Li’s Critique of DeepSeek
During his address at the conference, Robin Li articulated that DeepSeek’s functionalities were essentially restricted to text processing, conspicuously lacking the capability to comprehend and generate image and audio data. He asserted that this fundamental limitation severely constrained DeepSeek’s versatility and applicability across a spectrum of diverse use cases, thus undermining its potential as a comprehensive AI solution. Li further underscored that DeepSeek exhibited a notably elevated hallucination rate, which inevitably eroded user confidence due to the generation of inaccurate and unreliable results. This propensity for generating flawed outputs, according to Li, presented a significant impediment to user adoption, as it made it exceedingly difficult for users to repose trust in the model as a dependable and authoritative resource. The high incidence of hallucinations, in his assessment, rendered DeepSeek less suitable for mission-critical applications where accuracy and reliability were paramount.
Baidu’s Counter-Offerings
In stark contrast to the perceived shortcomings of DeepSeek, Robin Li strategically showcased Baidu’s latest and most advanced AI models, specifically the Wenxin Large Model 4.5 Turbo and the Deep Thought Model X1 Turbo. He vehemently asserted that these cutting-edge models possessed superior capabilities, including the ability to seamlessly support multi-modal inputs, a feature conspicuously absent in DeepSeek. Furthermore, Li emphasized the robust logical reasoning capabilities embedded within Baidu’s models, enabling them to process information with greater precision and derive more accurate conclusions. Perhaps most compellingly, Li highlighted the significant cost advantages associated with Baidu’s offerings. He revealed that the Wenxin 4.5 Turbo was priced at a mere 40% of the DeepSeek – V3 model, while the X1 Turbo was an astonishingly low 25% of the DeepSeek – R1 price. By offering such competitively priced alternatives, Li expressed his sincere hope that developers would be empowered to substantially reduce their deployment expenses and foster the widespread practical application of AI technology across various industries and sectors. He envisioned a future where Baidu’s cost-effective models would serve as a catalyst for innovation, enabling developers to experiment and build novel AI-powered solutions without being constrained by exorbitant costs.
Public Skepticism
However, Robin Li’s criticisms and self-proclaimed technological supremacy did not universally resonate with the public. A significant segment of online users voiced their skepticism, with some even accusing Baidu of exhibiting signs of “sour grapes.” One commenter sarcastically remarked, “Didn’t you just integrate DeepSeek? Now you’re criticizing it,” implying a perceived inconsistency in Baidu’s stance. Critics further suggested that Baidu lacked a degree of self-awareness, emphasizing that the company’s energies would be better spent on improving its own products rather than engaging in blame games with its competitors. The prevailing sentiment among these critics was that Baidu’s focus should be on addressing its internal challenges and enhancing its own capabilities, rather than attempting to undermine the perceived strengths of its rivals. This skepticism highlighted the importance of demonstrable results and tangible improvements in product quality, as opposed to mere rhetoric and pronouncements of superiority.
Li’s Emphasis on Applications and Cost Reform
Despite the discernible pushback from certain quarters, Robin Li remained steadfast in his conviction that the true value of AI lay not merely in the sophistication of the models themselves, but rather in their ability to be effectively applied to real-world problems and use cases. He posited that even the most advanced and computationally powerful AI models were rendered essentially worthless if they lacked the capacity to support practical applications that delivered tangible benefits to users. In Li’s perspective, applications constituted the very core of the future AI ecosystem, and lowering the barriers to entry, particularly those related to cost, was absolutely crucial for fostering innovation among both developers and enterprises. He argued that by making AI technology more accessible and affordable, a greater number of individuals and organizations would be empowered to experiment, innovate, and create novel solutions that could transform industries and improve lives. This emphasis on practical applications and cost-effectiveness underscored the importance of democratizing AI technology and making it available to a wider audience.
Baidu’s AI History and Missed Opportunities
Baidu embarked on its journey into the realm of large language models in early 2023 with the highly anticipated launch of Wenxin Yiyan, which was widely hailed as one of China’s pioneering generative AI models. However, the company’s initial charging strategy, which was perceived by many as being overly restrictive and expensive, inadvertently hindered its growth and adoption. Baidu’s monthly subscription fee of 49.9 yuan, while intended to attract high-quality users, ultimately failed to achieve the necessary scale to compete effectively in the rapidly evolving AI landscape. As a consequence, Wenxin Yiyan’s monthly active users (MAU) stagnated, preventing the model from gaining the widespread adoption and critical mass necessary for sustained success. In stark contrast, DeepSeek adopted a radically different approach, embracing an open-source, free model that quickly attracted a large and vibrant community of developers. This strategic decision proved to be highly successful, as DeepSeek rapidly amassed a significant user base, fueling its growth and technological advancement. By January 2024, DeepSeek’s MAU had reached an impressive 33.7 million, significantly outpacing Wenxin Yiyan’s 13.05 million. This vast user base facilitated a virtuous cycle of data collection, feedback, and iterative improvement, enabling DeepSeek to continuously enhance its model and stay ahead of the competition.
Cost and Performance Advantages of DeepSeek
DeepSeek has strategically leveraged its innovative Mixture of Experts (MoE) architecture and FP8 quantization technology to achieve significant reductions in inference costs, bringing them down to a mere 30% of the industry average. This remarkable cost efficiency has made DeepSeek an exceptionally attractive option for enterprises and developers seeking to deploy AI solutions without incurring exorbitant expenses. Multiple enterprise tests have consistently demonstrated that using Baidu Smart Cloud’s Qianfan platform incurs an average cost of 1.2 yuan per 10,000 requests, with a single conversation costing approximately 0.8 yuan. In contrast, DeepSeek’s model costs only 0.25 yuan, representing a substantial cost savings that significantly reduces budget expenses and labor costs by over 40%. This significant cost advantage has further solidified DeepSeek’s position as a leading player in the AI market, attracting a growing number of users and enterprises seeking to maximize their return on investment.
Wenxin Yiyan’s Early Setback and DeepSeek’s Rise Through Open Source
Baidu initially had a golden opportunity to collaborate with Apple, the world’s most valuable technology company, to integrate Wenxin Yiyan into iPhones sold in China, which would have significantly expanded its influence and reach. However, this potentially transformative collaboration ultimately fell through due to a confluence of privacy and technical concerns, leading Apple to choose Alibaba’s Qianwen model instead. This setback proved to be a significant blow to Baidu, causing its market value to plummet by over 10 billion yuan in a single day, underscoring the critical importance of effective technology implementation and seamless commercial execution. The failed collaboration with Apple served as a stark reminder that even the most advanced technology can fail to achieve its full potential without a robust and well-executed commercial strategy. DeepSeek, on the other hand, capitalized on its open-source approach to gain traction and establish itself as a major player in the AI landscape.
Baidu’s Strategic Shift
Faced with increasing open-source pressure and intensifying price competition from DeepSeek and other emerging AI players, Baidu announced the open-sourcing of its Wenxin Large Model in early 2025, signaling a significant strategic shift. This decision marked a departure from Baidu’s previous proprietary approach and reflected a growing recognition of the importance of open collaboration and community-driven innovation in the AI space. During the conference, Robin Li stated that he hoped the new model would lower the barriers to using AI technology, promote application development, and foster ecosystem growth. This strategic shift towards open-source was widely seen as an attempt by Baidu to regain its competitive edge and revitalize its position in the rapidly evolving AI market.
Baidu’s Future in AI
Despite its early start in AI, Baidu has undeniably missed several key opportunities for market transformation, allowing competitors like DeepSeek to gain ground. The company now faces the imperative to abandon its short-sighted mindset and focus on strengthening its technology and product competitiveness. This requires a renewed commitment to innovation, a willingness to embrace open collaboration, and a relentless focus on delivering value to its users. Only by accelerating innovation and fostering a robust developer ecosystem can Baidu regain its leading position in the global AI landscape. The company must also address its past missteps, learn from its competitors, and adapt to the changing dynamics of the AI market.
The Core of the Debate: Cost, Performance, and Accessibility
The core of the “involution” spat between Baidu and DeepSeek, as it’s being called in China, fundamentally centers on three key factors: cost, performance, and accessibility. These three pillars are inextricably linked and collectively determine the overall competitiveness and market viability of any AI model. DeepSeek, with its pioneering open-source model and unwavering focus on efficiency, has seemingly gained a significant edge in the accessibility department. By making its model freely available for use and offering a cost-effective solution, DeepSeek has successfully attracted a large and vibrant developer base, fostering innovation at a grassroots level. This accessibility has enabled DeepSeek to democratize AI technology and empower a wider range of individuals and organizations to experiment and create novel solutions.
Baidu, conversely, initially adopted a more traditional, proprietary approach with Wenxin Yiyan. While the model may have possessed certain performance advantages over its open-source counterparts, its pay-to-use model inadvertently hindered its adoption and ultimately allowed DeepSeek to surpass it in terms of user base and market traction. This strategic decision highlighted the importance of accessibility in driving adoption and building a strong community around an AI model.
Robin Li’s criticisms of DeepSeek’s hallucination rate and limited multi-modal capabilities can be interpreted as attempts to highlight potential performance shortcomings. However, these criticisms have been largely overshadowed by concerns about Baidu’s perceived lack of competitiveness and its initial failure to embrace the open-source movement. The public perception is that Baidu focused too heavily on proprietary technology and failed to recognize the growing importance of open collaboration and community-driven innovation.
The Importance of a Strong Developer Ecosystem
The ultimate success of any AI model is inextricably linked to the strength and vibrancy of its developer ecosystem. A vibrant and engaged community of developers can contribute to the model’s continuous improvement, expand its capabilities, and create innovative applications that drive wider adoption. The developer ecosystem serves as a critical feedback loop, providing valuable insights and contributing to the ongoing evolution of the AI model.
DeepSeek’s open-source model has been instrumental in building a strong developer ecosystem. By providing developers with free access to its model and the tools to build upon it, DeepSeek has fostered a collaborative environment that encourages innovation and accelerates development. This collaborative environment has attracted a diverse range of developers, each contributing their unique skills and expertise to enhance the capabilities of the DeepSeek model.
Baidu’s initial closed-source approach with Wenxin Yiyan limited its ability to build a similar ecosystem. While the company has since embraced open-source, it faces the challenge of catching up to DeepSeek and other companies that have already established a strong foothold in the open-source community. Baidu must actively engage with developers, provide them with the resources and support they need, and foster a collaborative environment that encourages innovation.
The Shift Towards Multi-Modal AI
Another key aspect of the ongoing debate is the growing importance of multi-modal AI. Robin Li’s criticism of DeepSeek’s lack of support for images, audio, and video highlights the increasing demand for AI models that can seamlessly understand and process different types of data. Multi-modal AI represents a significant advancement in the field of artificial intelligence, enabling models to interact with the world in a more natural and intuitive way.
Multi-modal AI has the potential to unlock a wide range of new applications in diverse areas such as healthcare, education, and entertainment. For example, a multi-modal AI model could be used to analyze medical images, generate personalized learning experiences, or create immersive virtual reality environments. The ability to process and integrate different types of data opens up new possibilities for AI-powered solutions that can address complex real-world problems.
Baidu’s Wenxin Large Model 4.5 Turbo and Deep Thought Model X1 Turbo are specifically designed to address the growing need for multi-modal AI. By supporting multi-modal inputs and possessing strong logical reasoning capabilities, these models are strategically positioned to compete effectively in the rapidly evolving AI landscape. Baidu’s investment in multi-modal AI reflects a recognition of the importance of this technology in shaping the future of artificial intelligence.
The Broader Implications for China’s AI Industry
The “involution” spat between Baidu and DeepSeek has broader implications for China’s AI industry as a whole. It underscores the increasing competition in the Chinese AI market and the relentless pressure on companies to innovate and adapt to changing market dynamics. The Chinese AI market is one of the most dynamic and competitive in the world, with a large number of companies vying for market share.
The rise of DeepSeek and other open-source AI models has challenged the dominance of traditional players like Baidu. These new models are forcing companies to rethink their strategies and embrace new approaches to development and deployment. The open-source movement has democratized AI technology,empowering a wider range of individuals and organizations to participate in the development and deployment of AI solutions.
The Chinese government has made AI a strategic priority and is investing heavily in the development of the industry. This government support is fueling innovation and driving growth in the Chinese AI market. As the market continues to grow and mature, competition is likely to intensify further, creating new opportunities for both established companies and startups.
The Future of AI in China
The future of AI in China is undoubtedly bright. The country possesses a large and growing economy, a supportive government, and a vibrant ecosystem of researchers, developers, and entrepreneurs. These factors collectively create a favorable environment for the continued growth and development of the Chinese AI industry.
As AI technology continues to advance at an accelerating pace, it is poised to have a profound impact on all aspects of Chinese society. From healthcare and education to manufacturing and transportation, AI has the potential to improve efficiency, productivity, and overall quality of life. The widespread adoption of AI technology is expected to drive significant economic growth and improve the competitiveness of Chinese industries.
The spat between Baidu and DeepSeek serves as just one example of the intense competition that is driving innovation in the Chinese AI market. As companies continue to compete and collaborate, the Chinese AI industry is poised for continued growth and success, solidifying its position as a global leader in artificial intelligence.
The Global Race for AI Dominance
The events unfolding in China are also indicative of the larger global race for AI dominance. Countries around the world are investing heavily in AI research and development, recognizing its potential to transform their economies and societies. The development and deployment of AI technology are seen as critical for maintaining economic competitiveness and national security.
The United States, China, and Europe are the leading players in the global AI market. Each region has its own strengths and weaknesses, and the competition between them is likely to intensify in the coming years. The global race for AI dominance is not just about technological advancements, but also about attracting and retaining top talent, securing access to vast amounts of data, and establishing ethical guidelines for the responsible development and deployment of AI.
The global race for AI dominance is not just about technology. It is also about talent, data, and ethical considerations. Countries that can attract and retain top AI talent, gather and process large amounts of data, and develop ethical guidelines for the use of AI will be best positioned to succeed. These factors are critical for building a sustainable and responsible AI ecosystem.
The Ethical Considerations of AI
As AI technology becomes more powerful and pervasive, it is increasingly important to consider its ethical implications. AI has the potential to be used for both good and evil, and it is crucial to ensure that it is developed and deployed in a responsible and ethical manner. The ethical considerations of AI are becoming increasingly complex as the technology continues to advance.
Some of the key ethical considerations of AI include:
- Bias: AI models can perpetuate and amplify existing biases in data, leading to unfair or discriminatory outcomes. It is crucial to ensure that AI models are trained on diverse and representative datasets to mitigate the risk of bias.
- Privacy: AI models can collect and process vast amounts of personal data, raising concerns about privacy and security. Robust data protection measures and privacy-enhancing technologies are essential to safeguard individuals’ privacy.
- Transparency: AI models can be opaque and difficult to understand, making it challenging to ensure accountability and fairness. Explainable AI (XAI) techniques can help to improve the transparency and interpretability of AI models.
- Job displacement: AI models can automate tasks that are currently performed by humans, leading to job displacement and economic inequality. It is important to invest in education and training programs to help workers adapt to the changing job market.
It is essential to address these ethical concerns proactively and to develop regulations and guidelines that promote the responsible and ethical use of AI. Collaboration between governments, industry, and academia is crucial for developing a comprehensive framework for ethical AI development and deployment.