Yuyang comes from the temple
Qubits | Official account QbitAI
Guess what, Wen Xin evolved one by one, first turning Baidu upside down.
Now, what temporary pull, you have to read 5k+ words of documents first, Baidu employees are not afraid at all.
Poke the "intelligent insight" window in the document, no matter how long the document is, the large model can give a summary in seconds, so that participants no longer have to be "confused".
Programmers also took advantage of Wen Xin's words, and everyone was equipped with code assistants.
Knocking on a function and waiting for AI to automatically complete has become the norm.
This wave ah, the main thing is one:
Such a scene of great efficiency improvement is being staged every day on Baidu's intelligent work platform.
At the Zhongguancun Forum not long ago, Li Yanhong emphasized that Baidu should reconstruct all products based on a large model such as Wen Xin Yiyan and under the AI native thinking.
Ruliu took the lead in experiencing the changes set off by large models in the office scene.
Now, we have obtained more first-hand internal testing data, how the big model can reconstruct the form of daily work, we can find out this time~
Wen Xin reconstructed office daily life in a word, like a stream taking the lead in new AI native "artifacts"
Jarvis in "Iron Man" seems omnipotent and fully satisfies people's imagination of AI assistants. And Ruliu, a "super work assistant" who can coordinate the overall situation, has a preliminary taste of Jarvis.
Super assistant is a new product of the big model that fully absorbs enterprise knowledge and workflow. The so-called "overall planning" means that in any work scene, it can be aroused in time, and through normal dialogue, help employees directly obtain knowledge and complete work arrangements.
△ In the search box or the lower right corner of the mobile terminal, one click can be summoned
For example, if you want to find a working document but can't remember the file name, ask the super assistant like this: Find the document that XX sent at the XX meeting.
AI can directly get your intentions and quickly find targets in the enterprise knowledge base.
For example, if you want to ask a colleague for a meeting, but you don't know when the other party has a schedule, the super assistant can communicate and understand in minutes, and even the schedule is directly arranged.
With the blessing of Wen Xin, this super assistant is also "omniscient". Whether it is a general question such as "how to learn AI native applications", or want to specifically understand a certain welfare policy of the company, from external Internet information to internal enterprise knowledge, you just ask, it can answer immediately.
And can directly locate specific learning materials.
The changes brought about by Wen Xin's words are more than that.
In more segmented work scenarios, large models are creating more and more new possibilities.
Scenario 1: Programming development
In the general scenario, the blessing of Wen Xin's words has significantly improved the efficiency of information retrieval and transmission.
But the big model of Wen Xin is not a pure "liberal arts student".
Based on the understanding and reasoning ability of the Wenxin big model, Baidu Ruliu has launched a code assistant called Comate (Coding Mate).
After the release of Comate, there are constantly new smart coding black technologies.
The latest development is that you write requirements and Comate automatically generates code.
It is worth mentioning that support Chinese - to some extent, it is implemented Chinese writing code (manual dog head).
Conversely, the two most annoying things for programmers: commenting on their own code and others not commenting on their code, can now be solved by Comate.
Unit testing is also an important part of the software development process and is key to improving the quality and maintainability of code. However, writing unit tests is also a more tedious and boring part of the work for programmers.
Such a job, thrown to AI automatic generation, is very appropriate~
At present, Comate covers more than 30 mainstream programming languages and frameworks, and supports mainstream IDEs such as VS Code and Jetbrains series.
It can be used as a plug-in in many common IDEs.
In addition, it is understood that Comate's training data comes from Baidu's internal accumulation of excellent programming field cases for many years, which makes the code it recommends and generates more in line with the needs of real research and development scenarios.
It's not just code assistants like Comate that can help programmers.
If you want to write a R&D summary for the submitted code every day, query DevOps usage problems, and use the small K R&D assistant to assist.
According to the code and circulation card submitted every day, intelligently summarize the week's work:
Get a quick DevOps tip with conversations:
No more struggling to check documents and turn records, it can be said that the programmer is ecstatic.
Scenario 2: Live chat
Besides writing code, what is the most common job of a programmer? It is probably the common pain point of Internet people, "alignment".
For example, planning an event is likely to brainstorm a group, collect needs a group, and execute another group.
For every 99+ chat record moved, there are innocent workers climbing stairs until they are bald.
In this scenario, Wen Xin's long-term conversation summary ability comes in handy: for example, the new IM intelligent summary function on the stream can automatically generate summaries for the merged and forwarded chat records.
The operation is also very simple, select the chat history normally, check "Generate summary based on message and display" when merging and forwarding, and then wait for AI for a few seconds.
In this way, the person forwarding the message no longer needs to type to explain the background, and the person reading the message can quickly get the focus, and the communication efficiency has really improved a lot.
Scenario 3: Meeting scenario
The same large model long dialogue summary and analysis capabilities are also integrated into the video conference.
Now, even if you join halfway, you don't have to worry about not being able to keep up. Based on Wen Xin's words, the Ruliu Intelligent Conference can not only summarize the content of the meeting, but also mark the key points in real time through "meeting AI insight", and divide the discussion topics, views, and key speeches according to time.
Even the to-do items mentioned in the meeting, AI can automatically @ people and make arrangements.
In addition, the AI native application brought by Wen Xin's words has also penetrated into the daily life of Baidu employees in more work links.
For example, in the direction of knowledge management and content creation that Baidu has always emphasized, Ruliu recently launched the "intelligent creation" and "intelligent insight" functions of the knowledge base.
Relying on the text generation and creation ability of Wen Xin's words, as well as the ability to read and summarize documents, combined with the internal knowledge of the enterprise, users only need to click on the required content, they can wake up the functions of text optimization, rewriting, summary, translation and Q&A, accurately meet the needs of optimizing creation, obtaining information, accelerating understanding and other needs in the knowledge base scenario, and intelligently inspire employees' thinking and stimulate innovation.
For example, when employees deal with time-consuming and repetitive problems in the workflow, in the "intelligent assistance area", AI can quickly understand the current page work content and employee roles in the form of "co-pilot", and provide employees with personalized business knowledge recommendations and intelligent Q&A.
This wave, the large model is indeed a bit of a direct hit on the pain point of the work, and the practicality is full of taste.
Big models enable "singularity moments"
In fact, standing on the front line of the wave of large models, Li Yanhong has already made a judgment: big models are about to change the world, and the future trend is that all applications will be developed based on large models.
For Baidu itself, it is necessary to "redo all products, not integration, not access, redo, refactoring".
Wen Xin's words were released for 3 months, and the speed of evolution can be described as "thousands of miles a day". Now, this application change has been first implemented in Baidu, and has penetrated into every detail of the work of Baidu employees.
We learned that the flow of Wen Xin's blessing has received a lot of praise within Baidu. Especially with features like chat history intelligent summary, the big guys have felt the leap of experience from it.
But for Baidu, and even for the entire technology industry that is focusing on large models, more importantly, the reconstruction of such a stream has become a reference example in the exploration of large language models:
In the era of big models, how will people's working lives change? The answer to the current version has become clear in practices such as Ruliu.
All human-computer interaction will become more and more natural. Everyone can have a smart work assistant that is professional enough and real-time.
The office scene is taking the lead in ushering in "leapfrogging" experience changes and ushering in the "singularity moment". And such a "singularity" will continue to occur in more scenarios.
And when the prologue begins, the tide is unstoppable. The native application of AI, rooted in large model technology and AI technology, has begun to lead explosive growth around the world.
Midjourney, an AI painting app, achieved revenue of about $100 million in a year with a subscription payment model; OpenAI's ChatGPT iOS version exceeded 500,000 downloads in a week; The AI partner blessed by Baidu's large model capabilities exploded in the internal test, and the number of related search questions and answers surged...
Technology manufacturers, the pace of in-depth application of large models is getting faster and faster.
Abroad, caution such as Google has also added AI dialogue capabilities to search. Microsoft has further integrated Bing and ChatGPT, and while Bing is backed by GPT-4, ChatGPT can also use Bing's capabilities.
Domestically, taking Baidu, which has accumulated the deepest accumulation in large models, as an example, the internal flow provides "reference answers", and the new speed of external capability output is also remarkable:
As soon as Wen Xin said the "full moon", he quickly went to work, and showed "work skills" such as making PPTs, specifying travel plans, generating investment proposals, and linking digital live broadcasts at technical exchange meetings.
On May 25, at the Vientiane Conference, Baidu officially announced the use of large models to reshape search capabilities and launch AI partners.
……
Every technological explosion brings a lot of new opportunities. This time, it was AI native thinking that changed everything.
Those who are prepared have already taken the lead in handing over reference answers.
More "singularity moments", the future can be expected.
— End —
Qubits QbitAI · Headline number signed
Follow us and be the first to know the latest scientific and technological trends