Recommendation
Explore how AI search technology reshapes the way we obtain information.
Core content:
1. The impact of the evolution of search technology on human cognition and decision-making
2. How deep search technology can achieve more accurate retrieval results
3. Quark AI Superframe's innovative practice in the search field
Yang Fangxian
Founder of 53A/Most Valuable Expert of Tencent Cloud (TVP)
Since the birth of the Internet, the evolution of search technology has profoundly changed the way humans obtain information. Even after the transition from the web to the app, people's reliance on search has never weakened. To some extent, search has profoundly reshaped human cognitive patterns, decision-making methods, and social relationships.When we talk about search, we have to mention recommendation engines. These are exactly the two ways that humans interact with information. Search is initiated by users, while recommendation is passively accepted by users. Search is a one-time interaction, while recommendation is for the system to remember users' long-term interests and preferences. Recommendation engines are popular in the mobile Internet era because they are simpler and more convenient than search.Search is still a one-time way of obtaining information that requires users to make their own judgments in the search results. Problems such as search results being difficult to match needs and cumbersome operations have always existed. Therefore, in the LLM era, the emergence of AI search has become an opportunity for a leap forward in search. On the basis of the original search results, a part of the answers generated by the model capabilities is added. In some search scenarios, the generated answers are placed at the top, followed by the original search results, making the search results more clear at a glance.In the past six months, the relationship between humans and information has been evolving rapidly. According to a search big model expert, "AI engineers have realized that by incorporating long-term thinking and reasoning processes into search systems, more accurate and in-depth retrieval results can be achieved than ever before." Sure enough, DeepSeek-R1 has taken the lead in allowing users to experience how powerful AI that can search online can be without limit.However, AI is not a search engine after all. In order to take LLM a step further in search, Quark, as a disruptor in the search industry, took the lead and launched a new "deep search" product based on the AI superframe. Through deep thinking ability, intelligent retrieval technology and precise answering functions, it can help users solve diverse and complex problems with "high search quotient".As Alibaba's AI flagship application, Quark AI Super Frame also iterates the "intelligent image processing" function, continuously refreshing the user experience and capability boundaries of the AI all-round assistant, and pushing the search experience to a new level.From RAG to DeepSearch, the search experience has changed dramatically
Starting from traditional search engines and entering the realm of deep search, there are two major hurdles to overcome.The first thing to consider is the leap in search technology.Since the rise of Retrieval Enhancement Generation (RAG) technology, using LLM to improve search has become an industry consensus. Integrating search engine results into the content generation process of LLM has become a common practice in the industry, and Perplexity is one of the representatives.The drawback of the RAG system is that it generally only runs the search-generation process once, so the precision and recall of the generated results are limited. This is also the biggest dilemma faced by AI search in the past two years.DeepSearch is a new technology that can be upgraded on the basis of RAG. Its core concept is to continuously cycle through the three links of search, reading and reasoning until the best answer is found. By introducing a multi-step iterative mechanism, the search link uses search engines to explore the Internet, while the reading link focuses on detailed analysis of specific web pages. The reasoning link is responsible for evaluating the current status and deciding whether to break down the original problem into smaller sub-problems or try other search strategies.This cycle can continue until a certain business condition is met, and the best results can be obtained. This condition may be the number of attempts or the token limit. Under this mechanism, the results generated by DeepSearch are far better than those of ordinary AI searches.The second thing to consider is that this looping process takes much longer than traditional search. In the past, search engines relied on a pre-built web index database to quickly return a list of links through keyword matching and sorting algorithms. The entire process is highly optimized and usually takes less than 50 milliseconds. If there is no response within 200 milliseconds, it is a failure.But now users are willing to endure longer processing time for a better result. This is of course due to Deepseek’s popularization of CoT (chain of thought). Deep thinking has become an important reason why people are willing to use AI. Manus projects the code execution process in the virtual machine to the interactive interface in real time, so that users are willing to wait longer, which is also based on the same design concept.With both technology and users fully prepared, the time for commercial search transformation has finally come. AliQuark launched "deep search", which also provided the internal and external conditions to compete with traditional search.What are the characteristics of high emotional intelligence searches?
Quark AI Superframe has launched a new "deep search", the core advantage of which is its "high emotional intelligence". Under in-depth analysis, deep search surpasses deep thinking and conventional search in terms of thinking ability, content correctness and multi-modal capabilities.High emotional intelligence comes from two aspects: one is to correctly understand the user's search intent and understand what the user wants; the other is to be able to generate reasonable and reliable results and meet the user's needs.After receiving the user's question, deep search will deeply analyze the user's real needs, determine the search intent, and then intelligently break down the search task, thinking before searching like humans, and then intelligently summarize the searched content to generate a new answer.For example, I searched with the following prompt words: I have meniscus injury and plantar fasciitis. I want to recommend a pair of women's shoes that can be worn in spring and summer. I want a specific brand and model.Traditional searches will push me several web pages containing "meniscus injury, plantar fasciitis, women's shoes" under keyword matching. The content of these web pages may or may not meet my needs because these web pages are not customized based on my needs.In Quark's deep search, the first step is to use analysis, which can deeply analyze the complex questions raised by users, analyze the questions step by step and think thoroughly, so as to extract key information and true intentions. In this case, Quark's demand decomposition and deep thinking process is as follows:The deep search thinking process speaks out the needs of users that have not been expressed in the prompt words, including shoe types, medical advice, breathability of shoes in spring and summer, etc. It is no exaggeration to say that it is a caring little assistant.Secondly, "deep search" adopts a new approach of "think first, then search". It searches and matches high-quality information sources across the entire network, carefully reads the core information in hundreds of pages, and calls different agents according to the situation, making the search results more in-depth and comprehensive.Ultimately, "deep search" can deeply integrate key content and cross-domain knowledge to generate an accurate, reliable detailed explanation and solution. Especially in the fields of daily life, travel planning, medical health, learning and education, it can further help users solve practical problems and significantly reduce user decision-making costs.Based on 33 web pages with high-credibility sources, Quark generated an answer:This high EQ answer is particularly evident in three types of questions. The first type is personalized questions, for which there is no corresponding answer on the entire Internet. In regular searches, users can only refer to the search results to summarize. For example, when giving gifts on Mother's Day, we can ask Quark what gift to give based on the mother's usual preferences :My mother likes to dance in the square and plant and appreciate flowers. I have given her jewelry, bags and clothes before. Last year, I gave her a set of Armani cosmetics. She liked it very much. What gift should I give her for Mother's Day this year? The budget is within 2,000 yuan.
The second category is complex problems that require content support from different dimensions.
My child is in the second grade of primary school. The teacher assigned a handwork assignment. The theme is to protect nature. It can be in the form of painting, hand-written newspapers, origami, model making, etc. Help me think of the easiest way to do it with the least tools, and the content should be creative. Tell me in detail how to do this handwork assignment?
This problem may seem simple, but it is troublesome to actually do it . We need AI to not only give suggestions, but also provide detailed instructions, and even provide pictures and videos to help us complete such tasks. Having multi-modal processing capabilities and solving various complex problems is exactly what Quark Deep Search is best at.
The third type is the problem that the text description is not clear and requires fuzzy search. This kind of problem cannot provide much information, but you hope that the search engine can give a definite answer, and only deep search can resolve it.I forgot the name of the novel, but it's about the protagonist having to keep gambling his life with the zodiac in order to return to the real world. It's an infinite flow. Which novel is it?
These things that cannot be done or done well in traditional searches can now be easily solved by Quark Search with the technical empowerment of Alibaba Group's self-developed big model.
When a user initiates a search request, the system not only provides accurate initial results, but also continuously identifies and verifies intent during the interaction process - analyzing the user's potential needs, adjusting the search strategy based on real-time feedback, and intelligently determining whether it is necessary to expand the search dimension or deepen the vertical field.Not only are the answers traceable, but they also perform well in professional scenarios such as healthcare and academic research. Deep search has an intelligent closed loop of "search-verify-re-search". It not only improves the efficiency of information acquisition by more than 40%, but also significantly reduces information bias through multi-dimensional cross-validation to ensure the accuracy of answers.The relationship between humans and information is changing. The trouble and inconvenience of users actively searching for information online will be solved in the era of deep search. People who are good at using deep search will no longer be restricted by the information cocoon brought by the information flow of recommendation engines.Quark's first step towards becoming a super agent
In March this year, after the release of Quark's "AI Super Frame", it caused a huge response on the C-end user side. This is the first flagship AI super portal publicly announced at the strategic level by a major Internet company. Industry observers generally believe that this is Alibaba's key move in the consumer AI battlefield.After visiting many AI product professionals, Leifeng.com found that people have very different views on the existence of universal agents. Some people believe that future agents must be universal, while others are relatively pessimistic and always believe that the complexity of real-world scenarios determines that universal agents are difficult to achieve. However, everyone agrees that information retrieval, processing and planning are always the first step before agent execution.Compared with the "Tongyi Qianwen" large model and DingTalk's intelligent transformation that focus on enterprise services, Quark's "AI Super Frame" encapsulates the cutting-edge technologies of Alibaba Cloud Intelligence and DAMO Academy into a disruptive experience that can be perceived by C-end users.In the search scenario, Quark continues to innovate to create a new user experience. Deep search is just one step. Half a month ago, Quark AI Super Frame released a new AI camera and a new "Take a picture and ask Quark" function. Just take a picture and upload it, and Quark will understand the picture and answer questions in the real world. This is also an important measure to expand the search scenario.In building a closed-loop agent ecosystem, Quark has integrated over a hundred vertical field agents, covering scenarios such as scanning, learning, medical treatment, and creation, forming a full-link service closed loop.Quark's deep search function continues to deepen its intelligent upgrades. Relying on Alibaba's self-developed reasoning model and multi-agent collaboration mechanism, it can accurately break down complex needs and call exclusive agents among hundreds of vertical field agents to complete tasks.We learned from insiders that Quark will continue to launch the Deep Search PRO function based on Deep Search, which has more professional analysis and reasoning capabilities, can compress complex problems and tasks that take several days to complete into minutes, and deliver structured and systematic professional results. From search, analysis, decision-making to execution, a chain is being opened up, and a super agent is about to be born.Quark is redefining the value chain of search services - in the process of evolving towards a super agent, the subversion of search has become an important step in the form of new products. The launch of deep search is not only a key step in exploring universal agents, but also the first to open the door to a new era of search.
