The (AI) Arms Race of Our Times.

The emergence of DeepSeek R1 as a competitive open-sourced model challenging U.S. frontier (closed-source) models such as OpenAI adds a new dimension to the AI arms race between the U.S. and China, particularly in the context of AI ethics and safety. Here’s how this development might impact the landscape:

Here’s a LinkedIn post by Andrew Ng (Founder of DeepLearning.AI; Managing General Partner of AI Fund; Exec Chairman of Landing AI):

The buzz over DeepSeek this week crystallised, for many people, a few important trends that have been happening in plain sight: (i) China is catching up to the U.S. in generative AI, with implications for the AI supply chain. (ii) Open weight models are commoditising the foundation-model layer, which creates opportunities for application builders. (iii) Scaling up isn’t the only path to AI progress. Despite the massive focus on and hype around processing power, algorithmic innovations are rapidly pushing down training costs.

About a week ago, DeepSeek, a company based in China, released DeepSeek-R1, a remarkable model whose performance on benchmarks is comparable to OpenAI’s o1. Further, it was released as an open weight model with a permissive MIT license. At Davos last week, I got a lot of questions about it from non-technical business leaders. And on Monday, the stock market saw a “DeepSeek selloff”: The share prices of Nvidia and a number of other U.S. tech companies plunged. (As of the time of writing, some have recovered somewhat.)

Here’s what I think DeepSeek has caused many people to realize:

China is catching up to the U.S. in generative AI. When ChatGPT was launched in November 2022, the U.S. was significantly ahead of China in generative AI. Impressions change slowly, and so even recently I heard friends in both the U.S. and China say they thought China was behind. But in reality, this gap has rapidly eroded over the past two years. With models from China such as Qwen (which my teams have used for months), Kimi, InternVL, and DeepSeek, China had clearly been closing the gap, and in areas such as video generation there were already moments where China seemed to be in the lead.

I’m thrilled that DeepSeek-R1 was released as an open weight model, with a technical report that shares many details. In contrast, a number of U.S. companies have pushed for regulation to stifle open source by hyping up hypothetical AI dangers such as human extinction. It is now clear that open source/open weight models are a key part of the AI supply chain: Many companies will use them. If the U.S. continues to stymie open source, China will come to dominate this part of the supply chain and many businesses will end up using models that reflect China’s values much more than America’s.

Open weight models are commoditizing the foundation-model layer. As I wrote previously, LLM token prices have been falling rapidly, and open weights have contributed to this trend and given developers more choice. OpenAI’s o1 costs $60 per million output tokens; DeepSeek R1 costs $2.19. This nearly 30x difference brought the trend of falling prices to the attention of many people.

In a Nutshell 🥜

Energy Efficiency in Training: DeepSeek’s R1 model has significantly reduced energy consumption during the training phase by using a “mixture of experts” technique and improved reinforcement learning. This allows only a portion of the model’s parameters to be active at any given time, thereby lowering energy use.
• Inference Phase Challenges: While DeepSeek is more efficient in training, its performance during the inference phase (where the model is used by consumers) is not significantly better than other models. This means that the overall energy consumption could still be high due to increased usage.
• Market and Investment Impact: The efficiency gains from DeepSeek have led to a reevaluation of AI investments. Companies are now more focused on cost-effective and energy-efficient models, potentially reducing the need for massive data center expansions. However, the increased adoption of AI could still drive up overall energy demand
• Geopolitical and Economic Dynamics: The emergence of DeepSeek has highlighted the competitive landscape between the US and China in AI technology. It has also raised questions about the effectiveness of US export controls and the future of AI development globally.

Environmental Implications: DeepSeek’s success has sparked excitement as it suggests that AI development can be more sustainable. However, the Jevons paradox indicates that efficiency gains might lead to increased usage, potentially negating some of the initial environmental benefits.

1. Increased Competition and Innovation

• Positive Impact: Competition between the U.S. and China in AI development can drive rapid advancements in technology, including in areas like AI ethics and safety. Both nations may invest more in developing frameworks and tools to ensure their AI systems are safe, transparent, and aligned with ethical principles.
• Negative Impact: The pressure to outpace each other could lead to cutting corners in safety and ethical considerations, prioritising speed and performance over responsible development.

2. Divergence in Ethical Standards

• The U.S. and China have different cultural, political, and regulatory environments, which could lead to divergent approaches to AI ethics and safety. For example:
  • The U.S. may emphasise individual rights, transparency, and accountability.
  • China may prioritise state control, social stability, and collective interests.
• This divergence could create challenges in establishing global norms and standards for AI ethics and safety.

3. Global Influence and Governance

• As DeepSeek v3 and other Chinese AI models gain prominence, China’s influence over global AI governance frameworks may grow. This could lead to tensions with the U.S. and its allies over whose ethical and safety standards should prevail.
• The competition could also spur efforts to create international agreements or standards for AI development, similar to arms control treaties, to mitigate risks associated with AI misuse.

4. Dual-Use Concerns

• AI technologies developed by both nations have dual-use potential, meaning they can be used for both civilian and military applications. The arms race could exacerbate concerns about the militarization of AI, including autonomous weapons systems, and the ethical implications of such technologies.

5. Collaboration vs. Confrontation

• While competition is intense, there is also a growing recognition of the need for collaboration on AI safety and ethics, given the global nature of AI’s impact. Initiatives like the U.S.-China AI Dialogue or multilateral efforts through organisations like the UN could provide platforms for cooperation.
• However, geopolitical tensions and mistrust could hinder meaningful collaboration, leaving critical ethical and safety challenges unaddressed.

6. Impact on Global AI Ethics Leadership

• The U.S. has traditionally positioned itself as a leader in promoting ethical AI development, but the rise of Chinese models like DeepSeek v3 could challenge this leadership. If China demonstrates a commitment to ethical AI (albeit aligned with its own values), it could reshape global perceptions of who sets the agenda for AI ethics and safety.

Conclusion:

The competition between DeepSeek v3 and U.S. frontier models underscores the need for a balanced approach that fosters innovation while prioritising ethical and safety considerations. The U.S.-China AI arms race could either lead to a dangerous escalation with insufficient safeguards or catalyse global efforts to establish robust ethical and safety standards. The outcome will depend on whether competition is tempered by cooperation and a shared recognition of the risks posed by unchecked AI development.