From the initial search through books and materials, to the use of web** for search in the Internet era, and then to the search on social media platforms in the mobile Internet era, and now it has become a new mode that has been hotly discussed - using AI to search, and unconsciously people's search habits have completed many jumps.
Before talking about how AI search will change people's search habits and even living habits, we need to talk about China's first AI search product.
On August 23, 2023, Kunlun Wanwei released it in a low-key mannerTiangong AI, the first AI search in China,Received a lot of praise. Since then, domestic manufacturers of all sizes have followed up to launch AI search, but Tiangong AI has always maintained an absolute advantage in the product experience of various AI search products, perhaps we can find the core value of AI search from the development process of Tiangong AI.
In April 2023, Kunlun Wanwei released the self-developed 100-billion-level large language model Tiangong Large Model; In August, Tiangong AI Search was launched, the first AI search product in China that integrates the technical capabilities of large models; In September, in the authoritative reasoning list Benchmark GSM8K test, Tiangong model stood out with an 80% accuracy rate, and its reasoning ability was close to GPT-4; In the same month, Yan Shuicheng, a world-renowned AI scholar, joined Kunlun Wanwei as the co-CEO of Tiangong Intelligence and the president of the 2050 Global Research Institute. In October, Kunlun Wanwei open-sourced the "Tiangong Skywork-13B" series of tens of billions of large language models.
On February 6 this year, Kunlun Wanwei once again upgraded and released a new version of the MOE large language model Tiangong 20 and Tiangong AI intelligent assistant, which is also the first AI application in China equipped with MOE architecture and free of charge for all C-end users.
It can be seen that Tiangong AI carries Kunlun's iterative innovation and application exploration of AI technology all the wayIt is also an AI search engine that has always adhered to building based on large language models and has been rapidly iterating and self-transcendence. It can even be said that Tiangong AI is the "originator" of domestic AI search.
So why fuse AI large language models? After integrating the capabilities of AI large models, how strong is AI search? From Tiangong AI, in fact, you can also get a glimpse of the answer.
With the deep integration of Tiangong 2Tiangong AI intelligent assistant with 0 large language model capabilities as an example - because it adopts the industry's top MOE expert hybrid model architecture, compared with other AI searches, it will be obvious that it has stronger ability to cope with complex tasks, faster model response, higher training and inference efficiency, and stronger scalability.
In other words, with the AI search supported by large language models, not only the quality of answers and response speed will be greatly improved, but also with excellent visual understanding, reasoning and instruction following capabilities, which can meet the needs of various users such as dialogues and knowledge questions and answers, and can also support ultra-long context windows (more than 150,000 Chinese characters) up to 100k.
For example, when a user searches for "What are the home fitness tools?" "How do you play Blue and White Porcelain on the guitar?" When you ask this kind of question, you can get the best answers, and the results are more intuitive and richer.
For example, it can conduct multiple rounds of in-depth dialogue with users according to contextual semantics, whether it is entering questions, phrases or keywords, it can complete "follow-up", and more than 20 rounds of interaction can be carried out on a question topic, so as to achieve in-depth search, and all answers can be completed in seconds. In the future, in addition to text, users will be able to search for conversations in the form of **, documents, links, voice or **.
It is safe to say that only AI search equipped with large language model technology is the real AI search
From finding content, to getting answers. Traditional search provides users with a large amount of "information", and users need to screen and filter by themselves. AI search with large model capabilities like Tiangong AI can provide generative search, dig deep into the real intent of users, and keenly capture the contextual relationships in query statements, so as to obtain more accurate, more relevant, and more desirable search results, and no longer waste time filtering redundant or invalid information.
From obtaining vague information to acquiring knowledge with precision. In addition to search and dialogue, in different application scenarios such as reading and creation, it can also provide more accurate and specific answers and follow-up suggestions for different needs of users.
From a mixture of fish and dragons to traceability and trustworthiness. Different from traditional search, which often misleads users with misinformation or advertising, AI search using large model technology can identify and screen a large amount of content on the Internet, block false advertising content, and then introduce other influencing factors such as authority and reliability to clean out relatively pure and high-quality search results. Not only can it provide more accurate answers and suggestions for free, but it can also add a source index to all responses to ensure that the answers are traceable, verifiable and reliable.
There is no doubt that after entering the era of AI search, the search experience has been upgraded and reshaped. It is also expected that the products represented by Tiangong AI can bring more commendable and practical application functions.