Search engines may really be changing!
Google is finally about to usher in its biggest change in 25 years. Google announced the opening of the internal beta [Search Generative Experience (SGE)], and gradually abandoned those even the notorious ten blue links (10 Blue Links ).
In 1998, Larry Page and Sergey Brin launched the Google online search engine with the PageRank algorithm. It initially looked like this:
And Google is also almost all the search engines that we are familiar with and understand now. After entering the question, it will use PageRank or other more complex search recommendation algorithms to sort out the most relevant web pages for us as the answer. The forms of "blue links" are listed as follows. We read the titles of these web pages, browse the summary below, and use our brains to judge whether clicking on a certain blue link will give us the answer we expect
For 25 years, such search engines have shaped and even defined our online life. In the past 25 years, we have dealt with this small search box and those blue links that are true or false. Clicking, staying, sliding the mouse, locating key information, or extracting or collecting , search engines, as a different medium, have greatly extended the boundaries of our cognitive world. 25 years have made us familiar with search engines As if in my own body, there is a kind of certainty in my heart that after entering a certain string of characters, these neatly arranged links will appear in front of my eyes
Therefore, in December last year, shortly after ChatGPT debuted, when we mentioned in the article that ChatGPT may subvert all traditional search engines, we also thought that these large models revolutionizing traditional search engines were just an exaggerated rhetorical technique , everyone in the comment area also thought that talking about "revolution", "replacement", or even "threat" at that time was just a fantasy , although no one denied that this technology brought us a deeper understanding of the "essence of search". Thinking about this question and the almost unlimited imagination behind it, it forces us to gradually realize the question " the answer to the search is not necessarily just those ten blue links "
If time is fast forwarded, in just half a year, we have experienced a hurricane movement based on large-scale language models (LLMs) with you. New technologies, new applications, new problems, and new experiments have sprung up. When reporting [Zhang Junlin: Will ChatGPT become the next generation of search engine] from a technical perspective, we believed that the new generation of search engines would appear in the form of "user intelligent assistants", which cannot be realized through traditional search engines , its core must be based on the large model represented by ChatGPT , although we believe that there are still three unresolved problems between ChatGPT and the real search engine at that time, namely the authenticity problem caused by machine illusion, the knowledge update problem of the model and the cost issue .
After more than six months of technological blowouts, now, at the moment, Google is finally about to usher in its biggest change in 25 years. Google announced the opening of internal testing [Search Generative Experience (SGE)], And gradually abandon those even the notorious 10 Blue Links (10 Blue Links) . Traditionally, the whole set of search processes that we are familiar with for 25 years will be innovated. In the future, when we enter questions, such as if I asked, "If I have three children and a pet dog, which one is more suitable for our family trip, Bryce Canyon or Arches National Park?", the answer will no longer be a vague "Compare Bryce Canyon and Arches National Parks” travel notes link, instead telling us directly that “While both parks prohibit dogs on unpaved trails, Bryce Canyon has two paved trails that allow dogs .”
At the same time, what SGE has changed is not just simple search, e-commerce and advertising based on search will usher in a revolution , as shown in the figure below, when searching for products, we no longer need to face those " xx Folding Bicycle xx Inch Non-Speed Coupling Adult Student Children City Commuter Bike White Red" symbols piled up, but you can directly get product pictures, product descriptions, latest reviews, ratings, prices and other comparisons of related products.
It is no exaggeration to say that SGE can completely change the information integration method of search engines in the past. It is not just a rearrangement and combination of a display interface, but the key process of the entire "search" has changed, and the answer will be in a more direct way. Presented in front of users, we no longer need to shop around and jump back and forth between multiple links, and we don’t need to spend effort to judge the authenticity of the preset URL behind the link, let alone when faced with some personalized problems. Combining information between different pages to complete careful inference, the search will truly become a question-and-answer process, and the answer will be directly presented to us .
But going back to the three problems we mentioned in our report last December, the authenticity problem caused by machine hallucinations, the knowledge update problem of the model, and the cost problem, aside from the cost problem, it is clear that the future of SGE still faces " How does this AI-generated summary ensure authenticity" and "model update" issues . In terms of model update, we have also reported articles on "ChatGPT Networking" and "Seal Release". For example, we reported Google's Bard Almost has near real-time timeliness. At the same time, there are also discussions on the issue of model updating in the academic circles, such as [What should I do if the knowledge of the large model is Out? The Zhejiang University team explored a method for updating large model parameters—model editing], and Zhejiang University proposed that the method of model editing may become one of the future solutions.
As for all the questions that may be caused by the machine hallucinations that may cause the wrong answer of the large model, it may be a more critical question. There have been user responses, such as asking the large model "When will the new Zelda be released?" The answer is 2017 March, which is the release date of Breath of the Wild, not May 12, 2023, when Tears of the Kingdom will be released. For this problem, we once conceived a two-step process of large-scale model search engineization, that is, from the traditional search engine as the main model to the traditional search engine as the main model. Undoubtedly, when Google's SGE is officially opened to the public At that time, the search engineization of large models has really reached a new stage where large models are the mainstay and traditional search engines are supplemented, that is, large models are used as the main provider of search answers, and traditional search engines are used as an auxiliary to verify the authenticity of answers. tool , in SGE, you can see a set of cards on the right side of the answer provided by the big model, and these cards correspond to links as evidence of the answer to the big model.
Going back to our beginning, the search engine has changed. Yes, if we in December last year could not believe that the search paradigm that has accompanied us for 25 years will be replaced by a large model in the form of direct question and answer, then today with the SGE We can no longer deny that the new era of search engines has quietly and suddenly arrived . From the blue link to the current personal assistant-like answer, the arms race of these big companies and big models has also given birth to a huge AI application in the past 25 years. At the climax, it can be seen that the old problems have not been completely solved, and new problems are still being raised. Can SGE adapt to those blue links cultivated by Google search in 1998 in terms of search authenticity, privacy protection, model update, etc. According to the needs of discerning users, whether SGE can truly perfectly replace traditional search engines and become a truly ideal personal intelligent assistant, let us wait and see, the future will come soon!
SGE experience
At present, the SGE function needs to apply to join the Waitlist of Search Labs first, and then you can try and experience the SGE function after queuing to obtain the qualification. At present, the SGE function does not support Chinese, and can only be experienced in English.
Application address:
https://labs.google.com/search/login
