Give AI a month, what can it evolve into?

When Wen Xin said "full moon", Baidu held a technical exchange meeting and presented the following report card:

After completing 4 iterations, the cost of large model inference is reduced to one-tenth of the original cost.

Of course, this is not intuitive enough. According to the latest live demonstration, the function of Wenxin Yiyan is far more than a conversation and chat, but has really become a productivity tool .

Moreover, the requirements are entered on the spot and the results are produced in real time.

The onlookers were full of participants from all walks of life, and there were not enough chairs.

PPT is an important means of communication in the workplace. In the past, many people racked their brains and lost their hair for making PPT.

Now he can directly pass on the tasks assigned by the boss to Wen Xin, and the results will be announced directly.

The format layout is handed over to AI, no more energy is required, and the content can be customized and adjusted as needed.

After the outline is determined, press the Enter key, and in less than a minute, a PPT with a decent version is automatically generated.

Yes, in the rhythm of the intensive release of ChatGPT-like products, Baidu, which was the first to release large models in China and was the first to open internal testing, ran ahead again this time:

The exploration of the industrial landing of Wenxin Yiyan has been started.

It has to be said that when the time for Chinese large-scale model competition comes, for technology enthusiasts, there are new wonderful things happening every day.

Productivity tool test

The small partners who got into Baidu Wenxin Yiyan Enterprise Service internal test also revealed more information about the actual test to us.

Speaking of which, unlike the earlier NLP and CV large models, the reason why this round of large models can set off a storm is because of their versatility, they can go deep into various industries to produce chemical reactions.

Wenxinyiyan + tourism, generating a travel plan with pictures and texts is only a basic operation.

Wenxin Yiyan cooperates with airlines, and can also act in the real world, directly booking air tickets for you.

Wenxinyiyan + finance, it is not impossible for you to call it to write an investment proposal.

You don’t need to find the basic financial report information yourself, just ask Wen Xin, and it can find it online by itself. like this:

User: Has Baidu's financial report for 22 years been released? A word from the heart: On February 22, Baidu released its unaudited financial report for the fourth quarter and the full year ending December 31, 2022. [link to source of information]

Click the link, and the content of the financial report will be copied into the application interface immediately.

Next, it is the job of prompting the project.

Summarize important information, OK.

It is not difficult to analyze the health of a certain business.

In the final generated investment proposal, it is also possible to further modify and supplement the content through convenient human-computer interaction.

It only takes ten minutes to complete the first draft of a research report...

Then if it is more difficult, for example, let Wenxin Yiyan and an AI support a live broadcast?

It really does.

The copywriting can be generated in different ways, whether it is the literary style selected by Dongfang X, or the rap style of Laotie 666, Wenxin can do it with just a word.

Linked with the digital human, a pure AI live broadcast room can be launched.

Such a set of intelligent operations is equivalent to knocking down the original high cost of digital human application.

In addition, by accessing the graphite document of Wenxin Yiyan, you can automatically write the bidding document.

Provide guidance on how to do things:

It's not a problem either.

In this wave, it is true that productivity tools in government affairs, finance, e-commerce, and marketing are all covered.

Enterprises can't live without cloud services

As can be seen from the previous demonstrations, the large-scale access models of all walks of life will be directly upgraded whether it is internal office processes or external business forms.

Just like every industry in those years was worth redoing with the Internet, "every industry is worth redoing with AI" has gradually become popular.

So how can enterprises use Wenxin Yiyan's ability to develop their own applications?

At the end of March, Baidu reached a new result in a closed-door communication meeting with cooperative enterprises: Baidu Smart Cloud large-scale model platform launched the first batch of enterprise internal testing.

As the world's first one-stop enterprise-level large-scale model platform, it not only provides large-scale model services including Wenxin Yiyan, but also provides various tool chains and a complete set of environments for developing AI applications.

In addition, the Baidu Smart Cloud large-scale model platform not only provides a full set of Wenxin Yiyan large-scale model services, but also supports various third-party open source large-scale models, becoming a distribution center for large-scale model production and distribution.

For the specific service model, Baidu also provides three options , which are suitable for enterprises and developers with different capabilities.

Inference service , enterprises can directly invoke the core reasoning capabilities of general large models and output reasoning results;

For fine-tuning service , customers can inject a small amount of industry data according to their own needs, and fine-tune their own large-scale model at a small cost based on the basic capabilities of the general-purpose large-scale model.

Hosting services , general-purpose large-scale models or fine-tuned industry large-scale models can be directly hosted on the cloud of Baidu Smart Cloud. On this basis, enterprise customers only need to use the large model without worrying about complex deployment and management issues. The large model has better usability, efficiency and security.

In addition, Baidu Smart Cloud's six smart product series will also be fully upgraded based on Wenxin Yiyan, and will be launched after the security assessment is completed in the future.

It is not difficult to see that due to the existence of large models, no matter which model is inseparable from cloud computing services.

In the past, when enterprises chose cloud computing services, they paid more attention to hardware performance such as computing power, storage, and network.

In the new era of AI, the ability of the AI ​​model running on it will also be included in the key examination scope.

This is the origin of the recently hotly discussed concept MaaS (Model as a Service) in the cloud computing industry.

For example, Baidu Smart Cloud believes that the service model of cloud computing will change from IaaS (Infrastructure as a Service) to MaaS.

But in fact, this is exactly the situation that cloud computing vendors with AI capabilities like to see.

On the one hand, cloud computing is supported by computing power and data, which lowers the threshold for using large models

On the other hand, AI also makes cloud services more intelligent, improves efficiency, and expands more usage scenarios.

As a result, various cloud computing vendors have begun to connect and integrate cloud computing and intelligence, which originally belonged to two systems, and "integration of cloud and intelligence" has become an industry consensus for a while.

In fact, the integration of cloud and intelligence is not a new word, it first appeared in the strategy released by Baidu Smart Cloud in 2020.

The goal back then was to build computing power, frameworks, models, and scenario applications into standardized products through the integration and innovation of cloud computing and artificial intelligence. Lower the threshold for enterprises to acquire and use artificial intelligence. Subsequently, Baidu Smart Cloud upgraded the cloud-intelligence integration strategy twice in 2021 and 2022 respectively.

And on April 11, Alibaba Cloud also publicly stated that the integration of cloud and intelligence makes intelligent innovation within reach and enables the industry to move towards intelligence in an all-round way.

The unanimous statements of the two major cloud vendors mean that cloud-intelligence integration has become a consensus in the cloud computing industry, and it also heralds the arrival of a big industry opportunity for cloud-intelligence integration.

From the perspective of industry trends, there are two types of models in the transition from cloud computing to AI:

One is the cooperation between cloud computing vendors and third-party AI vendors. Needless to say, the combination of Microsoft Azure cloud + OpenAI, Amazon has also successively reported news of in-depth cooperation with open source AI forces such as StabilityAI and HuggingFace.

The second category is that cloud computing vendors with strong AI capabilities launch their own AI models, such as Google Bard + Google Cloud.

On the domestic front, Wenxin Yiyan + Baidu Smart Cloud is a representative case of the second category.

In the new era, what kind of cloud service should we choose?

Under the new form of cloud-intelligence integration, the standards for measuring the quality of cloud services will naturally have to adapt and change accordingly.

In addition to the most direct computing power, the strength of AI capabilities, and the price and cost, there is another point that is very critical:

It is the synergy between the chip, framework, model, and application four-tier architecture. On March 16, at the Wenxin Yiyan press conference, Baidu Group CEO Robin Li once proposed that "in the era of artificial intelligence, the technology stack in the IT field has undergone fundamental changes."

That is to say, the original three layers of "chip-operating system-application" are upgraded to four layers of "chip-framework-model-application" .

For example, Baidu is the only company in the world that has self-developed technologies on these four-layer architectures, from the Kunlun chip chip, to the deep learning framework of the flying paddle, to the Wenxin pre-trained large model, to search, intelligent cloud, and automatic driving. , Xiaodu and other applications.

Integrating the four-layer architecture together becomes the base of Baidu AI .

The Baidu AI Big Base implements a platform-based feedback closed-loop mechanism. There are many feedbacks between each layer, and by continuously obtaining feedback from real business scenarios, it realizes end-to-end optimization and provides enterprises with ultimate performance.

Enterprises no longer need to spend a lot of money on building infrastructure, but can use it on demand like using hydroelectric energy, truly realizing the cost reduction and efficiency increase of the whole process of AI production.

At the Boao Forum for Asia not long ago, Shen Dou, executive vice president of Baidu Group and president of Baidu Smart Cloud Business Group, made a further interpretation, "Efficient collaboration between the four-tier framework is the key to solving the bottleneck of computing power."

If we can increase the performance of an application by 100% with the same computing power input, it is equivalent to reducing the computing power demand by 50%.

In fact, Wenxin Yiyan has taken this step to explore the industry, but it is not Baidu's first attempt in AI enterprise services.

Technically, Wenxin Yiyan did not appear out of thin air, but developed on the basis of the Wenxin large model family.

As early as 2019, Baidu launched the knowledge-enhanced large-scale model ERNIE, which has developed into a unique technical advantage with massive Chinese data and knowledge graphs.

On the basis of general large-scale models, Baidu has also gone deep into industry scenarios and launched a series of large-scale models for industries such as finance and electric power.

It is also in the past explorations in many industries such as industry, energy, and finance that Baidu discovered the pain points of the previous generation of AI customization solutions that were costly and difficult to replicate.

Now, the general large model is expected to greatly reduce the threshold for enterprises to use AI.

From a more macro perspective, large models bring new opportunities for industrial transformation, and MaaS will leverage the trillion-level market.

For the reason, Shen Dou explained it as a "double-effect improvement" in production efficiency and user experience.

In terms of production efficiency, the production process is nothing more than dealing with people, machines, and systems. In the future, it is very likely that a smart person will manage a bunch of robots, and these robots will execute his instructions and reshape the entire production line.

In terms of user experience, the era of graphical interfaces gave birth to the mouse and keyboard, and the touch screen in the mobile era has once again greatly improved the user experience. This time, it will liberate human-computer interaction more completely, and completely freely use natural language to make demands on the machine. , the whole interaction will be more direct and efficient.

Ultimately, AI will accelerate the "intelligent transition" of the whole society.

Specific to the development of China in the AI ​​era, Shen Dou believes that there are still special advantages.

China's huge industrial chain has a very strong "multiplicative effect": even if the efficiency of each individual link is only improved a little, the overall efficiency of all links is superimposed beyond imagination.

And now, it's time to get ready for everything.