Tsinghua Xueba competes against Baidu AI partners, bluntly saying that the college entrance examination may fail AI again

On June 7, the first day of the college entrance examination, the news of AI challenging the Chinese language of the college entrance examination swiped the screen.

In a live broadcast about the college entrance examination, Baidu Search's "AI partner" who is undergoing internal testing challenges the Chinese exam for the college entrance examination on the spot, including composition, micro-writing, appreciation and analysis of ancient poems, classical Chinese translation, reading and filling in the blanks and other question types. A composition in classical Chinese even confused Zhuang Linxu, a senior Chinese teacher, who mistakenly thought it was written by a real person, and gave a high evaluation, thinking that the level of the students exceeded 95%.

"Bi Dao", a popular science student of Tsinghua University who participated in the live broadcast connection, also said bluntly: "I may not pass it in the Chinese college entrance examination again." He also sighed, "The era of per capita academic masters is coming."

03428dda931e66ada30f126497697328.jpeg

This AI partner is an assistant brought by Du Xiaoxiao, an AI digital person who successfully got out of the circle last year because of writing Chinese essays for the college entrance examination. Last year, Du Xiaoxiao answered an argumentative essay titled "Own Hands, Skilled Hands, and Common Hands" in Volume I of the National New College Entrance Examination, and scored a high score of 48, beating over 75% of the candidates. Swipe the screen.

In the AI ​​era, mastering prompt words is a core competency. No matter what question is input, the large model can always give the answer, but to make it generate high-quality content, the user needs to input several high-quality prompt words. For the same large model, giving different prompt words will generate different results. It can be said that whoever has the engineering ability of prompt words will have the key to unleash the powerful production capacity of large models. Robin Li, founder, chairman and CEO of Baidu, once boldly predicted in a speech: 10 years later, 50% of the jobs in the world will be prompt word projects.

In this online live broadcast of the college entrance examination, Xiao Yang, the vice president of Baidu Group and the person in charge of the search platform, taught the audience how to use prompt words correctly so that AI can produce better results. It can be summarized into a formula: prompt word = task + Character + contextual background + detail supplement. For example, using "AI partner" to read and comprehend Du Fu's "Sick Cypress" ancient poems, without optimizing the prompts, the answers obtained are relatively comprehensive and satisfactory; but after the prompts are optimized, the quality of the answers is significantly improved.

def03c07cdac293c993bbf3d9cf32331.jpeg

In this regard, Xiao Yang said, "In the future, we will not only measure IQ and EQ, but also 'search quotient'. Just like in the Internet age, people who can use search better will have greater and greater advantages, and the gap will widen. AI partners are part of the intergenerational change in search. Better use of prompt words and better use of AI partners will gain more and more advantages. Prompt word engineering should be a compulsory course for everyone in the AI ​​era. Use large models to Create more, get more."

At the 2023 Vientiane·Baidu Mobile Ecosystem Conference held at the end of May this year, Xiao Yang demonstrated the ability of "AI partner". This all-round "Al partner" can understand any questions of users and interact with voice. It can also help users mark the key points in the answers, provide authoritative sources, and support calling various tools and services, as well as according to user intentions. Make pictures or create copywriting, like a caring partner who is always with the user, allowing users to talk to the future anytime, anywhere.

AI partner is the crystallization of using Baidu semantic retrieval technology + large language model, and this is of epoch-making significance. Xiao Yang said at the Vientiane Conference that the combination of semantic retrieval technology and large language models can further release the potential of search understanding, reasoning, organization and creation, which is expected to lead to subversive innovation in many aspects and the advent of the search singularity.

And this is due to Baidu's ten-year dedication to artificial intelligence. Baidu is one of the few artificial intelligence companies in the world that has a full-stack layout of artificial intelligence four-layer architecture. It has key self-developed technologies and products at all levels. The chip layer has high-end chip Kunlun chips, the framework layer has the deep learning framework of the flying paddle, and the model layer There is Wenxin pre-trained large model, and on the application layer, Baidu search is also the first in the industry to fire the "first shot" of large model application.

It is reported that if you search for "gaokao" in Baidu, you can enter the special page of Baidu's college entrance examination. Baidu takes advantage of its platform and technology to provide various information and services for the college entrance examination. Search for "gaokao" in the Baidu APP, and you can experience the one-stop AI test assistance service. . The services include but are not limited to test preparation services, online score checking, voluntary reporting, admission inquiries, etc., covering the three major stages of pre-test, mid-test, and post-test. In addition, Baidu has comprehensively upgraded the search experience of the college entrance examination based on AI technology this year, launched the college entrance examination search index for the first time, and also upgraded services such as AI volunteer assistants to help the college entrance examination throughout the process.

Guess you like

Origin blog.csdn.net/ZabeNbRdit36243qNJX1/article/details/131136603