UPDATE — OCTOBER 2025: Since the release of the CSET report on China’s critiques of large language models (LLMs), Beijing has deepened its commitment to pursuing diversified pathways toward general artificial intelligence (GAI). In mid-2025, the Ministry of Science and Technology (MOST) expanded its Brain-Inspired Intelligence and Hybrid AI initiative, creating new funding lines for “biologically plausible learning” and “machine consciousness frameworks.” At the city level, Beijing’s June 2025 AI Plan added a goal to develop “embodied cognitive agents” capable of both physical interaction and abstract reasoning, reflecting the expert critiques of LLM limitations. The Chinese Academy of Sciences (CAS) also launched the General Intelligence Collaboration Network (GICN), linking neuroscience, robotics, and symbolic reasoning labs under a unified framework.

Chinese academic leaders have advanced these priorities. Zhu Songchun presented a prototype “agentic cognition” framework that integrates symbolic reasoning with visual perception, while Tang Jie at Tsinghua University released hybrid symbolic–neural datasets and architectures as open-access alternatives to purely transformer-based designs. Meanwhile, Zhang Yaqin’s AI Industry Research Institute partnered with the Beijing Academy of Artificial Intelligence (BAAI) on studies that tie embodied learning to robotics and digital twin simulations.

China’s tech firms continue to scale LLMs but are shifting toward hybrid architectures. iFlytek’s Spark Cognitive Large Model 3.0, launched in April 2025, integrates speech, vision, and motor feedback, illustrating the state-backed push for multimodal, embodied cognition. New AI governance rules implemented in May 2025 now mandate national-lab testing before companies can claim “general cognition” capabilities, signaling tighter oversight and curbs on LLM hype.

Globally, China’s approach is influencing other regions. CSET’s follow-up briefs noted that DARPA in the U.S. and Horizon Europe have introduced funding streams for neuromorphic computing and cognitive AI, echoing aspects of China’s diversified strategy. Analysts increasingly describe China’s vision for GAI as one of multi-system hybridization—fusing LLMs with neuromorphic hardware, symbolic reasoning, embodied AI, and state-defined ethical frameworks into a cohesive ecosystem.

Would you like me to prepare a short comparative update that shows how China’s multipath AI strategy contrasts with the U.S. and EU approaches as of late 2025? That could make the global divergence clearer for readers.

ORIGINAL NEWS STORY:

China Challenges Large Language Models as the Sole Path to General AI

A new report from the Center for Security and Emerging Technology (CSET) sheds light on China’s critical stance toward large language models (LLMs) as the dominant approach to achieving general artificial intelligence (GAI). While the United States and Europe have heavily invested in LLMs, such as OpenAI’s GPT models and Google’s Gemini, China is pursuing a diversified AI strategy that incorporates alternative pathways, including brain-inspired AI and embodied intelligence.

The report, titled “Chinese Critiques of Large Language Models: Finding the Path to General Artificial Intelligence,” compiles statements from China’s top AI researchers, government strategies, and academic publications, revealing skepticism over LLMs’ ability to reach human-like reasoning and problem-solving capabilities.

China’s AI development differs from Western approaches by avoiding an overreliance on a single technological pathway. While Chinese companies and institutions have developed LLMs similar to ChatGPT, Chinese scientists have raised concerns about their inherent limitations, including their high computational costs, reliance on vast amounts of training data, and persistent issues such as hallucinations and lack of reasoning abilities.

Chinese researchers argue that GAI requires more than just scaling up LLMs. Instead, they emphasize integrating brain-inspired architectures, real-world sensory experiences, and a stronger emphasis on knowledge graphs and symbolic reasoning. Many top AI institutions in China, including the Beijing Academy of Artificial Intelligence (BAAI) and the Chinese Academy of Sciences, are actively exploring these alternative approaches.

Several leading general AI experts in China have publicly criticized the limitations of LLMs and called for more diversified research efforts. Among them is Tang Jie, a professor at Tsinghua University and a key figure in China’s AI community. He has argued that LLMs, while powerful, do not sufficiently replicate the cognitive mechanisms of the human brain. Instead, he advocates for a deeper exploration of biologically inspired AI.

Similarly, Zhang Yaqin, former president of Baidu and founding dean of Tsinghua’s AI Industry Research Institute, has pointed out LLMs’ inability to interact with the physical world effectively. He suggests that future AI systems must integrate principles from the physical sciences and knowledge graphs to bridge this gap.

Another major critic, Zhu Songchun, director of the Beijing Institute for General Artificial Intelligence, has outright dismissed LLMs as a viable path to GAI. He argues that simply increasing the scale of current models will not achieve true intelligence and that AI development must shift toward new paradigms rooted in cognitive science.

The Chinese government has aligned its AI strategy with these expert critiques, calling for a more comprehensive approach to GAI. In May 2023, Beijing’s municipal government issued a directive promoting research into “new paths” for general AI, including brain-inspired intelligence and embodied learning. The directive emphasized that future AI systems should be capable of autonomous decision-making, interacting with dynamic environments, and operating under a unified theoretical framework.

At the national level, China’s Ministry of Science and Technology has also pushed for a multi-pronged AI development strategy. A speech in March 2024 by Wu Zhaohui, vice president of the Chinese Academy of Sciences, underscored the need for “synergy between large and small models” and called for research into embodied intelligence, group intelligence, and human-machine hybrid intelligence.

China’s emphasis on diverse AI research approaches could give it a strategic advantage in the race to GAI. While Western companies remain heavily invested in LLMs, China’s broader research portfolio ensures that it is not locked into a single AI development path. If LLMs ultimately fail to achieve GAI, China’s investment in alternative models could place it ahead in the long-term AI competition.

Moreover, China’s AI research is increasingly tied to national security and governance priorities. Unlike Western AI development, which is largely driven by private sector initiatives, China’s AI strategy is state-led, ensuring alignment with broader economic and strategic goals. This includes efforts to embed state-approved values into AI models, a concept championed by Zhu Songchun and other Chinese AI leaders.

The CSET report warns that while Western AI development is fixated on scaling up existing LLMs, China is methodically pursuing a broader research agenda. If LLMs prove insufficient for achieving GAI, China’s approach may offer a more sustainable and adaptable roadmap.

The findings of the CSET report highlight a critical divergence in global AI strategy. While Western AI research remains dominated by LLMs, China is positioning itself for long-term success by investing in a wider range of technologies. If China’s diversified approach proves successful, it could shift the balance of AI leadership and redefine the future of artificial intelligence on a global scale.

Need Help?

If you have questions or concerns about China’s General AI stance, or any global guidelines, regulations and laws, don’t hesitate to reach out to BABL AI. Their Audit Experts can offer valuable insight, and ensure you’re informed and compliant.