Author|Gong Zhouzhou
Produced by unnamed science and technology
At the DingTalk 2023 Ecological Conference, Ye Jun, President of DingTalk, announced the latest progress of DingTalk intelligence: 17 product lines and 55 scenarios have been fully connected to the big model. In addition, it announced the intelligent foundation of AI development environment and tools, AI PaaS, to help enterprises develop and deploy AI applications.
DingTalk AI PaaS is a cloud computing infrastructure that supports large-scale applications, which can provide developers with AI development tools and resources, such as APIs, data processing and model training, as well as cloud computing services needed to integrate and deploy applications.
At the meeting, DingTalk gave the industry's first commercialization plan for large-scale model landing application scenarios: on the basis of the annual fee of 9,800 yuan for DingTalk Professional Edition, an increase of 10,000 yuan can obtain 200,000 large model call quotas; On the basis of the exclusive DingTalk annual fee, an increase of 20,000 yuan can obtain 450,000 large model call quotas.
"Whether it is DingTalk or DingTalk ecological partners, we should really face the opportunities brought by productivity change to think about how we should be a cool Talk, and how to effectively solve the problems of small and medium-sized enterprises in the macro world." Ye Jun believes that this is the key.
100 days after announcing the access to the Tongyi model, DingTalk decided on a path of ecological development.
At the ecological conference, Ye Jun announced the opening of the intelligent base (AI PaaS) to ecological partners and customers, and used large models to help ecological partners redo products. That is, DingTalk makes basic capabilities and basic products, and opens these capabilities and products as a base to ecological partners, who upgrade more targeted application scenarios according to their own industry know-how.
Based on this, Ye Jun, president of DingTalk, and Luo Yihang, founder and CEO of Pinplay, launched a unique thinking storm on the practical significance and core capabilities of AI PaaS.
01
Industry dead volume model but lack of application
Luo Yihang: Today, I see that DingTalk AI assistants, digital employees, and "magic wands" actually support many product suites, even more than a dozen scenarios. This thing actually I think is very good, because now we see too many large models in China, but only a few are applied to the model, such as AI model training environment, do vector database, do this kind of thing very few people, everyone is rolling models, 100, 1000 models, a few ecological tools, less geometric applications, the whole is such a form, what do you think of this?
Many people in the industry make models, roll models, and roll models with their lives, but they do not do applications and do not do ecology. This thing seems to be a little different from when we first started the mobile Internet more than 10 years ago, what do you think of this problem?
Ye Jun: My view has always been very firm, I think the application scenario is more important. Application scenarios give us a lot of opportunities in all walks of life, just like Watt, everyone says that Watt invented the steam engine, but in fact, Watt is just a person who improves the steam engine, and he really applies the steam engine to the scene, so I think the application scenario will be more important.
Luo Yihang: Many people feel that the application scenario of AI is not clear, so it is not easy to do the application now. Let's think about DingTalk, DingTalk itself is a super application, maybe we don't want to say that we are an application, we prefer to say that we are a PaaS platform. But now, we say that DingTalk should be based on a big language model to do applications, and DingTalk now thinks that it is enough to do a good job?
Ye Jun: Actually, this answer is also very simple in our minds. DingTalk now has 600 million users, 200 million monthly active users every month, and there are only two such applications in China that can be called super APP - WeChat and DingTalk, so I don't have the confidence to talk about making applications, and it is most suitable to use the DingTalk ecology to prosper the application together.
Luo Yihang: WeChat and DingTalk, Douyin may not be convinced.
Ye Jun: Right. We're a development platform that will become more suitable for app creation.
Luo Yihang: Just now DingTalk mentioned more than ten scenes, such as meeting scenes, PPT scenes, and so on. Before, I discussed with many VCs why you only look at the model and not the application for large language models. He said that we don't have a scene in our application now, we can't see what AI native is, we can't see it, how we vote. So it's quite interesting that DingTalk announced more than 10 scenes today, such as the scene of the final order, I think the happiness is over, how did you find these scenes?
What scenarios did you create to prove that you have the ability to make applications? What scenarios did you assume might have a prospect, and how did they come about?
Ye Jun: Actually, there are many scenes. Demonstrate these scenarios as they are basically the real needs of customers and employees. Many scenes are PaaS by me, and I don't think it's very reliable. For example, the digital avatar, originally I also wanted PaaS, in fact, the boss didn't like this kind of thing, but the employees thought it was good, so we thought we could do it. I also believe that more and more such scenarios will occur in the future.
02
Enterprise data reliability drives AI application reliability
Luo Yihang: Actually, I also talked about a thing, when DingTalk was doing a demonstration just now, I counted the process of several demonstrations, and I probably helped 3 or 4 small places. Then you say that the big model is not very reliable, and many people are applauding at the bottom. I think they should be welcoming or accommodating these small pools of difference that emerge from the big model.
Of course, there are indeed some problems with large models, such as model illusions, resulting in nonsense or misoperation, but our work cannot tolerate mistakes, we can be held accountable for employee problems, there is a 360° environmental assessment, and HR can not be renewed. But what if future mistakes are caused by models, not people? If you can't be held accountable, accountability has to be taken down, and everything that is actually done is done for nothing.
So how do you see the contradiction between the illusion problem and the problem of work certainty in the model itself, and how to solve the problem of reliability?
Ye Jun: Yes, I think this is a prerequisite for any big language model to be applied to real scenarios. If you can't solve the reliability problem now, you can't solve the safety problem, in fact, it is almost impossible to put it into production, it is a toy. Therefore, the AI PaaS made by DingTalk today largely solves the problem of exclusive data tuning capabilities in serious scenarios of enterprises. With this ability, the enterprise technology department and the information digital department can do a lot of localization tuning, and after tuning, it becomes a real productivity tool available.
Luo Yihang: That is to say, it is actually to use some internal tuning and debugging to make it more reliable.
Ye Jun: Yes, combined with DingTalk's existing enterprise-level data with strong certainty, including address books, schedules, agencies, meetings, etc. Use the determinability of this internal data to accelerate model determinability.
Luo Yihang: So that it is not just a tool for brain teasers.
Ye Jun: Right.
Luo Yihang: So data will become more and more important in the future, because data is an asset.
Ye Jun: Yes, there is a lot of enterprise data that has been applied early, and it will be very helpful to the intelligence of his own enterprise in the future. Because it also completes a certain intelligent idea.
Luo Yihang: Just now we said that the certainty of internal data of the enterprise should be used to make the model itself more certain, of course, it is also necessary to do tuning. Therefore, I want to confirm again, not to mention the open platform for enterprises to use, that is, our own dozen products, DingTalk AI assistant, digital employees, document functions, etc., is it possible to access the platform in the future? For example, Wen Xin Yiyan, Spark, etc.
Ye Jun: We can reply clearly, definitely access. We will acquiesce to Tongyi Qianqianqing, because Tongyi is currently the most mature with us.
However, in the future, whether it is Tsinghua simple or Baichuan big model, etc., it has its own advantages in different scenarios, and it will also form some industry solutions, and we can have some docking of more targeted scenarios. At the moment, we are doing tests in many specific scenarios.
Luo Yihang: Just now, DingTalk, like many consumer goods conferences, announced a price package at the end. The middle gear is 118,000/year, including 450,000 intelligent service calls, do you think the cost is high or low?
Ye Jun: Actually, we are now drawing conclusions based on statistics, and there is room for further adjustment in price in the future. I think IT costs will continue to decrease, and currently it is 5 cents per minute. Even cheaper than the price of a phone call.
03
Be truly ecological, not in the name of ecology
Luo Yihang: You said that we only have 1,500 people, so we can't do an industry application. So, can we think that because there are fewer people, you have to do ecology?
Ye Jun: One of the affirmative answers is that the lack of people is a very important reason. But the other reason is that we do think that China may not lack a good software company, but China lacks a very cool company, a cool platform company with strong ecological support. We want to be No. 1 in this type of company.
Luo Yihang: Little DingTalk is cool DingTalk. In case there are more people, do you start to do vertical applications?
Ye Jun: We will encounter this problem in the future, and one day there will be more people as we grow. But I think I can't be sorry for the ecology, because to accompany us all the way up, 2B companies must have such feelings. In addition, we want more people to invest in the construction of DingTalk experience, so that the DingTalk experience can be used more easily and without any training, which is something we want to do now.
Luo Yihang: It is to let everyone use it more.
Ye Jun: Right. We also encourage our ecology to not only connect with DingTalk one platform in the future, but also multiple platforms.
Luo Yihang: This is also quite interesting, let's talk about ecology, and finally you released the AI app store, do you think it looks like that thing? Unlike OpenAI's recently released plugin store, the logic is the same.
In fact, every of its ecological partners is in it, whether it is HR or contract finance, it seems to have become a plug-in. For developers, in the past mobile store era, each of them was a native APP, and now each of them has become a plugin, a plugin like a browser. How do you think the developers themselves should accept this thing, and how can they re-refresh their cognition and change their own definition?
Ye Jun: We have communicated with many ecological partners about plug-in. I think there is a consensus that plug-in is a trend that everyone agrees with, because in fact, it is ultimately to reduce the cost of partners and make the use easier, so plug-in can be used for multiple platforms, this trend I think everyone has a consensus.
Luo Yihang: That is to say, a set of plug-ins can not only be adapted to DingTalk, but can also be adapted to other platforms, and can even enter different scenarios of the same platform. I think this may be more important, not every company has to do a login registration, WeChat scan code login DingTalk scan code login is also very good.
Luo Yihang: Actually, there are a few points that I am very impressed by today. The first is that the basic model of access is not necessarily a general meaning; The second is the DingTalk AI Open Store. The last time the information technology industry discussed opening up on a large scale was when the mobile Internet was slowly born in China, and everyone said that propaganda should be an open platform. The opening of the mobile Internet has come to today, 13 or 4 years, although it is more open than the PC era, but in fact there are still some problems. What do you think of openness in the AI era? Do you do something else in the name of openness, or do you really believe in openness.
Ye Jun: I think it's really open. The first is indeed faith, and since you believe, you are responsible for what you say; Second, in fact, employees also believe in me; The third is the intelligent era, if not for the interconnection of data and systems, the value is very thin, which is also a problem faced by China's enterprise service software industry. To solve this problem, I think it needs to be open, and there is no shortage of software companies in the industry.
Luo Yihang: Yes, there is no shortage of a software company, especially a large-scale software company, but there is a small company that only has a good vision that can do cool things and cool ecosystems and cool tools, and make others prosper.
Ye Jun: We want to become such a company. Common expectations.