Completing the Business Loop for Agents

Written by Chen Dengxin

Edited by Li Ji

Typeset by Annalee

The Wenxin Large Language Model Empowers Entrepreneurs with Dream Wings.

Since the launch of the second season of Baidu's 'Wenxin Cup' Startup Competition on April 16, 2024, nearly 1,600 domestic and international projects have fiercely competed in just a few months, representing a 60% year-on-year increase.

This signifies that Baidu's 'Wenxin Cup' Startup Competition has gained widespread recognition in the startup community.

On September 25, at the 2024 Baidu Cloud Intelligence Conference, the competition results were officially announced. Eight outstanding startup teams received millions of yuan in "cash + resources" investments from Baidu, while 13 exceptional university teams were awarded nearly one million yuan in cash prizes.

These winning teams have emerged from the cream of entrepreneurs across various industries, becoming the 'dark horses' of China's AI-native applications and agents, with the potential to evolve into AI-era 'unicorns' in the future.

Robin Li, Founder, Chairman, and CEO of Baidu, stated, 'Baidu hopes to work with all entrepreneurs to build a thriving ecosystem for AI-native applications and agents, helping entrepreneurs find their Product-Market Fit (PMF) earlier and create super apps for the AI era.'

Undoubtedly, Baidu has accelerated its deployment of large language models into practical applications.

AI-native Applications Bloom, Reshaping Industries

Entrepreneurs have always been the 'pioneers of the times' driving technological progress.

Only when entrepreneurs continuously 'roll out' applications can large language models thrive in an open and prosperous ecosystem. After all, 'In the AI-native era, we need millions of AI-native applications, but not a hundred large language models.'

The Wenxin Large Language Model is the Best Choice for Startups

To this end, Baidu provides comprehensive support in technology, funding, and resources to explore AI-native applications and industry innovations. The 'Wenxin Cup' Startup Competition emerged as a result, becoming the highest-profile and most substantial startup competition in China's large language model field.

Compared to the first season, the second season of the 'Wenxin Cup' Startup Competition boasts three notable changes.

First, the proportion of application projects has increased.

More and more entrepreneurs recognize that foundational models cannot cater to all applications. Coupled with the limited resources and energy of startup teams, it's unnecessary to develop both models and applications simultaneously. Focusing solely on applications often yields higher efficiency.

This consensus is reflected in the increased proportion of large language model application projects from 80% in 2023 to over 90% in 2024 in the current 'Wenxin Cup' Startup Competition.

In fact, Baidu has blazed a new trail for large language models by seeking to 'empower everything,' actively supporting, incubating, and encouraging third-party AI-native applications and agents, thereby continually expanding its core capabilities.

Second, the entrepreneurial directions have become more diverse.

While approximately 30% of participants in 2023 focused on general office and marketing, the 2024 competition showcased a myriad of directions, including entertainment, e-commerce, marketing, healthcare, office work, hardware, and enterprise services.

For example, the finance industry, with its dual attributes of technology- and data-intensity, naturally aligns with large language models. Wenzhi Technology's AI-Native financial investment research platform offers functions such as intelligent stock recommendations, opinion aggregation, and technical analysis.

Chen Fei, CEO of Wenzhi Technology, told Znker, 'The investment market is rife with rumors, many of which may differ from reality. We can bridge the information gap for users, achieving AI equity, akin to the phrase 'The swallows that once nested in the halls of the rich now fly into ordinary households.'"

Third, a new university group competition has been introduced.

The university group attracted 631 entries from 300 universities, with 13 outstanding teams selected to showcase their entrepreneurial prowess on the final stage.

Among the finalists, Aiuni AI stood out prominently.

With academic backgrounds from Tsinghua University's Yao Class, Aiuni AI leverages Wenxin Large Language Model-based AI technology combined with CG art to generate 4D content, revolutionizing traditional video creation and earning the title of the Chinese version of LUMA.

It is evident that Baidu is sparing no effort to promote the 'blossoming' of AI-native applications, while numerous outstanding entrepreneurs are fully committed to this generational entrepreneurial opportunity. This two-way endeavor has sparked a dazzling synergy.

Cultivating New Horizons Around Agent Entrepreneurship

Among these, agents are the most favored direction for AI-native applications.

Robin Li once stated, 'As foundational models become increasingly powerful, developing applications becomes simpler. Among them, agents are the simplest. By clearly articulating workflows in plain language and pairing them with specialized knowledge bases, valuable agents can be created, easier than building a webpage in the Internet era.'

As such, low-barrier agents present an entrepreneurial opportunity accessible and creatable by all.

Especially as killer AI-native applications have yet to emerge, as more people build agents upon large language models, once someone paves the way, the 'WeChat' and 'TikTok' of the AI era could emerge.

Against this backdrop, agents have gained widespread recognition among entrepreneurs.

'We fully embrace Baidu's concept of agents,' said Chen Fei. AI-Native financial investment research platforms, including features like 'Should I Buy' and 'Five Daily Stock Picks,' are all agents. 'In practice, we've found agents highly suitable for financial scenarios, as they can decompose complex financial problems into manageable tasks for agents to solve.'

More importantly, whereas code-based application development was once inaccessible to those unfamiliar with coding, now agents can be developed with little to no coding required.

Baidu has introduced AgentBuilder and AppBuilder, tools for creating agents. AgentBuilder has a lower barrier to entry, while AppBuilder offers more powerful features, catering to different developer groups. 'AgentBuilder's toolbox is like a stationery box or jewelry box, where each tool is readily accessible. Developers no longer need to write code from scratch.'

AppBuilder Architecture

In fact, over 60% of participants in the 'Wenxin Cup' focused on agent applications, and over 30% had no professional programmers on their teams.

Jingxiang Technology is a prime example, focusing on agents. Leveraging the Wenxin Large Language Model, they effortlessly created AI evaluators for digital human video consultations, AI counselors for individuals with mild psychological sub-health, and AI psychologists for mild to moderate patients.

Huang Li, Founder and CEO of Jingxiang Technology, stated, 'Users simply answer a series of questions on their phones, and the AI agent quickly analyzes the data to provide personalized psychological counseling and consultation solutions.'

It's worth noting that Baidu has become synonymous with agents.

On September 12, 2024, the Baidu Alliance Conference announced that Baidu had established a closed-loop agent ecosystem with a daily distribution exceeding 10 million. Currently, over 16,000 merchants have developed agents on Baidu, covering industries such as education, law, and B2B.

In other words, whether on the consumer or business side, agents developed based on the Wenxin Large Language Model are increasingly accepted by users and merchants alike, fostering a thriving agent ecosystem.

In the Agent Era, Why Baidu Leads the Way

There are three reasons why Baidu leads the charge in the agent era and stays ahead of the curve.

First, the continuous progress of the Wenxin Large Language Model.

Baidu is the only company in China with a full-stack AI technology architecture, boasting key self-developed technologies across chip, framework, model, and application layers. This seamless collaboration between layers is an unparalleled advantage.

Baidu is the only company in China with a full-stack AI technology architecture

In short, Baidu has Kunlun AI Chip, PaddlePaddle Deep Learning Framework, Baidu Intelligent Cloud for computing power, AgentBuilder and AppBuilder for agent development, and ModelBuilder for enterprise-level large model platforms.

Baidu Intelligent Cloud was the first to pass the large model platform application system certification

Through continuous accumulation of full-stack AI technologies and iterative AI optimization mechanisms, the Wenxin Large Language Model has achieved sustained progress.

This is evident from its growth data. In Q2 2024, the Wenxin Large Language Model recorded over 600 million daily calls and processed over one trillion tokens, both topping domestic records.

With the launch of Wenxin 4.0 Turbo, training efficiency improved 5.1 times and inference efficiency soared 105 times compared to the initial Wenxin Ergnie version.

As such, the continually evolving and increasingly effective Wenxin Large Language Model has become the optimal 'soil' for AI-native applications and agents to take root, enabling entrepreneurs' innovative ideas to materialize seamlessly.

Second, constructing a closed-loop agent business model.

Merely successfully deploying agents is insufficient; it's akin to 'having a car with no road or a boat with no sea.' For agents to thrive, distribution channels are crucial.

Put simply, agents must be easy to develop, distribute, and monetize. Only by bridging the development, distribution, and monetization links can agents truly capture the public's imagination, accompanying them in work, study, life, and beyond, thereby reshaping industries.

Coincidentally, this is where Baidu excels.

In Q2 2024, Baidu App had 703 million monthly active users and processed billions of search queries daily, forming an astonishing natural traffic pool.

'Search is the primary gateway for agent distribution,' said Robin Li, citing college entrance exam volunteering as an example. 'During peak periods, Baidu's college entrance exam agent answered over 2 million student questions daily, out of a total of 10 million examinees, indicating a significant proportion of students utilized this agent.'

Moreover, Baidu Search launched an interactive channel in May 2024, opening up a new incremental field for agent distribution and further enhancing user reach.

Currently, Baidu Search distributes over 10 million agents daily.

Thus, centered on search and complemented by advertising services from Baidu Alliance, Baidu has successfully constructed a complete 'development + distribution + monetization' business loop, fostering ecological collaboration, shared interests, and mutual benefits.

Third, a pioneering and steadfast strategic vision for agents.

In a September 2024 internal speech, Robin Li revealed, 'To date, agents are not a consensus. Few companies, like Baidu, view agents as the most crucial strategy and development direction for large language models.'

In essence, many large language model players haven't recognized the importance of agents, underscoring Baidu's far-sighted strategy and down-to-earth approach.

Indeed, Baidu was an early mover in the agent space.

As early as September 2023, Baidu launched the 'Lingjing Matrix' Wenxin Ergnie plugin ecosystem platform, which evolved into the 'Lingjing Matrix Agent Platform' and then the 'Wenxin Agent Platform' at the 2024 Create Conference.

Clearly, Baidu continues to invest heavily in agents.

Retrospectively, Baidu has consistently demonstrated forward-thinking strategic vision, investing consistently in AI, positioning it as a pioneer, evangelist, and leader in China's AI landscape.

By the first half of 2024, Baidu had invested nearly 160 billion yuan in R&D, with Q2 2024 R&D expenses amounting to 5.9 billion yuan, accounting for 17.4% of revenue, underscoring its R&D-driven identity.

Baidu Q2 2024 Results

This massive and long-term investment has fortified Baidu's strategy to empower agents.

According to the 'Baidu AI Innovation and Patent White Paper 2024,' by the end of 2023, Baidu had filed 19,308 AI-related patent applications and received 9,260 authorizations, ranking first domestically for six consecutive years.

In conclusion, while the second season of Baidu's 'Wenxin Cup' Startup Competition has concluded successfully, it marks just the beginning of Baidu's collaboration with entrepreneurs to build a thriving ecosystem for China's AI-native applications and agents.

In the future, Priority selection Wenxin Large Language Model will gradually become the consensus among entrepreneurs. Standing on the shoulders of giants, it will be easier to create blockbuster applications, while Baidu's large model ecosystem will continue to infuse fresh blood, ushering in Baidu's 'iPhone moment.'

Thus, both Baidu and entrepreneurs have much to look forward to.