By Wang Huiying

Edited by Ziye

With the explosive demand for large model applications, everyone is scrambling in this arena.

Among them, search is one of the most closely related landing applications of large models, and discussions about "AI search" are increasing. Especially with the emergence of Perplexity, a Silicon Valley star company, domestic internet players have also seen new opportunities.

This AI startup, with its first AI-native conversational search product Perplexity, has nearly 10 million monthly active users and a valuation of $2.5-3.5 billion.

When users input a question to Perplexity and click search, they can get an answer that has been filtered and integrated by AI. This answer reduces the hallucinations of generative AI and provides precise traceability for each sentence.

Perplexity's success has validated the industry's main narrative over the past year for large model applications: Compared to model capabilities, the application itself is a more important matter.

The latest entrant in China is Zhihu, which started as a Q&A community. On June 29, Zhihu released a new AI product, "Zhihu Direct Answer," at the 10th Salt Club Youth Conference. Based on creators' real Q&A data, it can generate answers based on user questions, making it Zhihu's first independent AI product.

In the view of Zhihu founder Zhou Yuan, Zhihu Direct Answer is a new productivity tool and should also become a connector for people to discover the world.

This also signifies that Zhihu has officially entered the "AI search" arena for large model applications.

In fact, since last year, Zhihu has been preparing for the launch of independent AI products. In April 2023, Zhihu released "Zhihai Tu AI"; in March this year, Zhihu released the new feature "Discovery·AI Search".

In this global competition for the landing applications of AI large models, Zhihu, which started as a community, is a special existence. As a knowledge-based Q&A platform with 100 million monthly active users, but for a long time, the secondary market's pricing of Zhihu has been difficult to truly reflect its community ecological value. Can AI search reactivate Zhihu's community vitality and increase investors' expectations?

The answer is full of variables. As Zhihu Direct Answer's official account answered in a question about "Zhihu Direct Answer": "The AI era has just begun, and the road ahead is still long."

1. What's Different about Zhihu Direct Answer?

The competition for the landing of large model applications is in full swing, and AI search has become a contested territory. Players like 360 AI Search, Kunlun Wanwei's Tiangong AI Search, Tencent's Yuanbao, Kimi from the Dark Side of the Moon, and Baidu are racing against each other.

Now, Zhihu, a community Q&A platform, has also joined the AI search table.

At the Zhihu Youth Conference on June 29, Zhihu founder Zhou Yuan released Zhihu's new AI product - Zhihu Direct Answer. Based on Zhihu's content library and Q&A genes, Zhihu Direct Answer can provide users with an experience of asking questions, searching, result generation, and summarization.

It is reported that "Zhihu Direct Answer" is the official productization of Zhihu's AI search function, with an independent domain name "zhida.ai" and service entry, which has been launched on the PC side. In the future, Zhihu Direct Answer will gradually promote App development and introduce multi-modal capabilities, while integrating with the Zhihu community and actively exploring external cooperation.

Source: Zhihu Direct Answer PC version

Zhihu's decision to create Zhihu Direct Answer was not surprising to the market. From last year's large model arms race to this year's large model application landing competition, Zhihu has been one of the first players to layout in China.

Strictly speaking, whether it's Zhihai Tu AI or Zhihu's AI search function, Zhihu's previous actions were all preheating for the landing of large model applications. Zhihu Direct Answer is Zhihu's first independent AI product, marking Zhihu's formal participation in the battle for the landing of AI large models.

The first thing Zhihu Direct Answer needs to answer is differentiation, which is its uniqueness. More importantly, Zhihu Direct Answer needs to differentiate itself from traditional Q&A.

Zhihu told Lianxian Insight that in the traditional Q&A model, users who want to get answers are often limited by time and space and need a certain creation cycle. In contrast, Zhihu Direct Answer is based on Zhihu creators' real Q&A data, integrating multiple scenarios of search, generation, follow-up questions, finding content, and finding people, further shortening the distance between content demand and quality answers, and amplifying the circulation effect of community creators and their content.

Specifically, Zhihu Direct Answer has two major characteristics: one is to reduce generative AI hallucinations through traceability, and the other is that each answer allows users to selectively trace their sources.

Zhihu Direct Answer Traceability, Source: Zhihu Direct Answer PC version

Currently, the biggest problem with AI search is the authenticity of content. Since the decision-making process of AI search is often difficult to explain, users may have doubts about the search results, affecting their trust in AI search. Especially in the field of Q&A, the authenticity and accuracy of answers are even more important.

In the content of Zhihu Direct Answer, users can see which answerers contributed the answer and where the answer is quoted from. At the same time, multiple data sources are introduced as auxiliary to provide more accessible information channels for search. Ma Kui, the product manager of Zhihu Direct Answer, said that Zhihu Direct Answer strictly traces its sources to ensure the reliability and authenticity of the answers, while also reducing the hallucination problem in the AI field.

In addition, Zhihu Direct Answer provides simplified and in-depth generation results. When asking questions, users can choose between in-depth Q&A or concise Q&A. The original intention of this function design is that Zhihu has found that some users want to do more in-depth research, while others want to know the answer quickly.

Behind this is Zhihu's years of Q&A genes. As the largest online Q&A community in China, Zhihu has a large number of professional answerers and high-quality Chinese corpus. According to official data, as of Q1 2024, Zhihu has accumulated about 804 million pieces of content, with a total of 73.6 million content creators on the platform.

AI search has already become an industry-recognized trend under the landing of large model applications, and Zhihu Direct Answer has no shortage of competitors. Perhaps, more importantly than speed, Zhihu Direct Answer needs to run with its own characteristics.

2. Community Remains Zhihu's Top Priority

On the internet, discussions about what Zhihu is have lasted for a long time.

In the era when Zhihu was still being compared to Baidu Knows, Zhou Yuan responded: "Baidu Knows seems to be Q&A on the surface, but actually, it's search behind it. Zhihu is Q&A on the surface, but actually, it's an SNS (Social Networking Service) behind it."

The attribute of the community determines that, unlike traditional search, Zhihu has had a different definition of "knowledge" from the beginning. Zhihu's "knowledge" is not standardized knowledge, nor are the answers formulaic answers, but more authentic answers provided by answerers based on their own experience and professional knowledge.

However, as a community product, it is inevitable that with the popularization and sinking of the platform, the overall content tone has shifted or even become biased. Some users have found that on Zhihu, in order to grab popular replies under a certain topic, professional replies of several thousand words are often less popular than a "witty" reply.

From the perspective of creators, discovering in more topic traffic pools and "showing wit" at the lowest cost in each traffic pool has become an efficient and realistic choice. This has led to significant problems of "content watering" and "creator exodus" faced by Q&A platforms like Zhihu.

Zhihu has also realized that the traditional content ecosystem needs to change. If it relies solely on UGC content, then watering down and soft advertising are inevitable; if it only allows answerers to earn money through methods such as Q&A consulting small classes, the enthusiasm of answerers is also not high.

Therefore, the first step of Zhihu's transformation is to establish a PGC content ecosystem. Zhihu continuously expands the number of official accounts and the content matrix screened by Zhihu's official team, and ensures their exposure, allowing users to see the "strictly selected" content from Zhihu; secondly, Zhihu separates the "storytime" content and establishes a "novel" channel, and expands a large number of content sources through purchasing, signing, and converting.

Under the new content ecosystem, the essence of Zhihu has not changed. Q&A remains the most basic model, and the community ecosystem is also the foundation, but new commercialization methods have emerged from the community.

Zhihu's financial report for the first quarter of 2024 showed that during the statistical period, paid reading, marketing services, and vocational education contributed 46.8%, 34.4%, and 15.1% to revenue, respectively. Among them, paid reading revenue reached 450 million yuan, firmly occupying the top source of income.

Judging from the data, both the paid reading business driven by SaltSelect articles and the vocational education business driven by Zhixuetang are inseparable from Zhihu's community ecology and users. Financial report data shows that Zhihu had an average of 89 million monthly active users and 14.8 million monthly subscription members in the first quarter of 2024.

As early as the Zhihu Discovery Conference in March this year, Zhihu emphasized that community ecological construction is the foundation of Zhihu, and Zhihu needs to rely on the wisdom and strength of more community users. Therefore, Zhihu proposed a long-term strategy to build a trustworthy community. In Zhou Yuan's view, the trustworthiness of the community lies in professional and authentic human sharing, serious and friendly interpersonal communication, and transparent and efficient information mechanisms.

The launch of Zhihu Direct Answer, by introducing direct answers to nourish the platform's content ecology, is also a thought of Zhihu.

When asked how Zhihu Direct Answer interacts with other Zhihu community functions, Zhihu told Lianxian Insight that Direct Answer provides a new way to access Zhihu community content. While absorbing more community content, Direct Answer will also nourish the community, whether it's more content creation or incentives for creators, building a high-quality information sharing platform with the Zhihu community.

In other words, amplifying the value of the Zhihu community through AI technology and realizing Zhihu's vision of "starting from the community, but not stopping at the community" may be the longer-term goal of Zhihu Direct Answer.

3. Direct Answer Must Also Play the Role of "Connection"

The essence of Zhihu Direct Answer is AI search. What was search like before AI search arrived?

According to a Microsoft survey, people use search almost more than 10 billion times a day, covering a wide range of topics, from astronomy to geography, humanities to cyberpunk, and more.

That is to say, when we want to acquire new unknown information, we need to search, and the process of satisfying the demand requires a search tool to complete. In addition to providing answers, the most important thing for search is to complete the connection between the asker and the answerer.

The same is true for AI search. At the Zhihu Youth Conference, Zhou Yuan emphasized several times that Zhihu Direct Answer should play the role of "connection".

As Zhou Yuan said, "Zhihu Direct Answer" is a new productivity tool and should also become a connector for discovering the world, as well as a bridge connecting people. "Zhihu Direct Answer" will help everyone better "discover the world through questions" and build a more imaginative, creative, dynamic, and vibrant community around the tool and everyone.

"In the AI era, some answers are also compressed into AI large models, and we need tools to reconnect." Zhou Yuan said that Zhihu Direct Answer is an attempt to integrate search, Q&A, and follow-up questions.

Zhihu founder Zhou Yuan, Source: Zhihu WeChat public account

After experiencing Zhihu Direct Answer, you will find that Zhihu Direct Answer can not only help users find content, but more importantly, help users connect with the people behind the answers, mainly answerers.

As a community, interaction has always been one of Zhihu's major features. The "find people" function of Zhihu Direct Answer aims to interact with the Zhihu community and enhance Zhihu's interactivity as a community.

Zhihu has a large number of high-quality creators who were previously difficult to discover due to the platform's recommendation mechanism and traffic distribution. But now, through Zhihu Direct Answer, users can directly find them by searching for who the answerer is or what excellent answerers there are in a certain field.

Zhihu revealed to Lianxian Insight that in the future, Zhihu Direct Answer may incentivize users' high-quality creation on Zhihu Direct Answer in terms of content visibility and recommendation, interaction and traffic feedback, and copyright protection.

Supporting Zhihu Direct Answer's operation is the underlying capability of Zhihu's AI large model. In March 2023 and 2024, Zhihu has released the "Zhihai Tu AI" large model and the community-native AI function "Discovery·AI Search," completing two important progressions from model development to open product applications.

Zhihu found that users use the AI search function to ask and answer questions with an average of nearly three rounds. How to better link the logical relationship between multiple Q&As and achieve "multi-round dialogues" is the focus of its AI search model capability upgrade.

The upgraded "Zhihu Direct Answer" can more accurately understand the relationship between users' new questions and past questions in the "follow-up question" function, accurately identify intentions and referential relationships for multiple Q&As, and provide more accurate answers.

That is to say, compared to search engines that match information relevance, Zhihu Direct Answer does more work on understanding instructions, making search results closer to what users want.

In other words, with the support of AI large models, compared to traditional search, Zhihu Direct Answer is more like a dialogue interaction between people.

Zhihu told Lianxian Insight that creators are the core of Zhihu, and Zhihu Direct Answer not only connects users with content but also tightly integrates users, content, and creators, making every user's question a collision with a large number of excellent creators.

It is worth mentioning that Ma Kui, the product manager of Zhihu, revealed, "Our model now has multi-modal capabilities, and we are developing related functions because Zhihu content is not just text, but also a large number of images and videos."

Under Zhihu Direct Answer, Q&A is not a limitation, but rather an expression of unlimited possibilities. What Zhihu aims to do is not just Q&A, but to build a public forum and create a new connection.

Returning to the question of what Zhihu is, Zhou Yuan's description is more apt: "To explain Zhihu in one sentence, it's about moving a lot of knowledge that hasn't been transferred to the internet from our minds onto the internet."

Driven by AI, this statement will be implemented more thoroughly. For a content community like Zhihu, how to share wisdom in the AI era is a topic worth exploring. This also means that Zhihu still needs to continue valuing the community's importance and find ways to activate its value.