, "Foreign manufacturers have actually been leading the domestic market because they are indeed more mature. The model capabilities of China's C-end large-scale model companies are actually similar. For example, in AI search and AI emotional companionship, foreign companies are still leading the way. HoweverAs for the application direction, everyone is still in the exploratory stage. The current AI is still in a very rudimentary form, and the whole world is trying to find nails with a hammer.”

He pointed out that the Chinese market cannot completely imitate and learn from foreign countries, because China's situation is very special and it also has its own market advantages. "Take B-side products as an example. Many AI giants have their own huge application needs. In addition, highways, water conservancy, and finance all have strong application scenarios."

Compared with foreign manufacturers who focus more on products and underlying R&D, which are capabilities from 0 to 1, Chinese companies are actually better at going from 1 to 10. China's mobile Internet industry is mature, and with the rich application scenarios brought about by the huge market scale, the implementation of large models must start with breakthroughs in scenarios. In fact, instead of volume marketing and price reduction, domestic large model companies should try to find value from scenarios. Such commercial exploration is healthier and more sustainable.

We must first step out of the marketing war and implement it into applications before we can explore more possibilities.

For media cooperation, please contact WeChat ID: ciweimeijiejun

If you want to communicate with us, you can reply "join the group" in the background to join the community