DeepSeek: A Game-Changer in AI and the Power of Sparsity
If you wondered why your tech stocks dipped recently, the answer is DeepSeek, a Chinese AI company that unveiled its R1 model and in the process has upended conventional thinking about how advanced AI models are built and deployed.
What exactly is DeepSeek? I asked DeepSeek to describe itself and its effect on the field of artificial intelligence. It said:
DeepSeek is a groundbreaking Chinese AI company that has rapidly emerged as a significant player in the global artificial intelligence landscape. Its latest model, R1, has demonstrated capabilities on par with top-tier models from industry leaders like OpenAI, but at a fraction of the cost. What sets DeepSeek apart is its innovative use of **sparsity**—a technique that optimizes neural networks by selectively deactivating non-essential components, reducing computational demands without sacrificing performance. This approach, combined with its open-source release of the R1 model, has the potential to democratize AI development, enabling smaller companies and startups to compete with tech giants.
But why is DeepSeek such a big deal in artificial intelligence? There are a few reasons.
- It’s Cheap
Companies had assumed that they needed cutting-edge computer chips to create capable artificial intelligence. Based on this assumption, President Biden restricted sales of AI-capable chips from nVidia and other manufacturers to China, to restrict China’s ability to compete with the US in the AI field.
That didn’t matter. DeepSeek managed to train R1 using less advanced (read: less expensive) computer chips that it could import. DeepSeek costs about $6 million in training costs, compared to the $500 billion cost President Trump announced for the AI joint venture Stargate. This lowered cost means that DeepSeek could make advanced AI technologies more accessible to a broader range of organizations, including smaller businesses and even nonprofit organizations that would previously turn to American companies. (This may also get China’s foot in the door to American data, which it could use to train AI or other purposes.)
- It’s Open Source
DeepSeek’s R1 is open-source software, like Linux or Mozilla Firefox. This means that anyone can access and modify its source code. Other companies and researchers can build on its work, which could accelerate the pace of AI advancements and, again, broaden R1’s appeal. However, this also raises AI governance and regulatory challenges, as organizations must navigate the complexities of open-source licensing while protecting their own innovations and managing the risks associated with AI.
- Sparsity
DeepSeek’s R1 uses an AI principal called “sparsity.” In simple terms, sparsity refers to the practice of using only the necessary parts of a neural network or dataset to achieve the desired outcome. Think of an AI model as a vast network of interconnected nodes, each contributing to the model’s ability to process information and generate outputs. Sparsity identifies and prunes away the nodes that have minimal impact on the final result.
In simpler terms, imagine that other AI programs are like your kids, who leave the lights on all over the house so they can always see where they’re going. R1 turns off the lights in the rooms it’s not using, which saves energy without compromising functionality.
DeepSeek’s method is obviously less resource intensive; that is, it’s more efficient. We can expect this efficiency to catch the attention of AI’s makers and users.
- It Raises Ethical and Regulatory Issues
The AI community has already focused on issues regarding misuse, security, and bias, and DeepSeek’s sudden rise will accentuate these issues. As AI technologies become more accessible, these concerns will grow. Policymakers and legal professionals will need to work closely with technologists to ensure that AI advancements benefit society while minimizing risks.
Regulatory issues will also assume urgency. State governments and agencies have already issued an array of statutes and regulations governing AI, and the release of a cheaper AI alternative will only spur more regulation. And this urgency will spread worldwide, which means that companies and other organizations that operate internationally will face increased pressure to comply with such laws as the EU AI Act, EU’s General Data Protection Regulation (GDPR) and China’s Personal Information Protection Law (PIPL).
For businesses, DeepSeek’s emergence offers an opportunity and a challenge. DeepSeek is not just a technological breakthrough; it is a call to action for businesses, policymakers, and legal professionals to shape the future of AI effectively and responsibly Companies that embrace these advancements can gain a competitive edge, but they must also navigate the regulatory complexities that accompany AI.
Sean Griffin is a partner at Longman & Van Grack. He is one of the world’s first experts certified as an Artificial Intelligence Governance Professional by the International Association of Privacy Professionals, and he is a Certified Information Privacy Professional within the United States. Additionally, Sean is a member of the International Association of Defense Counsel (“IADC”), where he Co-Chairs the AI Committee and serves as a Vice Chair of the Cyber Security, Data Privacy and Technology Committee. Sean is also a member of the Association of Defense Trial Attorneys (“ADTA”), where he Chairs the Artificial Intelligence Steering Committee. Sean litigates business disputes in Maryland, Virginia, and Washington, DC. You can reach Sean directly via email at sean@lvglawfirm.com or via phone at 202-836-7828.
