Since the beginning of this year, the AI industry has been surging. The large model accelerates the update iteration, the price drops rapidly, and extends its tentacles to a wider market and more subdivided fields. AI search is becoming a must for platforms in the development of large models.
Zhou Yuan, founder and CEO of Zhihu, recently officially released the AI search tool "Zhihu Direct Answer" at the "10th Salt Club New Knowledge Youth Conference", which is based on the "Discovery· AI Search", upgraded the function of questioning, content traceability and other functions.
Zhou Yuan
Is AI search on the eve of an explosion?Since the beginning of this year, investment in the AI industry has shifted, and AI search has become a hot spot in the industry.
As of October 2, 2023, startups focused on generative AI applications have received nearly $18 billion in venture capital funding that year, an increase of more than 360% from the entire year of 2022, according to the State of AI 2023 report produced by AI investor Nathan Benaich and the Air Street Capital team. Overall, a lot of investment in 2023 will be focused on basic large models.
In 2024, the reduction of computing power costs brought about by the accelerated evolution of technology and the intensification of competition among large model manufacturers have jointly catalyzed the price reduction of large models.
On the one hand, the cost reduction promotes the inclusiveness of the basic large model, and also provides assistance for platform manufacturers to improve their operations.
From the perspective of Zhihu's investment in large models, according to the financial report, in the third quarter of 2023, Zhihu's net loss will be 278 million yuan, one of the important reasons is the cost of large models, and by the first quarter of 2024, Zhihu's net loss will be 166 million yuan, which has been significantly narrowed from the same period last year. The cost reduction of large models has provided more opportunities for businesses such as Yanyan Story, which benefits from AI empowerment and efficiency improvement, and in terms of results, the proportion of its paid reading business continued to grow in the first quarter, ranking first as the largest source of revenue.
Regarding the role of AI in Zhihu's performance growth, Zhou Yuan once said that he believes that the potential of AI search function and the value of the trustworthy community will provide a new impetus for Zhihu's sustainable growth.
On the other hand, platform manufacturers are also looking for more model application areas outside of the price war. Acumen investment then turned to model applications.
Among them, AI search is one of the main model application areas to which investment has turned. Taking Perplexity, which focuses on AI search, as an example, it has started to make AI search applications long before the release of ChatGPT, and has received investment from Nvidia this year, and may have received another $10 million to $20 million investment from SoftBank in the near future, with a valuation of $3 billion, ranking first among all AI application companies.
There are also a number of domestic manufacturer platforms exploring AI search. In addition to "Zhihu Direct Answer", including Kunlun Wanwei, Baidu, 360, etc., have launched AI-related search products. AI search seems to be on the eve of an explosion.
The scene of the Salt Club New Knowledge Youth Conference
"Follow-up" enables multiple rounds of dialogueAs a content platform closely related to search and answers, search is important to Zhihu more than just capturing industry outlets.
Zhou Yuan explained his understanding of AI search at the New Knowledge Youth Conference, "AI search is a new productivity tool, which can not only help solve problems and improve efficiency, but also help find the principles behind answers. The answer is scattered in countless brains, and in the age of AI, part of it is compressed in an AI large model that needs to be reconnected by AI search tools. ”
What are the benefits of AI search? Zhu Keli, executive director of the China Information Association and founding president of the National Research Institute of New Economy, believes that the most obvious advantage of AI search lies in its application of deep learning and natural language processing technology, which can better understand users' search intentions and provide more accurate and personalized search results.
The ability to integrate "across domains" is another important advantage of AI search. Zhu Keli said that in the era of information explosion, users often need to obtain information from multiple fields to answer a question. By integrating knowledge and information from different fields, AI search can provide users with comprehensive and in-depth search results to meet users' needs for diversified information.
These advantages mean that AI search will play an important role in vertical search, Q&A systems, intelligent assistants, and other fields, providing users with a more intelligent and efficient search experience.
As a question-driven community, Zhihu has a significant demand for AI search functions in the vertical content field and Q&A system.
To this end, Zhihu has been actively investing in AI large models. In March 2023 and March 2024, Zhihu has released the large model of "Zhihaitu AI" and the community-native AI function "Discovery · AI Search", completing two important steps from model development to open product application.
In the statistics of the actual use of "Discovery and AI Search", Zhihu found that users have nearly 3 rounds on average, and how to better link the logical relationship between multiple Q&A and achieve "multi-round dialogue" is the focus of its AI search model capability upgrade. After the upgrade, the "Zhihu Direct Answer" function can more accurately understand the relationship between users' new questions and past questions, and can accurately identify the intention and referential relationship for multiple questions and answers, so as to give more accurate answers.
Upgrade content sources and traceability methodsFrom the perspective of the industry as a whole, due to the recent birth of AI search, maturity and related regulatory issues are still worthy of attention.
The explainability and content credibility of AI search are issues that users are concerned about. 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.
In response to these practical problems, "Zhihu Direct Answer" has been upgraded in terms of content source and traceability methods.
In terms of content, relying on community content to supplement the content of the whole network is an important advantage of "Zhihu Direct Answer". According to reports, "Zhihu Direct Answer" is based on the real Q&A data of Zhihu creators, and endorses the content with Zhihu's excellent answerers; At the same time, a variety of data sources are introduced as auxiliaries to provide more adequate and accessible information channels for searching.
At the same time, "Zhihu Direct Answer" has added the function of tracing the source of content. According to the product owner of "Zhihu Direct Answer", "Zhihu Direct Answer" will not only list the citation sources of the answers, but also trace the source at the paragraph level or even the sentence level, so that you can see in more detail which respondents' contributions are in the answers. Strict traceability can better guarantee the reliability of answers and alleviate the "illusion" problem in the field of AI.
In addition, it can also be found in the actual test experience that "Zhihu Direct Answer" can provide "brief" and "in-depth" answer generation results according to user needs, and supports "finding content" and "finding people", further shortening the distance between content needs and answers, and amplifying the circulation effect of community creators and their content.
It should also be noted that from the perspective of the entire AI search industry, it is still in the early stage of development, and it is necessary to continuously update the technical level of products and upgrade model capabilities. Zhu Keli pointed out that the current AI search may face risks such as design defects and operation errors, and how to ensure the stability and reliability of AI search is a long-term concern for related products.
Liu Chunsheng, an associate professor at the Central University of Finance and Economics, said that with the rapid development of technologies including AI search, how to effectively supervise them and prevent abuse and misuse is a problem to be faced by AI search in the future. At the same time, manufacturers need to take further protection measures against the privacy and security risks that may arise from AI search processing of user data.
(Sino-Singapore Jingwei)