DingTalk Ecosystem Conference Insight: Unicorns and Big Platforms Enter a New Era

06/27 2024 408

On June 26th, DingTalk held the "2024 DingTalk Ecosystem Conference" in Beijing, with "AI" and "going global" as the key topics of the conference.

At the conference, in addition to the launch of practical functions such as "AI Search", "AI Assistant", and "AI Meeting Hardware", it is noteworthy that DingTalk also announced that it will integrate Alibaba's Tongyi large model, as well as six unicorn large models: MiniMax, Dark Side of the Moon, Zhipu AI, LeoSpace, Lingyi Wanwu, and Baichuan Intelligence.

Although the move was significant, DingTalk unfortunately lost the initiative. Feishu, another office collaboration platform, announced its cooperation with multiple unicorn large models such as MiniMax, Dark Side of the Moon, Zhipu AI, Lingyi Wanwu, Baichuan Intelligence, and Jieyue Xingchen two days earlier.

It is worth noting that DingTalk relies on Alibaba, while Feishu relies on Douyin. Both Alibaba and Douyin are currently giant platforms in China, possessing the richest large model research and development resources, as well as leading general large model foundations. Among the large models cooperating this time, MiniMax and Dark Side of the Moon are also clearly general large models.

In theory, these giants' general large models were originally in a competitive relationship with the general large models of unicorn companies. So what are the commercial considerations of all parties in such cooperation? And why is this trend only emerging now?

01. The Ambitions and Bottlenecks of Unicorns

In fact, globally, it has become a new trend for large models to develop separately from their affiliated platforms. The reason for this is that if a large model is too closely tied to its affiliated commercial platform, with strong exclusivity, it is not conducive to the growth of both the platform and the model.

Specifically, when a large model is tightly integrated with a specific platform, it may limit the model's technological accumulation in other scenarios, and at the same time, the platform's own business model will only be confined to a specific large model.

That is to say, from the perspective of giant platforms, it is a general trend for their large models to go out, or even for other large models to come in. For example, Opera, under Kunlun Wanwei, has reached a deep cooperation with Google Cloud, integrating Google's Gemini large model into its native browser AI assistant Aria. It is worth noting that one of Google's main products is also a browser.

In the past year and a half, most commercial applications of large models have sought external cooperation, but the main businesses of previous partners often did not compete with each other, such as integrating large models into hardware partners. In fact, these unicorn large models themselves can also develop into an internet tool or platform. And for these unicorns, being able to develop into a new giant platform through technology is the ideal commercial vision. However, these unicorn large models still chose to cooperate with platforms like DingTalk.

Not to mention domestic unicorns, even OpenAI, the largest bellwether of global large model unicorns, has always wanted to rely on its own technology and exposure to platformize itself. These actions include but are not limited to releasing multiple multimodal large models, charging its platform for C-end users, and the OpenAI App Store.

But after a year and a half of exploration since its explosion, OpenAI seems to have finally realized that the bottleneck of large model implementation lies precisely in its ambition to platformize. Wanting users to simply perform multi-scenario tasks on the platform it has built and develop habits based solely on technology is ultimately too idealistic. OpenAI obviously also needs to bet on other paths.

On June 12th, OpenAI and Apple officially announced their cooperation, exposing each other's current dilemmas. From Apple's perspective, the cooperation with OpenAI will help the development of its voice assistant Siri and may promote iPhone sales. At least when other mobile phone manufacturers are emphasizing the need to create AI smart devices, Apple will not fall behind in this wave. From OpenAI's perspective, integrating its products into Apple products will allow it to reach Apple's global users, and OpenAI will gain significant exposure through this application scenario.

From this cooperation, it is not difficult to see that Apple's dilemma is its own backward AI, while OpenAI's dilemma is that user growth has peaked, or further speaking, its scenario data collection capabilities have peaked, and data is precisely one of the core competencies for the future development of large models. Judging from the various lawsuits OpenAI has faced in the past year and a half, it cannot be said that it lacks high-quality data sources. The controversy over content调用with the New York Times has not been resolved to this day. Although Apple itself is a hardware device, the essence of this cooperation is not the integration of large models into hardware, but rather OpenAI's entry into the Apple ecosystem platform.

All this is clearly seen by all domestic large model unicorns.

Since even the global bellwether has not been able to find a good path to platformize large models and has temporarily joined other platforms, unicorn large models that lack scenario data and cannot yet achieve platformization will naturally also temporarily "give up their dreams" and choose to cooperate with already mature platform ecosystems.

02. Both Platforms and Large Models are Opening Ecosystems

There are three types of cooperation models between DingTalk and large models this time, which largely预示the future cooperation methods between other internet platforms and external large models.

DingTalk's cooperation model with its own Tongyi large model involves it undertaking AI capabilities for products such as IM, documents, audio, and video. The cooperation models with the already officially announced models also combine the characteristics of their large models to explore the application of different model capabilities in products and scenarios. For example, DingTalk is working with Dark Side of the Moon to explore educational application scenarios based on the long-text understanding and output capabilities of large models.

In terms of AI Agent development, DingTalk has opened up an AI Assistant (AI Agent) development platform to large model ecosystem partners. When developers create AI assistants on DingTalk, in addition to the default Tongyi large model, they can also choose large models from different vendors based on their own needs.

For customers' personalized scenarios and needs, DingTalk will work with large model vendors to customize corresponding intelligent solutions for customers and provide services such as model training and optimization, AI solution creation, and AI customized application development. It can also achieve privatized deployment of models.

These three models are at three different levels of integration. Among them, the lightest integration is AI Agent, which essentially involves large models calling the platform's data interfaces to independently build additional applications on the platform. This will also be the most open model, and it is expected to integrate more manufacturers' large models in the future. This approach is truly born around the platform ecosystem, with the initiative in the hands of the platform.

The difference between the other two models lies in whether the platform and large model capabilities are integrated first and then matched with customers, or whether customers are matched first and then customized integration. The former initiative lies in the hands of both the platform and the large model, while the latter initiative lies in the hands of the customer. In fact, these are currently the two main commercialization ideas for large models and previous new technologies.

Specifically, matching capabilities first and then customers can generate greater commercial value, while matching customers first and then customized integration will not have a significant impact on the overall structure of the platform at the macro level. After all, the core capability of office collaboration platforms is to establish work and management processes for enterprises, and customized processes have limited impact on general processes.

These three approaches have laid a good framework for future cooperation between platforms and unicorns at the AI level.

As we all know, the three main online collaborative office management software on the market, DingTalk's advantage lies in workflow management, Feishu's advantage lies in content process management, and Enterprise WeChat lies in marketing process management. In terms of the original degree of ecological openness, Enterprise WeChat relies on the WeChat system, with a strong third-party ecosystem and C-end users. Most of its ecological cooperation models are also based on self-developed applications through interface calls rather than strategic cooperation. That's why we haven't seen Enterprise WeChat officially announce cooperation with any unicorn large model, as essentially these large models can be integrated into Enterprise WeChat at any time.

In fact, not only are office platforms opening up their ecosystems, but large models themselves are also opening up their ecosystems. After all, the aforementioned MiniMax, Dark Side of the Moon, Zhipu AI, Lingyi Wanwu, Baichuan Intelligence, etc., have not only chosen DingTalk but also Feishu, and may choose more similar platforms in the future. Not to mention that these large models also have cooperation with other different types of platforms.

Such a multilateral open ecosystem destines that large model unicorns and giant platforms actually stand on an equal and mutually selectable relationship. Especially since the parent companies behind these platforms have also invested in these unicorn large models, such as Alibaba and Tencent investing in MiniMax and Dark Side of the Moon. After all, for their parent companies, placing bets on multiple parties is the most rational choice. At the same time, from the current investment patterns, their parent companies will not swallow up these unicorns.

On this basis, large models themselves have more possibilities to be implemented in more scenarios and gradually develop into platforms. In short, domestic large model unicorns have now given up direct platformization and have chosen the path of first scenarioization and then platformization. The only unicorn that can simultaneously manage both platformization and scenarioization is OpenAI.

03. Conclusion

The relationship between unicorns and big platforms often has two types. One is that the unicorn endures all the pursuits and obstructions and becomes a big platform itself. The other is that the unicorn itself is completely incorporated into a big platform. There may be other types of special cases, but in the business world, these are the two main situations.

The signal being released now is the equality, healthy competition, and common development between unicorns and big platforms, all stemming from the innovations brought about by new technologies, giving unicorns sufficient initiative.

In fact, this is also the most ideal state of the business world. The emergence of every new technology will lead to such drastic changes in the business landscape, which is essentially a dimensionality reduction strike of technology on the business world.

After all, business models are easy to replicate, while core technologies are protected by law.

*The cover image and images in the article are sourced from the internet.

Solemnly declare: the copyright of this article belongs to the original author. The reprinted article is only for the purpose of spreading more information. If the author's information is marked incorrectly, please contact us immediately to modify or delete it. Thank you.