Shulou(Shulou.com)11/24 Report--

Thanks to CTOnews.com netizen AINSOFT for the clue delivery! CTOnews.com March 16 news, Baidu held a press conference at its headquarters in Beijing this afternoon. At the press conference, Baidu officially launched the big language model Wen Xin Yi Yan, and demonstrated the comprehensive ability of Wen Xin Yi Yan in a number of scenarios.

From the point of view of the on-site display, Wenxin Yiyan has the ability to understand human intention to some extent. However, Li Yanhong also mentioned many times that this kind of big language model is far from the stage of development and there is a lot of room for improvement.

Baidu also announced the invitation test scheme of Wen Xin. Starting from March 16, the first batch of users will be able to experience the product on Wenxin Yiyan's official website through the invitation test code, and will be opened to more users one after another. In addition, Baidu Intelligent Cloud will soon open the API interface to enterprise customers to invoke services. Officially open the appointment from March 16, search for "Baidu Intelligent Cloud" to enter the official website, you can apply to join the Wenxin one-word cloud service test.

At the press conference, Robin Li, founder, chairman and CEO of Baidu, demonstrated the performance of Wen Xin in five scenarios, including literary creation, commercial copywriting, mathematical calculation, Chinese understanding and multimodal generation.

In the scene of literary creation, Wen Xin summarizes the core contents of the well-known science fiction "three-body" according to the dialogue, and puts forward five suggestions for the continuation of the "three-body". It reflects the comprehensive ability of dialogue question and answer, summary and analysis, and content creation. In addition, Wen Xin accurately answers the factual questions such as the author of "three-body" and the role player of TV series. Faced with such questions as "what do Yu Hewei and Zhang Luyi have in common" and "who is taller between Yu Hewei and Zhang Luyi", Wen Xin also got the answer based on his reasoning ability.

In the commercial copy creation scene, Wen Xin successfully completed the creative task of naming the company, writing Slogan and writing press releases. CTOnews.com learned from Baidu that the training data of Wenxin Yiyan model includes trillion-level web page data, billions of search data and picture data, 10 billion-level daily voice call data, and 550 billion factual knowledge graph.

According to Baidu, Wen Xin Yiyan also has a certain thinking ability, can learn mathematical deduction and logical reasoning and other relatively complex tasks. In the face of classic problems such as "chickens and rabbits in a cage", which exercise human logical thinking, Wen Xin can understand the meaning of the topic and have correct ideas for solving the problem, and then, like students doing the problem, work out the correct answer step by step.

On this basis, Wenxin Yiyan also shows the ability of Chinese understanding and multimodal generation. In the on-the-spot display, Wen Xin correctly explained the meaning of the idiom "Luoyang Paper Gui" and the corresponding economic theory of "Luoyang Paper Gui", and created a Tibetan head poem with the word "Luoyang Paper Gui".

In terms of multimodal generation, Robin Li demonstrated the ability of Wen Xin to generate text, pictures, audio and video. It is worth mentioning that Wenxin word can even generate dialect pronunciation such as Sichuan dialect; the video generation ability of Wenxin word is not open to all users at this stage because of its high cost, and will be gradually connected in the future.

Data show that since Baidu announced "Wenxin Yiyan" in February, more than 650 enterprises have announced access to Wenxin Yiyan ecology. Wang Haifeng, Baidu's chief technology officer, said that Wen Xin's words are a new generation of knowledge-enhanced big language models, which are developed on the basis of ERNIE and PLATO models. Its key technologies include supervised fine tuning, reinforcement learning with human feedback, hints, knowledge enhancement, retrieval enhancement and dialogue enhancement. The first three items are the technologies that will be used in this kind of large language model, and they have also been applied and accumulated in ERNIE and PLATO, and further strengthened and polished in Wen Xin; the latter three are the re-innovation of Baidu's existing technological advantages.

Welcome to subscribe "Shulou Technology Information " to get latest news, interesting things and hot topics in the IT industry, and controls the hottest and latest Internet news, technology news and IT industry trends.