Source | BohuFN

Recently, Kimi quietly launched its exploration version, which even briefly became a trending topic on social media. The team behind it, Moon's Dark Side, revealed that the Kimi exploration version was the result of overtime work during the National Day holidays, but even so, they did not anticipate the overwhelming enthusiasm from users.

It is reported that the Kimi exploration version boasts AI-powered autonomous search capabilities, capable of simulating human reasoning processes, decomposing complex problems into multiple levels, performing in-depth searches, and immediately reflecting on and improving results. This enables users to efficiently complete complex tasks such as analysis and research.

However, Kimi is not the only player focusing on AI search. Recently, Baidu announced that its AI language model, Wenxin Yiyan, has been upgraded to "Wenxiaoyan," positioning it as an intelligent assistant for "new search."

With the support of large language models, the traditional search industry has frequently blossomed with new innovations. There are native AI search tools like Perplexity and Mita, as well as AI chatbots like Kimi and Tiangong AI vying for market share. The search market is no longer dominated solely by Baidu.

While new players continue to explore search scenarios, veteran player Baidu is also actively defending its position. Does AI search stand a chance in transforming the traditional search industry?

01 Targeting search scenarios, Kimi aims to "learn from others' strengths and compensate for its weaknesses"

After the official launch of the Kimi exploration version, its product manager stated, "If Kimi cannot find certain information, it is highly likely that users will also struggle to locate it through traditional search engines." Clearly, Kimi aims to revolutionize traditional search paradigms.

In traditional search engines, represented by Baidu, users pose queries, and the engine then crawls webpages for information, sorts it according to its own algorithms, and presents the results to users.

This process harbors two hidden pain points: firstly, users need to phrase their questions simply and directly, potentially breaking down complex queries into multiple simpler ones; secondly, the ordering of search results often cannot escape the influence of advertising mechanisms.

The Kimi exploration version seeks to address these pain points by decomposing complex queries into simpler steps, guiding the model to reason autonomously and solve problems, thereby shifting the large language model's approach from "fast thinking" to "slow thinking." Notably, the o1-preview model recently released by OpenAI also adopts a "slow thinking" approach.

In the past, while large language models were "intelligent," they did not truly think like humans; they merely mimicked human-like behavior through extensive data training. However, under the logic of "slow thinking," future large language models will genuinely possess the ability to think and reflect autonomously, akin to humans.

According to examples showcased by Moon's Dark Side, when users wish to know how many companies on the top 50 technology list have headquarters in the capital city, they used to need to search for each company individually. In contrast, the Kimi exploration version can rapidly scan webpages and summarize the answer. It is reported that the search volume of the Kimi exploration version is ten times that of the standard version, enabling in-depth reading of over 500 pages with a single search.

Beyond its ability to integrate information, the Kimi exploration version also excels in conducting in-depth searches and analytical research. For instance, when asked which stock – BYD or Moutai – would yield higher returns if purchased during the same period, Kimi can directly calculate the results.

In this query, Kimi demonstrates its search and learning capabilities by decomposing the complex problem, locating relevant data for both stocks, learning how to calculate "returns," and ultimately arriving at an answer.

Furthermore, the Kimi exploration version enables fuzzy searching. For example, when asked "Which Silicon Valley company has produced a product similar to the iPhone?", Kimi can infer the company's name based on key details, even self-reflect and supplement additional information.

With the launch of the Kimi exploration version, Moon's Dark Side has also expressed its intention to further delve into search scenarios. As early as July this year, Kimi introduced a browser extension version, offering two key functions: highlighting and asking questions, as well as summarizing abstracts, helping users enhance their search efficiency.

In recent years, the search market at home and abroad has undergone significant changes, with new players like large language model titans, native AI search tools, and AI chatbots entering the AI search arena. Even OpenAI, a giant in large language models, released an AI search product called SearchGPT in July this year, presenting an opportunity for Kimi, which is still seeking a commercial path.

On one hand, search scenarios can further leverage Kimi's strengths. Considering that Kimi's core user base primarily comprises knowledge workers and university students with high-frequency search needs, search is the most suitable scenario for AI chatbot functionalities. It not only rapidly answers user queries but also provides clear links to relevant sources.

On the other hand, traditional search engines are gradually declining in popularity as younger generations begin to assign specific search tasks to different platforms, such as "Dianping" for restaurant recommendations and "Xiaohongshu" for travel planning. This presents an opportunity for Kimi.

However, the Kimi exploration version is not without its shortcomings. Firstly, the "hallucinations" of large language models are still difficult to avoid. For instance, Kimi may cite data from articles published by other users as sources, which cannot fully guarantee the credibility and timeliness of the information.

Additionally, based on feedback from some users, when queries are relatively vague, the Kimi exploration version struggles to accurately understand user intent, and its overall reasoning ability is still weaker than OpenAI's o1-preview model.

02 Focusing on "new search," Baidu wages a defensive battle

Nonetheless, Kimi's imperfections do not hinder its challenge to Baidu, given that many users have long been frustrated with the latter in the search market.

As a gateway to traffic, search engines have always been a coveted territory among internet companies. Dominating the search engine market equates to having the power to distribute information and connect various service ecosystems.

In the early days of the internet, Baidu, as a pioneer in the search market, leveraged its technological leadership and content ecosystem to establish a dominant position in Chinese language search. Other search engine competitors, hindered by factors such as brand recognition and regulatory policies, failed to challenge Baidu's status as the "search giant."

However, when there was only one search entry point, users began to feel "captive," with Baidu's "pay-per-click" advertising mechanism being the most criticized aspect. As long as advertisers were willing to pay, their information would appear at the top of search results, potentially distorting the authenticity of information to some extent.

According to Baidu's 2005 prospectus, its online promotion revenue once accounted for 90% of its total revenue, but this trend is changing. From the second quarter of 2023 to the present, Baidu's online marketing revenue has increased by 15%, 5%, 6%, and 3%, respectively. In the second quarter of this year, Baidu's online marketing revenue was RMB 19.2 billion, a year-on-year decrease of 2%, marking the beginning of negative growth.

The slowdown in Baidu's advertising revenue can be attributed to two main factors. Firstly, under the impact of emerging social media platforms like short videos and live streams, traditional search advertising's share of internet advertising has continued to decline. Secondly, users are being diverted away from search engines, as AI search gradually becomes an alternative to traditional search, eroding Baidu's market share.

According to the latest data from Statista, while Baidu still ranks first in China's search market in May 2024, its market share has dropped to 55.85%. In November 2021, its market share peaked at 86.82%.

Naturally, Baidu is aware of the market challenges. Since the dawn of large language models, Baidu's Chairman Robin Li has emphasized that all products should be rebuilt using large language models, with search services poised to become the "killer app" of the AI era.

Baidu has also attempted to reinvigorate its search business with AI. In May this year, Baidu's management revealed that approximately 11% of search results are now calculated using AI technology, with the proportion of AI-generated content in Baidu search set to increase in the future.

However, despite Baidu's early start in large language models and its exploration of AI search capabilities years ago, it has yet to deliver significant breakthroughs.

In May this year, Baidu launched an AI-powered Q&A function, acting as an intelligent assistant when users search for information, refining webpage content and summarizing answers. Nevertheless, compared to other AI search tools, Baidu's responses lack depth and breadth, particularly when dealing with questions requiring reasoning, where AI-powered Q&A struggles.

However, a month ago, Baidu upgraded Wenxin Yiyan to "Wenxiaoyan," positioning it as an intelligent assistant for "new search." Unlike Kimi's focus on deepening search capabilities, Baidu, which originated from search, chose to introduce an AI assistant tailored for "new search."

Currently, the primary differences between "Wenxiaoyan" and other AI search products lie in two aspects. Firstly, "Wenxiaoyan" benefits from Baidu's rich ecosystem support. For instance, when users inquire about transportation routes or travel itineraries, "Wenxiaoyan" can directly access Baidu Maps, providing a broader range of services such as transportation and dining options.

Secondly, "Wenxiaoyan" offers five core scenarios: asking questions, chatting, writing articles, drawing images, and assigning tasks. Beyond AI search, Baidu also provides users with a plethora of personalized AI services, essentially serving as Baidu's "AI suite."

While both Baidu, the veteran player, and Kimi, the newcomer, are vying for market share, their strategies differ. While Baidu pursues diversification, Kimi focuses on specialization. Nevertheless, their ultimate goal is the same: to address their respective ecological shortcomings.

03 The odds of "Kimi-like" players penetrating Baidu's stronghold

In the long run, AI search represents the prevailing trend, attracting numerous cross-industry players seeking to capitalize on this opportunity. As a traditional player, Baidu is actively integrating AI capabilities and exploring new product models.

However, transforming the traditional search industry through AI search is far from straightforward. Firstly, while the entry barriers for search products are lowering due to large language models, most AI search products adhere to a simplified search path and enhanced information matching approach. This results in AI engines appearing convenient but often falling into the trap of "useless information overload."

Moreover, users turn to AI search products hoping to streamline their search process. Yet, considering issues like information source pollution and large language model "hallucinations," users still need to double-check critical information. Building absolute trust in AI search remains a formidable challenge.

Secondly, commercialization poses a hurdle. Both Baidu and Kimi face pressure to monetize their search services. Kimi is still in the "burn money for scale" phase, and most large language models follow only two monetization paths:

TOB (Business-to-Business) faces fierce competition and thin margins, while TOC (Business-to-Consumer) relies heavily on subscription fees. Cultivating user willingness to pay requires outstanding product capabilities, which explains Kimi's continued refinement of its search capabilities.

Furthermore, insiders revealed that AI search is just the first capability of the Kimi exploration version, with other new capabilities, potentially multi-modal abilities, to follow. This suggests that Kimi, like Baidu, aims to continually enrich its "AI suite."

Baidu's stance is more nuanced. From a commercial perspective, AI search represents a significant disruption to traditional search models. Since AI search can directly provide answers, it becomes challenging to embed advertising content, potentially transforming Baidu's once lucrative search business into a second startup endeavor.

Therefore, until these two challenges are fully addressed, it is unlikely that AI search will completely replace traditional search engines. Nonetheless, enhancing search efficiency and accuracy through large language models, thereby reshaping the search market, is already underway.

Currently, "Kimi-like" players are attempting to penetrate Baidu's stronghold, but Baidu will not respond passively. The evolution of search engines is inherently a process of industry innovation. Each breakthrough in search technology and the entry or exit of different players has the potential to alter the search market landscape.

Before these issues are resolved, AI search products must actively explore product and monetization models, striking a balance between commercialization and technological advancement. Furthermore, they must enrich their content ecosystems to provide more precise and personalized search results, thereby effectively attracting and retaining users.

In this transformational battle for search engines, it remains uncertain who will emerge victorious. However, one thing is certain: the search market will no longer be dominated solely by Baidu.

The cover image and accompanying illustrations belong to their respective copyright holders. If the copyright owner believes that their work is not suitable for public browsing or should not be used free of charge, please contact us promptly, and we will immediately rectify the situation.