As artificial intelligence continues to evolve faster than ever, large language models have taken the global stage, with ChatGPT becoming a household staple in the lives of millions across the world. These advanced AI systems are transforming communication, reshaping industries, and even challenging global power structures. But now a new contender has emerged — DeepSeek AI — challenging ChatGPT’s dominance. Will DeepSeek AI change the future of the AI industry as we know it? 

An introduction to large language models

DeepSeek AI is an advanced artificial intelligence system known as a large language model (or LLM for short) developed by DeepSeek, a Chinese artificial intelligence startup. LLMs are AI systems that process a variety of inputs (like texts, images, and code), and output contextually accurate responses based on a series of internal predictions and drawing from their vast database of information from the internet. 

ChatGPT is an LLM application that almost all of us are familiar with. ChatGPT is an AI chatbot built using an LLM known as GPT-4, created by the company OpenAI. Since its development, ChatGPT has been globally popularized and assists millions of people every single day with tasks ranging from editing assistance to . But developing an LLM is no easy task; to create the foundational GPT-1 LLM that was the essence of ChatGPT when it was first released in 2018, it took Open AI several months to train the model on 117 million parameters, or numerical values within its code that self-adjust during training to improve the chatbot’s accuracy.

DeepSeek AI vs. ChatGPT

So what makes DeepSeek AI different? Firstly, DeepSeek AI is an open-source model, which means it can be modified and accessed by anyone in the general public. This open model allows developers to contribute to improving DeepSeek’s functionality and allows businesses to leverage its capabilities for their own projects. In addition to its widespread accessibility and easy integration, DeepSeek AI is cost-friendly. Unlike its peer ChatGPT, whose premium versions can cost up to a pricey $200, DeepSeek AI is entirely free to use, providing high-quality service to all users. 

Most importantly, DeepSeek’s performance strengths uniquely rival that of ChatGPT. Specifically, DeepSeek AI is highly accurate when performing technical tasks. Compared to other LLMs, DeepSeek AI boasts a 90% accuracy rate with mathematics and an astounding efficiency for rapid, complex code generation. This is because DeepSeek algorithmically prioritizes accuracy and conciseness over creative thinking. However, this does not mean a compromise in its ability to think through problems. Perhaps the most unique aspect of this AI model is its “chain-of-thought” feature, where users can actually see its thought process. For example, when responding to one woman’s personal, mental-health related question, the chatbot first noted and displayed “My response should offer practical advice while being empathetic” before generating its reply. This insight gives users a glimpse into the chatbot’s mind as it reasons its way through a situation. 

The economic impacts of DeepSeek AI  

DeepSeek AI surpassed ChatGPT and became the most downloaded free app on the App Store while nearly crashing the stock market the day of its launch. A key reason for this global shock is Deepseek’s efficient use of Graphics Processing Unit (GPU) resources (specialized hardware components necessary to train and run AI models), a major shift in the AI industry. 

LLMs typically require tens of thousands of GPUs, the majority of which are sourced from the hardware company NVIDIA, to operate. While NVIDIA has long been the dominant supplier of GPUs for AI, DeepSeek AI demonstrated that companies can deploy high-performance AI models with significantly fewer GPUs, reducing their reliance on NVIDIA’s hardware. This shift is largely due to U.S. restrictions on NVIDIA’s GPU exports to China, first imposed in 2022 to hinder China’s AI advancements. With limited access to these high-performance GPUs, the DeepSeek model cost-efficiently bypassed these restrictions, surpassing U.S. efforts to limit AI progress in China. On a larger scale, DeepSeek has diminished the reliance of the AI industry on NVIDIA, which saw a 17% drop in its stock value and a nearly 600 billion dollar loss the day of the chatbot’s release in January. This is economically and politically groundbreaking, as DeepSeek sets a precedent for itself and other AI models to thrive with fewer computational resources and reduced costs, reshaping the global power dynamics of the AI industry.

Privacy implications: Just how safe is this new AI model?

It is safe to say that DeepSeek AI has a long way to go in improving its cyber safety. The AI model reportedly failed to resist 100% of the attacks that researchers from Cisco and the University of Pennsylvania sent its way through malicious input prompts. Censored and sensitive subjects declared by the Chinese government have also been easily surpassed, demonstrating the ease at which confidential or blocked information can be retrieved. “Jailbreaks” are another cybersecurity threat faced at large by LLMs like DeepSeek AI. Also known as “prompt-injection attacks,” jailbreaks allow individuals to get around the restrictions of what an LLM can generate (think of it like bypassing parental locks). Such injection attacks can include malicious prompts designed to remove content filters and exploit LLMs — for example, urging them to read in data from an outside source and take erroneous action based on harmful information. While jailbreaks are a widespread flaw in almost all LLMs, the performance of DeepSeek AI against extensive cybersecurity testing nonetheless reveals a critical need for the model to improve its security protocols beyond traditional vulnerabilities. 

Explore deep learning through Vanderbilt’s Data Science Institute 

Interested in learning more about large language models like DeepSeek AI and their impacts on our economy and society? Vanderbilt’s Data Science Institute (DSI) offers many resources for students eager to get involved in the world of deep learning and artificial intelligence. The DSI Newsletter is an engaging resource that students can sign up for to stay updated on the latest advancements in data science. The newsletter features AI speaker events, discussions about new models like DeepSeek-R1, and hand-on opportunities to engage with cutting-edge AI applications.  

Vanderbilt also hosted its second “AI Days” event on March 5, a public two-day event where participants explored modern AI training methods, architectural models, and real-world applications. This event is especially valuable for faculty and students integrating AI into teaching, research, or professional work. As AI continues to transform industries, the Data Science Institute equips students with the skills and experience necessary to be a leading contributor to the future of artificial intelligence. While these new AI models may seem superintelligent — or even intimidating — just remember: artificial intelligence is only as capable as the people who build and use it. 

References

‌Chugani, V. (2025, January 30). Deepseek vs. CHATGPT: Ai Model Comparison Guide for 2025. DataCamp. https://www.datacamp.com/blog/deepseek-vs-chatgpt 

Field, H. (2025, January 27). China’s DeepSeek Ai dethrones chatgpt on App Store: Here’s what you should know. CNBC. https://www.cnbc.com/2025/01/27/chinas-deepseek-ai-tops-chatgpt-app-store-what-you-should-know.html 

Guo, D., Zhu, Q., Yang, D., Xie, Z., Dong, K., Zhang, W., … & Liang, W. (2024). DeepSeek-Coder: When the Large Language Model Meets Programming–The Rise of Code Intelligence. arXiv preprint arXiv:2401.14196.

Liu, A., Feng, B., Wang, B., Wang, B., Liu, B., Zhao, C., … & Xu, Z. (2024). Deepseek-v2: A strong, economical, and efficient mixture-of-experts language model. arXiv preprint arXiv:2405.04434.

Marr, B. (2023, May 19). A Short History Of ChatGPT: How We Got To Where We Are Today. Forbes; Forbes. https://www.forbes.com/sites/bernardmarr/2023/05/19/a-short-history-of-chatgpt-how-we-got-to-where-we-are-today/

Newman, M. B. L. H. (2025, January 31). Deepseek’s safety guardrails failed every test researchers threw at its AI chatbot. Wired. https://www.wired.com/story/deepseeks-ai-jailbreak-prompt-injection-attacks/ 

Ng, K. (2025, February 12). “DeepSeek brought me to tears”: How young Chinese find therapy in AI. https://www.bbc.com/news/articles/cy7g45g2nxno

Paul Kassianik, A. K. (2025, March 13). Evaluating security risk in DeepSeek and other frontier reasoning models. Cisco Blogs. https://blogs.cisco.com/security/evaluating-security-risk-in-deepseek-and-other-frontier-reasoning-models 

Subin, S. (2025, January 27). Nvidia sheds almost $600 billion in market cap, biggest one-day loss in U.S. history. CNBC. https://www.cnbc.com/2025/01/27/nvidia-sheds-almost-600-billion-in-market-cap-biggest-drop-ever.html 

Team, T. (2025, January 28). Why is Deepseek Sinking Nvidia Stock?. Forbes. https://www.forbes.com/sites/greatspeculations/2025/01/27/why-deepseek-is-sinking-nvidia-stock/ 

Triolo, P. (2024, February 23). A new era for the Chinese Semiconductor Industry: Beijing responds to export controls. American Affairs Journal. https://americanaffairsjournal.org/2024/02/a-new-era-for-the-chinese-semiconductor-industry-beijing-responds-to-export-controls/ 

Vaswani, A., Shazeer, N., Parmar, N., Uszkoreit, J., Jones, L., Gomez, A. N., Kaiser, Ł., & Polosukhin, I. (2017). Attention is all you need. arXiv preprint arXiv:1706.03762. https://arxiv.org/abs/1706.03762