陈奕强：中美人工智能大语言模型之争

在快速变化的人工智能世界中，美国和中国正处于一场超越科技的激烈竞争。这种竞争也涉及经济实力、伦理课题和全球影响力。在马来西亚推动数码转型过程中，了解这场人工智能竞争可以为我们的技术性目标提供重要的洞见。

OpenAI的ChatGPT (GPT-4) 和Anthropic的Claude 3.5 Sonnet等美国人工智能模型目前在效能方面处于领先地位。他们在MMLU（大规模多任务语言理解）和HumanEval等测验中得分很高，分别达到84.5%和83.7%，这显示了他们强大的语言能力和程式编码能力。这些成就也反映了美国坚实的研发体系，注重高品质和安全性。

DeepSeek-V3等中国人工智能模型也取得了不错进展，但采取了不同的方法。他们的培训成本较低（约550万美元／2457万令吉），培训时间较短，约6个月。尽管如此，他们仍然取得了稳定的表现，MMLU和HumanEval测试分数分别为82.3%和83%。这种效率显示中国如何善用资源，让先进的人工智能变得更便宜。

中美人工智能模型的优点和应用反映了两国不同的焦点。美国模型用途广泛，支援对话式人工智能、API（应用程式编程接口）整合以及聊天机器人、内容创建和研究分析等行业的各种应用。

相较之下，中国模式则专注于多语言能力、客户服务、电子商务和自然语言处理（NLP）。他们越来越多地用于重视人工智能伦理和成本效率的市场。

中美之间的人工智能市场动态也受到两国经济战略的影响。美国公司往往成本较高，但强调品质和安全。美国在Open AI API等生态系统以及与Meta等公司的合作伙伴关系的支持下拥有强大的全球影响力。

另一方面，阿里巴巴等中国公司大幅降低了大型语言模型（LLM）的价格，使广泛的受众更容易接触到先进的人工智能。这项策略帮助他们巩固了在亚洲的地位，同时扩大了全球影响力。

这场人工智能竞争带来的经济影响是深远的。预计到2030年，美国人工智能模型将为全球经济贡献约13兆美元／58兆令吉，这主要得益于对生成式人工智能新创公司和基础设施的投资，这将创造许多就业机会。

此外，中国模式则是透过降低成本使人工智能技术在中小企业（SME）更普及，从而支援智能制造和自动化工作。

从伦理角度来看，美国和中国的模式都在进步。美国模型著重持续更新、解决偏见问题以及遵守欧盟《一般数据保护条例》（GDPR）和美国《加州消费者隐私法》(CCPA) 等法规。至于，中国模式的目标则是在当地法律范围内实现伦理实践，同时也随著时间的推移而不断改进。

寻找大马策略定位

这场竞争的地缘政治面向也至关重要。尽管美国对先进人工智能技术实施出口管制以阻碍中国的进步，但中国企业找到了绕过这些限制的创新方法继续取得进展。这种适应性使中国能够专注于亚洲、非洲和中东等地区，以更快地推动这些地区采用人工智能技术。

展望未来，美国人工智能模型可能会扩展其API（应用程式编程接口）功能，同时在研究进展中优先考虑安全和伦理问题。反之，中国模型将加强多语言支持，以增强在亚洲市场的影响力。

在马来西亚踏上数码转型之旅时，观察美国和中国在人工智能领域持续的竞争，可为我们的未来提供宝贵的经验教训。美国在绩效指标方面处于领先地位，而中国在成本效率方面表现出色，这对我们的中小企业来说都是至关重要的因素。

通过了解这些动态，马来西亚可以寻找本身策略定位，利用人工智能促进经济成长，同时促进造福社会每个人的伦理创新。在这场争夺全球技术领先地位的高风险竞争中，成功不仅取决于技术技能，还取决于为所有相关国家创造包容性、伦理和经济效益的人工智能解决方案。

陈奕强〈中美人工智能大语言模型之争》原文：China US AI large language model fight

In the fast-changing world of artificial intelligence (AI), the United States and China are in a strong competition that goes beyond just technology. This rivalry also involves economic power, ethical issues, and global influence. As Malaysia moves forward with its digital transformation, understanding this AI race can provide important insights for our own technological goals.

U.S. AI models like OpenAI's ChatGPT (GPT-4) and Anthropic's Claude 3.5 Sonnet are currently the leaders in performance. They score high in tests such as MMLU and HumanEval, achieving 84.5% and 83.7% respectively, which shows their strong language skills and ability to generate code. This success reflects the solid research and development systems in the U.S., which focus on high quality and security.

Chinese AI models, such as DeepSeek-V3, are also making progress but take a different approach. They have lower training costs (around $5.5 million) and shorter training times of about 6 months. Despite this, they still achieve solid performance with MMLU and HumanEval scores of 82.3% and 83%. This efficiency shows how China is using resources wisely to make advanced AI more affordable.

The strengths and applications of AI models from the U.S. and China reflect their different priorities. U.S. models are versatile, supporting conversational AI, API integration, and various applications in industries like chatbots, content creation, and research analytics.

In contrast, Chinese models focus on multilingual capabilities, customer service, e-commerce, and natural language processing (NLP). They are increasingly used in markets that value ethical AI and cost efficiency.

The market dynamics between the U.S. and China in AI are influenced by their economic strategies. U.S. firms tend to have higher costs but emphasize quality and security. They have a strong global presence supported by ecosystems like OpenAI API and partnerships with companies like Meta.

On the other hand, Chinese companies like Alibaba have significantly lowered prices for large language models (LLMs), making advanced AI more accessible to a wider audience. This strategy has helped them strengthen their position in Asia while expanding their influence worldwide.

The economic impact of this AI race is significant. U.S. AI models are expected to contribute around $13 trillion to the global economy by 2030, driven by investments in generative AI startups and infrastructure that will create many jobs.

Chinese models are democratizing access to AI for small and medium-sized enterprises (SMEs) by lowering costs, which supports smart manufacturing and automation efforts.

Ethically, both U.S. and Chinese models are making progress. U.S. models focus on continuous updates, addressing bias issues, and complying with regulations like GDPR and CCPA. Meanwhile, Chinese models aim for ethical practices within local laws while also improving over time.

The geopolitical aspects of this competition are crucial. Although the U.S. has imposed export controls on advanced AI technologies to limit Chinese progress, Chinese firms continue to advance by finding innovative ways around these restrictions. This adaptability allows China to focus on regions like Asia, Africa, and the Middle East for faster AI adoption.

Looking ahead, U.S. AI models will likely expand their API capabilities while prioritizing safety and ethics in research advancements. Meanwhile, Chinese models will enhance multilingual support to strengthen their influence in Asian markets.

As Malaysia navigates its digital transformation journey, observing the ongoing rivalry between the U.S. and China in artificial intelligence offers valuable lessons for our future. The U.S. leads in performance metrics while China excels in cost efficiency—both vital factors for our SMEs.

By understanding these dynamics, Malaysia can strategically position itself to leverage AI for economic growth while promoting ethical innovation that benefits everyone in society. In this high-stakes race for global leadership in technology, success will not only depend on technical skills but also on creating inclusive, ethical, and economically beneficial AI solutions for all nations involved.