Wen Xinyiyan officially opens internal invitations! Robin Li: The experience is not perfect!

​Author | Xu Jiecheng

Reviewer | Yun Zhao

The early release of GPT4 really puts the pressure on Baidu! A few months after ChatGPT detonated the field of chatbots, we finally ushered in the first generative dialogue product in China.

At 14:00 on March 16, Baidu Wenxinyiyan press conference was officially held at Baidu’s Beijing headquarters. Robin Li, founder, chairman and CEO of Baidu, has also fulfilled his previous promise and turned cool technology into a product that everyone needs.

It is reported that Wenxin Yiyan (English name: ERNIE Bot) is a generative dialogue product launched by Baidu based on Wenxin large model technology, which can achieve natural and smooth human-computer interaction. . As one of Baidu's core technologies, Wenxin large model ERNIE has cross-modal and cross-language deep semantic understanding and generation capabilities. Wenxinyiyan, based on ERNIE technology, has successfully integrated artificial intelligence technology with human language communication capabilities, and can provide users with a high-quality, intelligent interactive experience.

1. Robin demonstrated Wen Xin Yi Yan's continuation of the Three-Body Problem, Luoyang Zhigui Poetry, and the calculation of chickens and rabbits in the same cage.

Currently, large-scale language models and generative AI are representatives of It has created a new technological paradigm and is an opportunity that every company in the world cannot miss. Baidu Wenxinyiyan is positioned as an empowering platform for artificial intelligence base models, and will assist the intelligent transformation of thousands of industries such as finance, energy, media, and government affairs.

At the press conference, Robin Li demonstrated Wen Xinyiyan’s five core capabilities and their performance in five usage scenarios, including literary creation, business copywriting creation, mathematical calculation, Chinese understanding and multimodal generation.

In the literary creation scene, Wen Xinyiyan summarized the core content of the well-known science fiction novel "The Three-Body Problem" based on dialogue questions and answers, and proposed multiple sequels to "The Three-Body Problem" 》Suggestions and ideas. In addition, Wen Xinyiyan also accurately answered the "Three-Body" author, "Three-Body" TV series character actors and other factual information based on the content of the question.

The main drawback of current generative AI is that it is difficult to guarantee the accuracy of answering questions, and Wen Xinyiyan continues It adopts Baidu's knowledge-enhanced large model concept and greatly improves the accuracy of factual questions. Faced with questions such as "What do Yu Hewei and Zhang Luyi have in common?" and "Who is taller, Yu Hewei or Zhang Luyi?" Wen Xinyiyan also relied on his information retrieval ability and reasoning ability to get the correct answer.

In the business copywriting scene, Wen Xinyiyan successfully completed naming the company, writing the company’s slogan, Writing company establishment press releases and other creative tasks. In three consecutive content creations, Wen Xinyiyan not only accurately understood user intentions, but also completed clear expressions of user feedback.

This is the "intelligence emergence" that occurs based on the huge scale of data. According to reports, the training data of the Wenxin Yiyan large model includes trillions of web page data, billions of search data and image data, tens of billions of daily voice call data, and a knowledge graph of 550 billion facts, etc., which makes it in In terms of Chinese language understanding and processing, it is better than almost all other large models in the world.

In terms of mathematical calculations, Wen Xinyiyan relied on his thinking ability, mathematical deduction ability and logical reasoning ability to successfully complete Answers to classic examples that exercise human logical thinking such as "chicken and rabbit in the same cage".

In this link, Wen Xinyiyan not only identified the problems in the question, but also successfully found the solution to the problem like a human being, and followed the correct steps step by step Deduced the accurate answer to the question.

## If literary creation, business copywriting creation, and mathematical calculation are common advantages and capabilities of large language models. Then the uniqueness of Wenxinyiyan is its better Chinese understanding and multi-modal generation capabilities.

As a large language model rooted in the Chinese market, Wenxinyiyan has the most advanced natural language processing capabilities in the Chinese field and has better performance in Chinese language and Chinese culture. In the fourth round of demonstrations, Wen Xinyiyan correctly explained the meaning of the idiom "Luoyang paper is expensive", combined with the connotation of the idiom to analyze the corresponding economic theory of "Luoyang paper is expensive", and even successfully created a poem using the four words "Luoyang paper is expensive" The first acrostic poem.

In terms of multi-modal generation, Robin Li demonstrated Wen Xinyiyan’s generation of text, pictures, audio and Video capabilities. Interestingly, Wenxinyiyan can even generate speech in dialects such as Sichuan dialect; Wenxinyiyan’s video generation capability is not currently open to all users due to its high cost, and will be gradually accessed in the future.

2. Technical interpretation of Wen Xinyiyan

Wen Xinyiyan, as a new generation of knowledge-enhanced large language model, is developed on the basis of the ERNIE and PLATO series models. Its key technologies include supervised fine-tuning, reinforcement learning with human feedback, prompts, knowledge enhancement, retrieval enhancement and dialogue enhancement.

The first three are technologies used by such large language models. They have also been applied and accumulated in ERNIE and PLATO, and have been further strengthened and polished in Wen Xinyiyan. ; The last three items are re-innovations of Baidu’s existing technological advantages, and are also the basis for Wen Xinyiyan’s growing strength in the future.

Wen Xinyiyan’s knowledge enhancement mainly involves the internalization of knowledge and the external application of knowledge. Knowledge internalization is based on semantic unit learning from large-scale knowledge and unlabeled data, using knowledge to construct training data, and learning knowledge into model parameters. Knowledge external application is to introduce external multi-source heterogeneous knowledge to perform knowledge reasoning, prompt construction, etc.

Wen Xinyiyan’s search enhancement comes from a new generation of search architecture with semantic understanding and semantic matching as core technologies. . By introducing search results, it can provide timely and accurate reference information for large models to better meet user needs.

Wen Xinyiyan’s dialogue enhancement is based on the accumulation of dialogue technology and applications. Wenxinyiyan has Memory mechanism, context understanding and dialogue planning capabilities to achieve better dialogue coherence, rationality and logic.

##3. Three major industry opportunities generated by generative AI

Robin Li mentioned that today’s AI technology has developed to a critical point, and all walks of life will inevitably be changed. China's AI market is about to experience explosive demand growth, and the release of its commercial value will be unprecedented and exponential.

Since Baidu officially announced “Wen Xin Yi Yan” in February, more than 650 companies have announced their access to the Wen Xin Yi Yan ecosystem. This means that many companies have understood that Wenxinyiyan and generative AI represent a new technology paradigm that will affect every company. According to Robin Li’s prediction, large language models will bring three major industry opportunities.

The first category is the new cloud computing company, whose mainstream business model will change from IaaS to MaaS. Wen Xin's words will fundamentally change the rules of the game in the cloud computing industry. In the past, enterprises chose cloud vendors based more on basic cloud services such as computing power and storage. In the future, more emphasis will be placed on whether the framework is good, whether the model is good, and the collaboration between the four layers of model, framework, chip, and application.

The second category is companies that fine-tune industry models. This is the middle layer between the general large model and enterprises. They can call the general large model capabilities based on insights into the industry. , providing solutions to industry customers. In this regard, Baidu Wenxin Model has released more than 10 industry models in electric power, finance, media and other fields.

The third category is companies that develop applications based on large model bases, that is, application service providers. Robin Li asserted that for most entrepreneurs and companies, the real opportunity is not to build basic large-scale models like ChatGPT and Wenxinyiyan from scratch. This is very unrealistic and uneconomical. This may be the real opportunity to preemptively develop important application services based on a general large language model. At present, based on text generation, image generation, audio generation, video generation, digital people, 3D and other scenarios, many entrepreneurial star companies have emerged, which may be new giants in the future.

Wen Xinyiyan will provide external services through Baidu Intelligent Cloud to help enterprises build their own models and applications in key areas such as agriculture, industry, finance, education, medical care, transportation, and energy. , the efficiency will be greatly improved, and new industrial space will be quickly formed in every industry to help realize Digital China. Robin Li predicted that Baidu Smart Cloud will hold a press conference in the near future, with the theme centered on Wen Xinyiyan’s cloud services and application products, which include both public cloud services and privatized deployment.

At the end of the press conference, Wang Haifeng said that Wen Xinyiyan is currently open for internal testing. From now on, the first batch of users can experience Wenxinyiyan products on the official website by inviting test codes. Baidu Smart Cloud will soon open Wenxinyiyan API interface calling services to corporate customers. You can apply to join the Wenxinyiyan cloud service test on the official website.

4. Wen Xin said that it is not perfect

Robin also admitted frankly that this release At that time, Wen Xin said that he was not completely ready.

Recently, many friends have asked me, "Why today? Are you really ready?" In fact, in a sense, we have been preparing for this for many years. We started investing in AI research more than ten years ago and launched the Wenxin Big Language Model in 2019. Today's Wenxinyiyan is a continuation of our efforts over the past many years. But we can’t say that we are completely ready. Wen Xinyiyan’s benchmark against ChatGPT and even against GPT-4 is very high. No major global manufacturer has yet made it, and Baidu is the first. From my own testing, I feel there are still many imperfections. Individual choice.

5. Write at the end

After watching this press conference, I believe everyone has My own different feelings. In the final analysis, for users, especially in the field of cultural content, we still need domestic technology companies to actually use massive Chinese corpus to train some localized high-quality large models that are more suitable for the Chinese environment.

Now, as the pioneer of domestic AIGC, Baidu has taken the first step in this field. Can Baidu go further and further on this path in the future? Will more outstanding domestic technology companies set foot on this soil? What changes will China's AIGC ecology undergo under the accumulation of these pioneers? let us wait and see. ​

The above is the detailed content of Wen Xinyiyan officially opens internal invitations! Robin Li: The experience is not perfect!. For more information, please follow other related articles on the PHP Chinese website!