Large Models Compete in Ecosystem Development, Baidu Takes Another Lead

[baoshi.rao]

"Without a rich ecosystem of AI-native applications built on foundational models, large models are worthless."

At Baidu World 2023 on October 17, Robin Li demonstrated AI-native applications hands-on to encourage more enterprises and developers to join the large model application ecosystem.

In the nearly one year since the release of ChatGPT in November last year, major tech giants have equipped themselves with "AI Copilots," intensifying competition around large model application scenarios. By reconstructing traditional flagship applications, these tech giants have shifted the competition in the AIGC era from the first phase of the "hundred-model war" to the current second phase of application implementation.

Microsoft CEO Satya Nadella's grand vision includes integrating large models into all of Microsoft's core businesses. Google CEO Sundar Pichai is also busy leveraging large models to revamp all Google products. OpenAI CEO Sam Altman, after launching the paid version of ChatGPT, quickly rolled out apps for iOS and Android.

Domestically, Baidu has acted earlier and more thoroughly in this regard. In May this year, Robin Li stated publicly: "Baidu aims to be the first company to completely rebuild all its products—not just integrate or connect, but reconstruct!" The newly unveiled Wenxin Large Model 4.0 at Baidu World 2023, with significant improvements in comprehension, generation, logic, and memory, was praised by Robin Li as "on par with GPT-4" in overall capability. Baidu's "reconstruction" efforts are set to accelerate further.

<p style="margin-top: 0px; margin-bottom: 28px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; line-height: 32px; text-align: justify; color: rgb(59, 59, 59); word-break: break-word; font-family: "PingFang SC", "Microsoft YaHei", Helvetica, "Hiragino Sans GB", "WenQuanYi Micro Hei", sans-serif; letter-spacing: 0.5px; text-wrap: wrap; background-color: rgb(255, 255, 255);">不仅如此，百度对大模型的布局，显然不止于应用层的“造林”。在百度集团<span class="spamTxt" style="margin: 0px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000;">资深</span>副总裁、百度移动生态事业群组总经理何俊杰看来，始于技术、强于应用的大模型，最终必将惠于生态。因此，<strong style="margin: 0px; padding: 0px; box-sizing: border-box; outline: 0px; border: 0px; --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; color: rgb(20, 20, 20); white-space-collapse: preserve; display: inline; background: rgb(249, 250, 251);">百度移动生态决定开启大模型的下一程，“从AI人人可用，到生态共荣”</strong>，希望基于百度的技术土壤，与用户、客户、伙伴共育大模型的“生态雨林”。大模型的应用浪潮，也将从全球大厂或AI公司的独角戏，拓展成为生态共荣的普惠机遇。</p><p style="margin-top: 0px; margin-bottom: 28px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; line-height: 32px; text-align: center; color: rgb(59, 59, 59); word-break: break-word; font-family: "PingFang SC", "Microsoft YaHei", Helvetica, "Hiragino Sans GB", "WenQuanYi Micro Hei", sans-serif; letter-spacing: 0.5px; text-wrap: wrap; background-color: rgb(255, 255, 255);"><strong style="margin: 0px; padding: 0px; box-sizing: border-box; outline: 0px; border: 0px; --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; color: rgb(20, 20, 20); white-space-collapse: preserve; display: inline; background: rgb(249, 250, 251);">A</strong></p><p style="margin-top: 0px; margin-bottom: 28px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; line-height: 32px; text-align: justify; color: rgb(59, 59, 59); word-break: break-word; font-family: "PingFang SC", "Microsoft YaHei", Helvetica, "Hiragino Sans GB", "WenQuanYi Micro Hei", sans-serif; letter-spacing: 0.5px; text-wrap: wrap; background-color: rgb(255, 255, 255);">从卷模型，到卷应用，再到卷生态，百度似乎始终“快人一步”。</p><p style="margin-top: 0px; margin-bottom: 28px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; line-height: 32px; text-align: justify; color: rgb(59, 59, 59); word-break: break-word; font-family: "PingFang SC", "Microsoft YaHei", Helvetica, "Hiragino Sans GB", "WenQuanYi Micro Hei", sans-serif; letter-spacing: 0.5px; text-wrap: wrap; background-color: rgb(255, 255, 255);">而卷应用，是过去几个月百度移动生态最关键的考题。5月份的万象大会上，何俊杰就宣布，要用AI原生思维重构百度移动生态，让AI人人可用。今年的百度世界上，他再次强调，<strong style="margin: 0px; padding: 0px; box-sizing: border-box; outline: 0px; border: 0px; --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; color: rgb(20, 20, 20); white-space-collapse: preserve; display: inline; background: rgb(249, 250, 251);">AI原生应用，是推动AI人人可用的最直接路径。</strong></p><p style="margin-top: 0px; margin-bottom: 28px; padding: 0px; box-sizing: border-box; outline: 0px; border-width: 0px; border-style: solid; border-color: rgb(229, 231, 235); --tw-shadow: 0 0 #0000; --tw-ring-inset: var(--tw-empty, ); --tw-ring-offset-width: 0px; --tw-ring-offset-color: #fff; --tw-ring-color: rgba(41, 110, 228, 0.5); --tw-ring-offset-shadow: 0 0 #0000; --tw-ring-shadow: 0 0 #0000; line-height: 32px; text-align: justify; color: rgb(59, 59, 59); word-break: break-word; font-family: "PingFang SC", "Microsoft YaHei", Helvetica, "Hiragino Sans GB", "WenQuanYi Micro Hei", sans-serif; letter-spacing: 0.5px; text-wrap: wrap; background-color: rgb(255, 255, 255);">在这场面向AIGC时代应用生态开发标准的抢位战中，不仅要比谁跑得更快，谁能拿到更多的反馈数据，建立更多的应用落地优势，而且要比谁的应用真正能够改变大众用户的工作和生活。两相叠加，大模型厂商才有望凭借“滚雪球效应”脱颖而出。</p>

Baidu's philosophy for application layer restructuring is simple. As He Junjie mentioned in a previous media interview, Baidu's concept for large models—'be the house'—permeates the entire strategy of Baidu's mobile ecosystem restructuring. 'This restructuring is not a minor adjustment for us, but a radical transformation. It's not simply about adding large models to our business, but about multiplying their impact.'

First, through restructuring, AI-native applications can solve problems that previous products couldn't address or didn't address well. For example, if we need to verify 'the ranking of industrial added value by country over the past 20 years,' in the past, we would have to search for data from different years and compare them manually. But if we pose this question to Baidu's new search engine, restructured with AI, it no longer presents a list of webpage links but instead generates a dynamic chart created by AI. This is the new capability of 'ultimate satisfaction' brought by Baidu's new search engine after being restructured with large models.

Additionally, large model technology can expand the capabilities of existing products, achieving what was previously impossible. This is particularly evident in the new Baidu Wenku. Today, this product, described by Robin Li as the 'most radical transformation,' can directly perform 'brain-intensive' tasks such as intelligent text generation and editing, document summarization and Q&A, text-to-chart generation and analysis, PPT chart generation, outline writing, and brainstorming.

For instance, during He Junjie's three-minute demonstration at the World Conference, starting with the instruction to search for a 'photovoltaic industry' report, Wenku quickly found the industry report, generated a 70-page summary in seconds, and then transformed the summary into a PPT, allowing for quick customization and presentation—completing the integration and presentation of work or study materials in no time.

The AI-powered new features of Baidu Wenku, evolving from a 'content tool' to a 'productivity tool,' are gaining increasing recognition from users. Within two months of its launch, it has attracted over 13 million users and surpassed 100 million functional uses.

Even before ChatGPT was launched, Robin Li predicted that in the next decade, AIGC would disrupt existing content production models, enabling AI-generated original content at one-tenth the cost while accelerating output by hundreds or thousands of times.

At Baidu World, the newly released AI editor in Baidu APP continues to focus on lowering the barrier for creators and establishing new productivity models. For example, "AI Same-Style Follow Shot" allows users to generate creative images with one click, while "AI Image Captioning" automatically identifies image information and produces multiple social media captions. This eliminates the struggle of crafting the perfect post for a travel photo, empowering more casual users to become creators.

The overhaul of Baidu's marketing platform delivers direct benefits to clients. As Robin Li puts it, "One person can now be an entire marketing team." Take the AI Native marketing platform Qingge as an example: based on natural language descriptions, it reduces the time for creative generation and marketing plan development to minutes while accurately targeting ads to broader audiences, streamlining the entire process from ideation to deployment.

He Junjie showcased a case study from IT education group Tarena. With a simple instruction like "Help me target ads to young people seeking IT jobs," the system automatically identifies the audience and generates compelling creatives. This approach increased commercial conversions by 23.3% year-over-year and boosted ad ROI by 22.7%.

Through this application restructuring initiative, Baidu's mobile ecosystem is progressing toward its initial goal of becoming a "super APP" in the AIGC era, thereby opening new traffic channels. With the enhanced user experience and efficiency brought by AI-native applications, Baidu's mobile ecosystem also has the potential to attract new users and increase engagement.

In a September media interview, He Junjie stated that since Wenxin Yiyan became available to the public, more highly educated and younger users have been returning. "Whether it's Baidu APP, Wenxin Yiyan, or Baidu Wenku, each platform is experiencing new traffic growth, more sustainable resource improvements, and enhanced monetization efficiency for businesses."

During the World Conference, He Junjie also revealed that a massive number of users are actively utilizing AI features daily, with significant improvements in DAU, MAU, usage duration, payment rates, as well as creator monetization and merchant ROI.

After launching a series of popular C-end applications like the new Wenku, new search, and Wenxin Yiyan APP, He Junjie has expanded the "house (dealer)" concept to the entire AI-native ecosystem.

Transitioning from a "tree planter" to a "forest cultivator," He Junjie aims to leverage the dual drivers of content and commercial ecosystems to create fertile ground for external enterprises to develop AI-native applications within Baidu's mobile ecosystem. Together, they aim to build an "ecological rainforest" in the era of large models, using application platforms to bring AI down from its "pedestal of technology" and into every household.

Robin Li once said, "In the era of large models, the biggest entrepreneurial opportunities lie in applications. During the mobile internet era, there were only a few operating systems, with the most successful being applications like WeChat, Douyin, and Taobao. In the next decade, opportunities ten times more valuable than these may emerge."

The primary birthplace of these tenfold opportunities is not the large models themselves but the applications developed based on them.

With content and commercial ecosystems well-prepared, Baidu has launched the plugin ecosystem—Lingjing Matrix—based on its ERNIE capabilities, aiming to attract more developers and entrepreneurs to join, shifting from 'planting trees' to 'cultivating forests'.

Robin Li mentioned that plugins are a special type of AI-native application with the lowest barrier to entry and are the easiest to get started with. Plugins make large models more practical and user-friendly.

The increasing emergence of plugin applications will be a crucial measure to address users' personalized and complex needs in the era of large models, becoming a key component of Baidu's strategy to build a large-model application ecosystem.

For example, when writing an analysis article on the 'new energy vehicle battery supply chain,' the first step is to query relevant company financial reports. The 'Financial Smart Assistant' plugin can analyze financial reports and summarize them into data charts, helping users quickly understand the company's profit distribution, assets and liabilities, cash flow, and even provide industry rankings as references.

Recently, Baidu partnered with CITIC Books to launch the 'CITIC Books AI Reading Assistant' plugin. When users search 'Who invented the light bulb?' on Baidu, the search engine directly calls the plugin, providing a reliable answer from the book The Temperature of History 1. If users are interested in the book, they can also place an order directly through the plugin, marking the beginning of a new digital marketing model within the plugin ecosystem.

Alternatively, when users shop online and find that the delivered product does not match the seller's description, they can use the 'Legal Smart Assistant' plugin provided by the China Judicial Big Data Institute to receive professional advice on how to claim compensation.

"If large models are intelligent brains, then plugins are their hands and feet, enabling them to 'read thousands of books and travel thousands of miles'," said He Junjie.

Notably, within just one month of launch, the Lingjing Matrix plugin ecosystem platform has received registration applications from 27,000 developers, with individual developers accounting for 30%. These developers will gradually become an important force in enriching Baidu's mobile ecosystem applications.

Baidu's mobile ecosystem, which connects over 1 billion users, combined with the technological appeal of its ERNIE 4.0 model, serves as a strong incentive for external developers to participate in Baidu's plugin ecosystem. Additionally, Baidu has launched the 'Starlight Program', offering support measures including billions in traffic exposure, millions in computing power, and hundreds of millions in funding across all scenarios to encourage more partners to co-build the plugin ecosystem.

This low-barrier, high-traffic environment for plugin ecosystem development can indeed attract more developers to join and collaborate. Baidu has thus obtained the 'key' to enriching its large model ecosystem.

More importantly, with more data feedback from plugin applications and user feedback from Baidu's AI-native mobile ecosystem applications, the capabilities of the ERNIE model can be further iterated. This will in turn attract more users to Baidu's AI-native mobile applications and more external developers to join the plugin ecosystem. Baidu's mobile ecosystem is building a virtuous cycle growth flywheel on its path to creating a collaborative ecological rainforest.

From advocating the reconstruction of the entire mobile ecosystem with AI-native thinking, to launching popular applications and features like the ERNIE Bot app, new Wenku, and new search - all based on large model capabilities - and now proactively building an AI-native application ecosystem, Baidu's mobile ecosystem has become Robin Li's important battleground for reconstructing the C-end domain with large models, exploring new traffic inlets, and creating new commercial value.

The rapid generational transformation of the mobile ecosystem in the era of large models owes much to the real-time navigation of leaders like He Junjie, who have steered this massive ship through its course change.

With the arrival of the large model wave, one concept He Junjie repeatedly emphasized to the entire mobile ecosystem team early this year was seizing this AI opportunity. He advocated for a comprehensive reconstruction of Baidu's mobile ecosystem—spanning the user side, commercial side, content production side, and alliance side—to make MEG, this 'old tree,' bloom with 'new flowers.'

Accelerating application deployment has been placed at the core by He Junjie. In the process of drastically restructuring existing internal applications, He Junjie demanded that everyone cultivate an AI-native mindset while simultaneously launching a competition mechanism, requiring all business units to overhaul their operations and capabilities based on large models.

What’s impressive is that even many long-standing, stable products have shown remarkable performance during this restructuring. For example, Baidu Input Method has leveraged large models to introduce intelligent cloud predictions and associations, along with the all-scenario AI creation assistant 'Super Writer,' offering features like high-EQ communication and inspiration notes—essentially bringing generative AI capabilities right to users' fingertips.

The prioritization from top leadership has propelled Baidu's mobile ecosystem to become one of the first among major tech giants to wage the application battle, and under this sense of urgency, it has stepped ahead into the latter half of the large model application ecosystem competition.

This is also a dual race around technology and speed. Whoever launches more application products first and attracts more users to participate and co-create will likely gain an early advantage through real feedback from ecosystem users, gradually forming an increasingly strong Matthew effect. As He Junjie put it, 'Only when more users apply and invoke this model, accumulating industry know-how and proprietary data, will the model grow stronger and outperform others.'

From a single tree to an entire rainforest, it requires the platform’s cultivation and nurturing. For Baidu, which is building an AI-native application ecosystem, the new challenge facing He Junjie and his team is how to attract a larger user base with better tool products as quickly as possible.

Fortunately, Baidu's mobile ecosystem is not short of 'early adopters.' According to data released at Baidu World 2023, millions of professionals and students have already begun using the new features of Baidu Wenku for research and office tasks, while hundreds of thousands of Baidu Marketing clients are leveraging 'Qingge' to automatically generate and place ads. Additionally, tens of thousands of developers are creating specialized plugins on the Lingjing Matrix. Baidu's strategy to reshape the mobile ecosystem has now extended to 1 billion users and millions of clients and partners, fostering industry-wide collaboration to build a robust ecosystem.

In the 'Crossing the Chasm' theory, there exists a significant 'chasm' between the early market and the mainstream market for high-tech products or innovative services. Successfully crossing this chasm and gaining the support of pragmatists is crucial for determining the success or failure of the product or service.

Currently, the responsibility of bridging this chasm falls on Baidu's mobile ecosystem.