Liang Wenfeng founded DeepSeek in 2023, and it quickly became well-known by outperforming ChatGPT in the App Store. The AI startup's low-cost business models put US companies to the test, raising concerns about the efficacy of US export restrictions on AI technology as well as the loss of market share.
By surpassing ChatGPT as the most popular free game on the Apple App Store, the Chinese AI business DeepSeek has created quite a stir. Although ChatGPT has maintained its lead in the AI race since its inception in late 2022, DeepSeek now seems to be the next big thing in the field, despite increasing opposition from Google, Meta, and Anthropic. Here's an overview of DeepSeek, including what it is, how to use it, why it worries markets, and more.DeepSeek: What is it?
Liang Wenfeng established the AI startup DeepSeek in the eastern Chinese city of Hangzhou in July 2023. The startup gained notoriety last month when multiple benchmarks revealed that, despite being constructed at a significantly lower cost, its V3 large language model (LLM) surpassed those of numerous well-known US computer giants.
In a number of benchmarks, DeepSeek's R1 reasoning model, which simulates some features of human reasoning, both matched and exceeded OpenAI's most recent O1 model.
Both DeepSeek R1 and DeepSeek V3 were in the top 10 on Chatbot Arena, a ranking run by UC Berkeley that assesses the effectiveness of top chatbots, at the time this article was written. Although ChatGPT 4-o and Google's Gemini 2.0 Flash both outperformed the DeepSeek models, they outperformed models from Amazon-backed Anthropic and Elon Musk's xAI.
What cannot be done via DeepSeek?
Although DeepSeek's chatbots are capable of competing with their Western counterparts on nearly all metrics, they are hesitant to respond to inquiries that raise doubts about China. The response to any such query is, "Sorry, that's outside of my current purview." Let's discuss another topic.
The widely held notion that running AI models demands ever-increasing amounts of processing power is refuted by DeepSeek's LLMs, which are constructed at a far lower cost. Researchers from DeepSeek said in a report last month that the V3 model cost less than $6 million and was trained using Nvidia H800 chips, which is a small amount in comparison to the billions of dollars that AI behemoths like Microsoft, Meta, and OpenAI have promised to invest this year alone.
Given its quick ascent, DeepSeek is probably going to reduce the market share of well-known US AI firms like OpenAI, Meta, Google, and others. In order to stay competitive, Western businesses will probably have to cut their pricing in response to the low-cost Chinese models.
How is DeepSeek used?
DeepSeek can be downloaded for free from the Google and Apple app stores. Additionally, the Chinese company's website chat.deepseek.com offers access to its most recent AI models.
However, DeepSeek declared on Monday that a significant malicious attack on its services was the reason for the temporary suspension of registrations.
How has DeepSeek affected markets?
As investors questioned the spending intentions of leading AI companies, DeepSeek's ascent to the top of the App Store caused a nearly $1 trillion crash in US and European technology stocks, according to Bloomberg.
Jensen Huang's Nvidia has been the most severely impacted by the DeepSeek AI disruption, but the situation is nasty overall—the S&P 500 fell, and the Nasdaq fell more than 3%.
Due in large part to Nvidia's drag, which saw it drop a record 17% overnight, the US tech-heavy Nasdaq fell 3.1% on January 27. Chip manufacturer Broadcom Inc. finished the day down 17.4%, ChatGPT backer Microsoft was down 2.1%, and Google parent Alphabet was down 4.2%, according to the Reuters report.
Who is Wenfeng Liang
Chinese businessman and entrepreneur Liang Wenfeng, 40, has experience in finance. He co-founded the quantitative hedge fund High-Flyer in 2015 following his graduation from Zhejiang University. He employed AI trading tactics with High-Flyer to forecast market movements and make investment choices.
Prior to the US government under Joe Biden limiting US shipments of AI chips to China, Wenfeng began purchasing thousands of Nvidia graphics processors in 2021 as a side project. His friends at the time thought this AI side project was a strange pastime with no real purpose.
Despite DeepSeek's original perception as a side project, Liang was actively involved in both the company and its research. He also intends to establish DeepSeek as a domestic AI leader.
How is DeepSeek being received in Silicon Valley?
In later posts on X, OpenAI CEO Sam Altman continued to commend DeepSeek's R1 model while pledging to develop stronger models going forward. "Deepseek's R1 is an impressive model, especially when considering what they can offer for the price," he stated.
Naturally, we will produce far superior models, and having a new rival is genuinely energizing! We'll retrieve a few releases.
"DeepSeek has benefited from open research and open source (e.g. PyTorch and Llama from Meta)," stated Yann LeCun, chief AI scientist at Meta, in a post on Threads. They developed new concepts and expanded upon the work of others. Since their work is publicly available and open source, anybody can profit from it. That is the power of open research and open source.”
Marc Andreessen, a venture capitalist, described DeepSeek as a "deep gift" to the world. "One of the most amazing and impressive breakthroughs I've ever seen — and as open source, a profound gift to the world," he wrote in a post on X.
In a post on X, Salesforce CEO Mark Benioff called the startup "Deepgold," stating that "Deepseek is now #1 on the AppStore, surpassing ChatGPT—no NVIDIA supercomputers or $100M needed." The user interface and model are no longer the true gems of AI; they have become commodities. Data and metadata are what really provide value and are what give AI its potential. The fortune of the future? Our data contains it. Deepgold. 😇
In later posts on X, OpenAI CEO Sam Altman continued to commend DeepSeek's R1 model while pledging to develop stronger models going forward. "Deepseek's R1 is an impressive model, especially when considering what they can offer for the price," he stated.
Naturally, we will produce far superior models, and having a new rival is genuinely energizing! We'll retrieve a few releases.
"DeepSeek has benefited from open research and open source (e.g. PyTorch and Llama from Meta)," stated Yann LeCun, chief AI scientist at Meta, in a post on Threads. They developed new concepts and expanded upon the work of others. Since their work is publicly available and open source, anybody can profit from it. That is the power of open research and open source.”
Marc Andreessen, a venture capitalist, described DeepSeek as a "deep gift" to the world. "One of the most amazing and impressive breakthroughs I've ever seen — and as open source, a profound gift to the world," he wrote in a post on X.
In a post on X, Salesforce CEO Mark Benioff called the startup "Deepgold," stating that "Deepseek is now #1 on the AppStore, surpassing ChatGPT—no NVIDIA supercomputers or $100M needed." The user interface and model are no longer the true gems of AI; they have become commodities. Data and metadata are what really provide value and are what give AI its potential. The fortune of the future? Our data contains it. Deepgold. 😇
0 Comments