- DeepSeek is a Chinese AI company whose newest chatbot shocked the tech industry.
- DeepSeek says its AI model rivals top competitors, like ChatGPT’s o1, at a fraction of the cost.
- DeepSeek’s rise has impacted tech stocks and led to scrutiny of Big Tech’s massive AI investments.
An artificial intelligence company based in China has rattled the AI industry, sending some US tech stocks plunging and raising questions about whether the United States’ lead in AI has evaporated.
The Chinese startup, DeepSeek, unveiled a new AI model last week that the company says is significantly cheaper to run than top alternatives from major US tech companies like OpenAI, Google, and Meta.
Here’s everything you need to know about the hot new company.
What is DeepSeek?
DeepSeek is a Chinese artificial intelligence startup founded in 2023.
It’s been the talk of the tech industry since it unveiled a new flagship AI model last week called R1 on January 20 with a reasoning capacity that DeepSeek says is comparable to OpenAI’s o1 model but at a fraction of the cost.
DeepSeek made the latest version of its AI assistant available on its mobile app last week — and it has since skyrocketed to become the top free app on Apple’s App Store, edging out ChatGPT.
Who is behind DeepSeek?
DeepSeek started as an AI side project of Chinese entrepreneur Liang Wenfeng, who in 2015 cofounded a quantitative hedge fund called High-Flyer that used AI and algorithms to calculate investments.
After buying thousands of Nvidia chips, Wenfeng started DeepSeek in 2023 with funding from High-Flyer.
The AI chatbot can be accessed using a free account via the web, mobile app, or API.
Why are investors worried about DeepSeek?
DeepSeek’s R1 model is built on its V3 base model. The company has said the V3 model was trained on around 2,000 Nvidia H800 chips at an overall cost of roughly $5.6 million.
And though the training costs are only one part of the equation, that’s still a fraction of what other top companies are spending to develop their own foundational AI models. Mark Zuckerberg, for example, announced that Meta plans to spend over $60 billion in capital expenditures this year as it doubles down on AI.
According to Bernstein analysts, DeepSeek’s model is estimated to be 20 to 40 times cheaper to run than similar models from OpenAI.
The relatively low stated cost of DeepSeek’s latest model — combined with its impressive capability — has raised questions about the Silicon Valley strategy of investing billions into data centers and AI infrastructure to train up new models with the latest chips.
Nvidia, a company that produces the high-powered chips crucial to powering AI models, saw its stock close on Monday down nearly 17% on Monday, wiping hundreds of billions from its market cap. Other Big Tech companies have also been impacted.
DeepSeek has also said its models were largely trained on less advanced, cheaper versions of Nvidia chips — and since DeepSeek appears to perform just as well as the competition, that could spell bad news for Nvidia if other tech giants choose to lessen their reliance on the company’s most advanced chips.
What are tech leaders saying about DeepSeek?
DeepSeek’s success is also getting top tech leaders talking.
Meta’s chief AI scientist, Yann LeCun, looked to temper some people’s panic over DeepSeek’s rise in a post on Threads over the weekend.
LeCun said it’s not so much that China’s AI advancements are leapfrogging ahead of the US, it’s more that “open source models are surpassing proprietary ones.”
Microsoft CEO Satya Nadella also weighed in on X.
“Jevons paradox strikes again!” Nadella posted Monday morning, referencing the idea that innovation breeds demand. “As AI gets more efficient and accessible, we will see its use skyrocket, turning it into a commodity we just can’t get enough of.”
Marc Andreessen, the cofounder of Silicon Valley venture capital firm Andreessen Horowitz said in a social media post that “Deepseek R1 is AI’s Sputnik moment,” referencing the Soviet Union’s satellite that shocked the US and helped launch the space race.
How does DeepSeek compare to ChatGPT and what are its shortcomings?
DeepSeek says that its R1 model rivals OpenAI’s o1, the company’s reasoning model unveiled in September.
Like o1, DeepSeek’s R1 takes complex questions and breaks them down into more manageable tasks.
R1’s proficiency in math, code, and reasoning tasks is possible thanks to its use of “pure reinforcement learning,” a technique that allows an AI model to learn to make its own decisions based on the environment and incentives.
Similar to ChatGPT, DeepSeek’s R1 has a “DeepThink” mode that shows users the machine’s reasoning or chain of thought behind its output.
Business Insider’s Tom Carter tested out DeepSeek’s R1 and found that it appeared capable of doing much of what ChatGPT can. The app looks similar to that of ChatGPT, with a sparse interface dominated by a text box.
One of the few things R1 is less adept at, however, is answering questions related to sensitive issues in China. For example, when Carter asked DeepSeek about the status of Taiwan, the chatbot tried to steer the topic back to “math, coding, and logic problems,” or suggested that Taiwan has been an “integral part of China” for centuries.
Read the full article here
