news

on september 4, wenxinyiyan app upgraded its brand to wenxiaoyan and simultaneously launched a series of new ai functions.

baidu introduced that wenxiaoyan has launched "new search" capabilities such as rich media search, multi-modal input, text and image creation, and highly realistic digital people, which can meet users' search, creation, and chat needs. at the same time, wenxiaoyan released new features such as memory and free subscription. in addition, wenxiaoyan will also open the wenxin 4.0 large model capabilities for free from now on, and all users can use it for free throughout september.

regarding the product innovation from "wenxinyiyan" to "wenxiaoyan", xue su, vice president of baidu and head of ai product innovation business and meg strategic operations center, recently said in an interview with the media, including the reporter of "daily economic news", that there is still a lot of room and possibility for ai in the entire to c field. to be honest, baidu is still exploring, because to c has many possible directions and uncertainties, "wenxiaoyan is our phased achievement, and it will continue to get better."

baidu innovates search

"the more important the decision, the more likely it is that there are both risks and opportunities. particularly obvious decisions are often not decisions. there is often a balance in making a decision, which depends on what we care about most." xue su said when asked about the considerations behind wen xiaoyan's name change and upgrade.

xue su believes that the first anniversary of the launch of wenxin yiyan is a turning point for the product to mature. secondly, wenxin yiyan has achieved good results in the tob market and technology. as the product enters the "coming of age ceremony", the company hopes that it can become a to c product for a wider range of people, and the characteristic of such products is affinity.

another reason is that wen xiaoyan will be positioned as a new search intelligent assistant. xue su said that traditional search products have strong tool attributes, while the new search intelligent assistant is a shift from tool-based to human-based. he believes that future ai products, especially generative ai products, will demonstrate capabilities that tool-based products did not have in the past - more humane. although it has not been fully achieved yet, this reflects the team's goals and vision. "wen xin yi yan has become wen xiao yan. we hope that it can enter the daily life of users, not only as a tool, but also as a close friend, as well as a think tank and confidant who accompanies users."

xue su positioned wen xiaoyan as "an explorer in baidu's ecosystem". he said that baidu has a lot of ai capabilities, which are the "big brother" and the main force of baidu's new search camp. wen xiaoyan is the "vanguard" in baidu's new search camp and the commando of ai-native. in his opinion, wen xiaoyan is a more extreme product than ai-native; it is also an explorer of more new capabilities and new technologies; at the same time, wen xiaoyan's application scenarios are more in line with mobile scenarios. in terms of ecosystem construction and technology exploration, wen xiaoyan will go to the extreme.

talking about his understanding of new search, xue su said that for baidu, new search must include old search, and "search", "create" and "chat" all rely heavily on the capabilities baidu has accumulated in the past in the search field, such as search enhancement and other business capabilities. "we can do better by standing on the shoulders of giants. the two are not contradictory, but go hand in hand and promote each other."

xue su believes that from the product dimension, past search products were mainly based on “search”, including many ai search tools on the market: users initiate requests and the system provides feedback. this is a very obvious feature of tool products.

however, the new search should be a humanized product, and the scenarios it covers are not limited to search, but also include "creation", which is an extremely powerful function in new search products. it can generate unprecedented information in the past, even information that users themselves have never expected. through multiple rounds of interaction with users, based on the understanding of users, it gradually optimizes and generates a result that meets the personalized needs of users; another scenario is "chatting". sometimes when users communicate with generative ai or new search products, it is not always for a clear purpose. in addition to "searching" and "creating", it may involve emotional communication or providing companionship. therefore, compared with traditional passive tool products, products that complete the "search", "create" and "chat" scenarios are richer, more active and personalized.

in terms of capabilities, xue su thinks that previous product capabilities were passive, while the current scenarios are more inclined to initiative and personalization. the realization of initiative first requires personalization as a foundation, which requires the product to have memory and personalized processing capabilities, and a lower communication threshold. in the past, when faced with tool products, users realized functions by inputting commands; now, facing a wider user group, products need to show more humane characteristics, and the interface design should be easier and easier to use.

from passive to active, from universal to personalized, from high-threshold voice and text communication to low-threshold and more attractive communication, these are all changes from instrumentality to humanity. combined with the changes in the scene just now, we have moved from the purely instrumental "search" to the more humane "search", "create" and "chat".

from instrumentality to humanity

xue su observed that many early products neglected to understand user value and even the degree of user awareness of ai. for example, ai has overcome the limitations faced by many ai products six months or a year ago, such as the inability to perform tasks other than language conversation. no matter what the user's needs are, technology is trying to meet them. this is technology catching up with human needs. many people may no longer be clear about how far technology has developed and what functions it can achieve.

this made him start thinking about how to better help users become familiar with these features through product design, concepts, and the development of capabilities. only by deeply understanding users can the product be made better. for example, in some cases, ai's capabilities may not be satisfactory, but sometimes they may exceed expectations. taking search as an example, long-term search habits may make users want to make questions as precise as possible, while ai assistants are good at dealing with fuzzy and complex logical reasoning. therefore, where its capabilities are bounded and how to form the best experience still requires continuous exploration and experience.

after the big model craze came, robin li said many times that as a basic foundation, big models can support the development of countless ai native applications. just like the mobile era gave birth to "mobile-native" applications such as wechat, tiktok, and uber, in the ai ​​native era, there will definitely be excellent ai native applications developed based on these big models.

"the difficulty of making ai-native products lies in its fragmentation," xue su said. the fragmentation lies in the fact that people who understand products find it difficult to touch the technical boundaries, while people who understand ai may not be able to gain insight into the specific needs of users from the evolution of large models. it is very difficult to effectively combine these two parts. however, truly good toc products in the future must come from teams that understand both.

what is a truly good toc product? xue su thinks that it is the next step of tool products, that is, a product with "humanity".

in xue su's view, the so-called "humanity" is not simply copying real humans, but creating a "more powerful person": not only must he have the compassion and emotional intelligence of ordinary people, but also must have intellectual abilities that humans do not possess.

to create such products, the paths are also different. for example, the way to create traditional tool products is to clearly assign tasks and then obtain results. the calculation process is completed through formulas. this is a tool product. however, when taking a step forward and turning to more humanized products, if the method of assigning tasks and letting them complete them is used, the products are also not perfect because they only provide one answer. this is similar to handing a task to a colleague, and if the other party only provides an answer, this is feasible, but not completely satisfactory.

xue su said that what he really wants to do is to make an ai assistant that does not provide one result at a time, but 20 different results at a time, using humanized means to complete superhuman interaction. he believes that this is the essence of an ai assistant. in many cases, the ability of ai is not limited to the results it outputs. although the results themselves are important, as long as the information it provides has a certain amount of information, it can also inspire users to think. "the ability of ai is not just the answer machine of the past, which is very traditional. the current ai's reasoning, understanding, and generative capabilities have many links and steps that can be mined and utilized."

ai products, although called ai-native products, are no longer solely driven by technology. in the early stages of product development, people may notice that the product design is so complicated that once general artificial intelligence (agi) makes a little progress, all previous efforts seem to be covered. xue su feels that the marginal value of products is limited now, but there must be some crucial product cognition and insights that can be combined with technology, and these insights still come from a common understanding of technology and user needs.

(cover image source: daily economic news file photo)

daily economic news