GuShiio.com Wen| Wang Zhiyuan
Two days ago, when WeChat DeepSeek R1 was first launched, a friend asked me:
When you got the qualification for internal testing, why didn’t you write about your experience as soon as possible? I said: It’s useless yet, I don’t understand it, but I don’t dare to write it indiscriminately. I have been trying out this new model for the past few days, and I only have one feeling after using it:It doesn’t work well.
01
First of all, this function is hidden too deeply. To use it, you have to first click on the search box at the top of the chat interface and then select “AI Search”.
But I am usually used to searching directly on the “Discovery Page”. Now every time I have to slide down from the chat interface to find the search box, and then click on AI search, which completely disrupts my original operating habits. This kind of repeated switching makes people feel quite troublesome and inconvenient.
Secondly, content generation is a bit useless.
The answers it generates are crude and verbose. I asked some business and AI-related questions, and it could give a large number of answers. At first glance, it seemed quite comprehensive, with everything from time, place, people, and reference materials.
But on closer inspection, the content structure is loose, the logic is messy, and there is even duplicate and redundant information. Although it may seem like a lot of information, I still have to spend my own time sorting and screening the really useful parts, which is really a headache.
What’s even more annoying is that the answer also quoted a bunch of reference materials. Some of the content cannot be directly used and has to be re-edited, and the data has to be checked one by one. It’s too inconvenient to operate with a mobile phone. I can only send the generated content to the WeChat Transport Assistant or my own dialog box, and then use my computer to open it and see it.
The result came the question:
The computer does not support viewing these content at all. In this case, why don’t I operate directly on my computer? It’s a bit ridiculous to have to do a lot of trouble on your mobile phone.
There is something even more speechless. Fourth point: Historical records are unreliable.
If you don’t click the Share button, send it to friends or save it in a conversation, the next time you open the history, you may still be able to see it on the first day, but you will find it regenerated the next day.
In other words, search history is sometimes not saved at all and is completely random: some can be kept, some cannot. I still don’t understand its logic until now: Will historical records be preserved? Therefore, every time the content generated, we have to share it quickly, for fear that it will be gone and the work will be in vain.
Let’s talk about the fifth point: the memory is too bad. I found it a little “forgetful” and couldn’t remember what we had talked about before.
For example, I asked first: Who is Wang Zhiyuan? It answered well. Then I added: The one who wrote “Compound Interest Thinking”. As a result, it seemed to completely forget the previous problem and only talked about what compound interest thinking was without mentioning its name at all. This makes me a little confused.
What is even more outrageous is that some of the materials it cites are copied from unknown public accounts, and the quality is very average, just like those “chicken soup for the soul” articles. I even wonder why it quotes unreliable content?
Sixth point: The knowledge base has not been opened up.
I think the knowledge base it uses now seems to have nothing to do with the content of Weixin Official Accounts. Most of the generated content is based on data from the DeepSeek R1 model itself. High-quality articles on the public account are rarely used, and many high-quality content is not even mentioned at all.
I feel that its search scope is particularly narrow and has not yet been truly integrated into the WeChat ecosystem. I don’t know if I am correct, but I just feel it personally.
02
Writing here, I couldn’t help but wonder: Under what circumstances would the deep thinking function be used? Or, when will I use System 2 in my brain: that rational, calm thinking model?
I think there are probably three situations:
- 1) When solving complex problems, such as: looking for a report on your mobile phone or searching for some key data;
- 2) When doing logical reasoning, such as: analyzing potential risks of the project during meetings;
- 3) When doing creative tasks, such as: suddenly wanting to find some creative inspiration on the road.
However, WeChat DeepSeek R1 does not yet support uploading files, which greatly reduces its practicality when handling these complex tasks. This raises a question: Now that there are so many specialized AI apps available, why do you have to search in WeChat? I still don’t understand it yet.
After all, DeepSeek R1 is really not suitable for in-depth thinking in the WeChat scenario. Why? Because of scenario adaptation issues.
What is WeChat? Spending on the circle of friends, replying to messages, and viewing public accounts are all “quick swiping” scenes. Now I have no patience to even read the public account, let alone think deeply. And what does deep thinking require? Be quiet, focus, and have enough time to process information.
Imagine:
You are checking data on your mobile phone, and suddenly a message pops up on WeChat. As soon as you get distracted, your train of thought is broken. After finally finding the answer, the content is long and messy. The mobile phone screen is crowded into a ball, and you have to swipe for a long time to copy a paragraph; What’s even more angry is that it can’t be opened on the computer, so you can only stare at the small screen.
How to do this? To put it bluntly, the “fast pace” of WeChat and the “slow demand” of deep thinking are like a fight between the rational brain and the emotional brain, and they cannot cooperate well at all.
Therefore, this kind of scenario adaptation not only amplifies the flaws in functional design, but also greatly reduces its value in practical applications; however, the combination of WeChat and DeepSeek R1 does have many highlights and places worth looking forward to.
For example, if it can improve the quality of content generation, reduce lengthy information, and optimize logic, DeepSeek R1 will be much more practical; at the same time, it can also make better use of high-quality resources in the WeChat ecosystem, such as public account articles and video account content, close to users ‘daily usage habits, and truly integrate into life scenarios.
Or: Compared with ChatGPT or Baidu Wenxinyan, the biggest advantage of DeepSeek R1 plus WeChat is that it has a large user base and rich and diverse application scenarios.
Functions such as payments and Mini programs have long been deeply embedded in everyone’s daily life. If DeepSeek R1 can be deeply integrated with these functions and provide intelligent support, its competitiveness will definitely be much better than other similar products.
Imagine that when you want to go to the Forbidden City to play but can’t find the ticket purchase entrance, DeepSeek R1 can not only directly help you call up the ticket purchase Mini programs, but also organize relevant information clearly, saving you the trouble of looking around. This “one-stop service” may be a highlight in the future.
Therefore, if we can do these two points well, WeChat and reasoning models may really become a lifestyle assistant.
03
I also noticed a lot of complaints outside. Some people believe that:
WeChat wants to grab Baidu’s search market share and even replace other AI search engines; however, I think that WeChat’s access to DeepSeek R1 should not have much impact on the AI search market.
Although WeChat has more than 1.3 billion monthly active users, competition in the AI search market has long been heated. Platforms such as Baidu, Google, and Bing have long built high walls in terms of technology, data accumulation and user experience.
According to a report by StatCounter GlobalStats, by the end of 2024, among the desktop search engine market share in China, Bing will account for 52.96%, ranking first; Baidu will account for 27.36%, ranking second; Haosou will account for 15.2%; and Google will account for 1.79%.
Although the number of daily active users of WeChat Souyisou has exceeded 700 million, its search function mainly serves content distribution within the WeChat ecosystem.
Moreover, WeChat also made it clear that AI search will not use users ‘personal information and private data, but will only integrate public accounts and public information on the Internet, which also limits its depth of data utilization to a certain extent.
Looking at the global market, Google is still the dominant force in search engines, with a global market share of 89.74%; in China market, Bing’s desktop market share has surpassed Baidu, reaching 52.96%, while Baidu still accounts for 58.21% of the mobile market.
These data are enough to show that traditional search engine giants are still very competitive in their respective fields. So, want to disrupt the market with one function? I don’t think it’s that simple and not that fast.
Another point is that if AI search wants to be successful, it is not enough to have user size alone. It must also continue to invest and optimize in many aspects such as algorithms, data accumulation, and application scenarios.
Therefore, I think that WeChat’s access to DeepSeek R1 is more an optimization of its own ecosystem than a direct impact on traditional search engines.
Therefore, from the perspective of market positioning, WeChat AI search is not intended to subvert traditional search engines, but focuses on serving the content distribution and intelligence needs within the WeChat ecosystem. This differentiated positioning determines that it will not affect Baidu and Google in the short term. and other giants pose a threat.
However, in the long run, if DeepSeek R1 can better integrate resources within the WeChat ecosystem, such as public account articles, video account content and Mini programs services, its potential is considerable.
04
Although there are still many problems with WeChat combined with DeepSeek R1, from a strategic perspective, WeChat access to this function may have deeper considerations. I think it is mainly highlighted in three aspects:
First, geopolitical strategic height.
The AI field has become the focus of global competition. There are many top AI companies in the United States, such as OpenAI, Google’s DeepMind, Microsoft’s Azure AI, Anthropic, Cohere, etc.
These companies are very strong in natural language processing, machine learning, computer vision and other fields, covering almost all aspects of AI applications;
They are not only technologically advanced, but also form a complete ecosystem. From basic research to commercial applications, from open source frameworks to enterprise-level solutions, the entire industrial chain has been opened up. This layout has made the United States ‘position in the AI field more solid.
In contrast, China’s AI industry is also struggling to catch up, and basic research and core technologies are accelerating development. I saw on GitHub that in addition to Ali’s Qwen series’s outstanding performance, we still need more power to join.
Open source projects can not only demonstrate the progress of China’s AI, but also provide more choices for developers around the world. More importantly, they reflect China’s determination to independently innovate in the field of AI.
The emergence of DeepSeek R1 shows that domestic AI technology has the potential to compete with international giants. At this time, WeChat, as a leader in the social field, connects to R1, which not only reflects the unity of domestic AI, but also sends a signal to the world:
China has the ability and determination to compete with international giants in the field of AI. In other words, WeChat’s access to DeepSeek R1 is not only a business decision, but also a key step in a technology game.
Second, the capital market is also optimistic.
Now Silicon Valley is talking about Agents, which is, combining reinforcement learning with a big language model. The big language model is mainly about understanding the user’s meaning, while intensive learning is about making decisions.
However, the big language model is still a little weak in complex decision-making, which limits its application in complex scenarios.
Before the emergence of real General Artificial Intelligence (AGI), AI was more a tool for Tencent to consolidate existing businesses and improve efficiency, rather than a disruptive innovation.
Stocks have risen sharply in this wave, largely due to DeepSeek’s access. The market has reassessed Tencent’s technical layout and future development potential, and confidence has suddenly risen. The optimism of the capital market is not due to the breakthrough of AI technology itself, but recognition of Tencent’s strategic value in the AI field.
In addition, the introduction of AI technology has also opened up new growth space for Tencent, such as data accumulation, algorithm optimization and application scenario expansion. These factors add up to Tencent’s share price in a short period of time.
Therefore, for Tencent, AI is not only an important force in promoting the company’s overall development, but also the key to enhancing market confidence.
Finally, from the perspective of user needs and ecological closed-loop, Tencent’s technological development is beneficial.
WeChat is a super App with rich and diverse functions: chatting, paying, reading articles, and using various services, covering almost all aspects of users ‘daily lives.
This access to DeepSeek R1 is to integrate advanced AI capabilities into the functions that 1.3 billion users use every day, allowing ordinary users to easily enjoy cutting-edge AI technology.
This seamless integration not only improves the user experience, but in turn inspires the team to speed up research and development, forming a virtuous cycle.
From the perspective of application scenarios, the deep thinking model has great potential. Tencent’s own products, such as Tencent Yuanbao, ima.copilot, documents, etc., are gradually accessing these AI capabilities. At the same time, this technology can also provide services to B-end customers, further expanding Tencent’s ecological boundaries.
A few days ago, Tencent also launched its own Thinker (T1).
In the current fierce competition in AI technology, Tencent can quickly add intelligent functions to its products with DeepSeek R1 ‘s mature natural language processing and reasoning capabilities.
Taking Tencent Yuanbao as an example, after connecting to DeepSeek R1, it can not only search and parse documents online, but also provide more accurate AI writing and reasoning functions. This not only makes it more convenient for users to use, but also helps Tencent accumulate more real data.
In the long run, this strategy has left Tencent valuable time and space for optimization. On the one hand, the access to DeepSeek R1 can quickly improve the intelligent level of products.
On the other hand, Tencent can collect a large amount of real data through actual use scenarios, thereby improving AI technology in a targeted manner, gradually creating a dedicated AI model that is more suitable for its own business and user needs, and can also use this time to explore more AI application scenarios, deeply integrating AI technology into various products and services.
Therefore, in this way, product capabilities are not only improved and valuable experience is accumulated, but also the ecological landscape is further expanded. Wouldn’t it achieve three goals with one stone? Letting users complain more may not be a good thing.
The above is purely personal. What do you think?
