Large language model companies are seeking lightweight deployment to lower user entry barriers and capture traffic entry points. Recently, ByteDance's Doubao and Moonshot AI have released browser plugins. Moonshot AI calls its latest Kimi browser plugin a "lightweight mini search," simultaneously launching two features: point-and-ask and summarizer. ByteDance's AI assistant Doubao has updated its browser plugin version to 1.3.0.
Behind the lightweight deployment is the traffic anxiety of large language model companies. According to Quantum Bit statistics, in June this year, ByteDance's Doubao AI's DAU exceeded 2 million, while Kimi and SkyAI's DAU surpassed 1 million. As of July 17, the Chrome store shows that Doubao browser downloads have exceeded 90,000, while Kimi browser plugin downloads have surpassed 30,000.
Reasons for large language model companies choosing AI search entry points include:
- Current large language model products are highly homogeneous with unclear technical differences
- The investment window for model-as-product has passed, increasing monetization pressure
- Becoming a search entry point through browser plugins reduces user costs
- Seeking commercial monetization models
A business leader from a large language model company stated that AI intelligent assistants are difficult to profit from in the short and medium term, serving more as brand promotion.
Domestic AI search market competitors also include Baidu, Quark Search, Zhihu, 360 AI Search, and Kunlun Wanwei's Tiangong Search. More players may enter in the future.
Compared to AI intelligent assistants, browser plugins better highlight the commercial ambitions of large language model companies. Search already has a clear advertising profit model. For example, in 2023, Google's website advertising revenue reached $206.5 billion, while Microsoft's Bing search and advertising business search ad revenue reached $12.21 billion.
Domestically, China's search engine advertising market size reached 106.3 billion yuan in 2020. Baidu's online marketing service revenue in 2020 and 2023 was 72.8 billion and 81.2 billion yuan, respectively.
AI search enhanced by large language models can simultaneously complete chart drawing, analysis, and network retrieval. An industry insider believes that good search should accurately grasp user intent and present desired information in one go.
A hard tech investor pointed out that large language model companies controlling models is beneficial for scenario optimization, cost control, and rapid product iteration. However, AI search faces challenges in business models.
The current search landscape is still dominated by traditional tech giants. According to StatCounter data, in January 2024, Google Search's desktop and mobile market shares were 82% and 95%, respectively. Domestically, in April 2024, the top three search engine market shares were Baidu, Bing, and Sogou, with 62%, 18%, and 4% respectively.
Lightweight deployment and vertical scenarios have become key words for large language model implementation this year. However, large language model companies still need a long time to achieve positive profitability.