Move Over, ChatGPT! DeepSeek Has Arrived

After showcasing innovative artificial intelligence (AI) models that provide performance comparable to the best chatbots in the world at what appears to be a fraction of the price, DeepSeek, a Chinese AI start-up that has been in business for just over a year, has shocked and alarmed Silicon Valley.

The rise of DeepSeek might provide evidence against the widely held notion that developing AI in the future will demand ever-increasing resources.

As the excitement surrounding DeepSeek’s breakthrough grew and investors started to consider the ramifications for its US-based competitors and their hardware suppliers, global technology stocks fell.

What is DeepSeek exactly?

Mr. Liang Wenfeng, the head of the AI-driven quantitative hedge fund High-Flyer, created DeepSeek in 2023. The business creates open-source AI models that are available for review and enhancement by the larger developer community. Following its introduction in early January, its mobile app shot to the top of the US iPhone download charts.

The application sets itself apart from other chatbots, such as OpenAI’s ChatGPT, by explaining its logic before responding to a prompt. According to the business, its new AI model, R1, performs on par with OpenAI’s most recent model. It has also issued licenses to anyone who wants to build chatbots using the technology.

What is the difference between DeepSeek’s R1 and OpenAI or Meta AI?

The cost of training and developing DeepSeek’s models is a small portion of what is needed for the top solutions from OpenAI or Meta Platforms, albeit the company has not provided a thorough breakdown. The model’s significantly higher efficiency raises doubts about the necessity of making large financial investments to purchase the newest and most potent AI accelerators from companies like Nvidia. The US export restrictions on these cutting-edge semiconductors to China, which were put in place to stop a breakthrough like the one DeepSeek seems to represent, are also brought to light by this.

According to DeepSeek, R1 performs on par with or better than competing models in several industry-standard benchmarks, including AIME 2024 for mathematical problems, MMLU for general knowledge, and AlpacaEval 2.0 for Q&A performance. It also appears among the top performers on Chatbot Arena, a leaderboard connected to UC Berkeley.

READ MORE:

Google to Invest $1B in OpenAI Rival Anthropic

ChatGPT is Now Just a Message Away on WhatsApp!

OpenAI confirms ChatGPT outage

What is alarming in the United States?

In an effort to slow China’s progress in artificial intelligence (AI), the crucial area in the US-China struggle for technological dominance, Washington has prohibited the sale of expensive goods like GPU semiconductors to the nation. However, DeepSeek’s development indicates that Chinese AI developers have bypassed the limitations by concentrating on increasing productivity with limited resources. Although the extent of DeepSeek’s access to sophisticated AI-training hardware is still unknown, the company has shown enough that the trade restrictions have not completely stopped China’s advancement.

When did DeepSeek become well-known around the world?

Since the release of its first model in 2023, the AI developer has been the subject of intense scrutiny. Then, in November, it unveiled its DeepSeek R1 reasoning model, which is intended to simulate human thought processes. This model serves as the foundation for its mobile chatbot software, which, along with the web interface, shot to international fame in January as a far less expensive OpenAI substitute. Investor Marc Andreessen dubbed this “AI’s Sputnik moment.”

According to data from market tracker App Figures, the DeepSeek mobile app had 1.6 million downloads by January 25. It was the top app in iPhone app stores in Australia, Canada, China, Singapore, the US, and the UK.

Who is the originator of DeepSeek?

Mr. Liang, who was born in Guangdong in 1985, graduated from Zhejiang University with a bachelor’s and master’s degree in electronic and information engineering. According to the company database Tianyancha, he used 10 million yuan (S$1.9 million) in registered capital to launch DeepSeek in 2023.

In an interview with the Chinese media site 36kr, Mr. Liang stated that US limits on access to the best chips impede further advancements, not additional fund-raising. He emphasized the need for China to create a domestic ecosystem similar to the one centered around Nvidia and its AI chips, stating that the majority of his top researchers were recent graduates from prestigious Chinese colleges.

Innovation may not always follow from more investment. If not, big businesses would control all innovation, Mr. Liang stated.

What is DeepSeek’s position in China’s AI market?

China’s top digital companies, including Tencent Holdings, Baidu, and Alibaba Group Holdings, have invested a lot of money and resources in the competition to get clients and hardware for their AI projects. In addition to Kai-Fu Lee’s 01.AI start-up, DeepSeek is notable for its open-source methodology. It aims to swiftly acquire the greatest number of users before creating revenue-generating plans for that sizable user base.

In China, where the larger competitors have been involved in a pricing war that has seen rounds of price reductions over the past year and a half, DeepSeek has already helped lower prices for AI developers due to the lower cost of its models.

What effects will this have on the global AI market?

The success of DeepSeek might force OpenAI and other US suppliers to reduce their prices to maintain their current advantage. If more effective models can compete with a much smaller outlay, it also raises doubts about the massive spending by corporations like Microsoft and Meta, which have each promised capital expenditures of at least US$65 billion (S$87.7 billion) this year, mostly on AI infrastructure.

As a result, investors sold off firms like Nvidia and ASML Holding that had profited from the surge in demand for AI services, upsetting international stock markets. Chinese companies connected to DeepSeek, such as Iflytek, saw an increase in share prices.

Developers worldwide are already experimenting with DeepSeek’s software and want to use it to create tools. This might accelerate the uptake of sophisticated AI reasoning models, but it might also raise more questions about the necessity of restrictions on their application. DeepSeek’s developments could speed up regulation to limit the growth of AI.

What are the drawbacks of DeepSeek?

DeepSeek self-censors on subjects considered sensitive in China, just as all other Chinese AI models. It sidesteps concerns concerning the Tiananmen Square demonstrations of 1989 or sensitive geopolitical issues like the potential for a Chinese invasion of Taiwan. The DeepSeek bot can provide in-depth answers about political people in testing, such as Indian Prime Minister Narendra Modi. Still, it will not respond to questions regarding Chinese President Xi Jinping.

The increased popularity of DeepSeek is going to put its cloud infrastructure to the test. After a brief severe outage on January 27, the company will have to deal with even more traffic as both new and returning users continue to flood its chatbot with questions.