DeepSeek AI Chatbot: Key Features and Insights

DeepSeek has taken the world by storm, rocketing to the top of the Apple App Store and Google Play charts. Their AI models, crafted with a keen eye on efficiency, have got Wall Street analysts and tech gurus buzzing about the U.S.'s AI dominance and the ongoing demand for AI chips.

But where did DeepSeek come from, and how did they shoot to global stardom so fast?

DeepSeek's trader origins

DeepSeek is backed by High-Flyer Capital Management, a Chinese quant hedge fund that uses AI to make its trading calls. AI enthusiast Liang Wenfeng, who started dabbling in trading while at Zhejiang University, kicked off High-Flyer in 2015. By 2019, it was a full-blown hedge fund, all about developing and deploying AI algorithms.

In 2023, High-Flyer spun off DeepSeek as a lab focused on AI research, separate from its financial operations. With High-Flyer as an investor, DeepSeek turned into its own company. From the get-go, they built their own data centers for training models. But, like other Chinese AI outfits, they've been hit by U.S. export bans on hardware. For one of their recent models, they had to use Nvidia H800 chips, which are a bit less powerful than the H100 chips available to U.S. companies.

DeepSeek's tech team is pretty young. They're on a mission to scoop up doctorate AI researchers from top Chinese universities. They even hire folks without a computer science background to help their tech grasp a broader range of subjects, according to The New York Times.

DeepSeek's strong models

DeepSeek unveiled its first set of models — DeepSeek Coder, DeepSeek LLM, and DeepSeek Chat — in November 2023. But it wasn't until last spring, with the release of their next-gen DeepSeek-V2 family of models, that the AI world started paying attention.

DeepSeek-V2, which can handle both text and images, did well on various AI benchmarks and was way cheaper to run than other models at the time. It made DeepSeek's domestic rivals, like ByteDance and Alibaba, slash their prices and offer some models for free.

Then came DeepSeek-V3 in December 2024, which only added to their fame. According to DeepSeek's internal tests, DeepSeek V3 outshines both open-source models like Meta's Llama and closed models like OpenAI's GPT-4o.

Their R1 "reasoning" model, released in January, is just as impressive. DeepSeek says R1 performs as well as OpenAI's o1 model on key benchmarks. As a reasoning model, R1 fact-checks itself, which helps it dodge some of the usual traps. It might take a bit longer — seconds to minutes — to come up with answers, but it's more reliable in fields like physics, science, and math.

There's a catch, though. Being Chinese-developed AI, DeepSeek's models have to pass China's internet regulator's benchmarks to ensure they reflect "core socialist values." For instance, in DeepSeek's chatbot app, R1 won't touch questions about Tiananmen Square or Taiwan's autonomy.

In March, DeepSeek hit 16.5 million visits. "For March, DeepSeek is in second place, despite seeing traffic drop 25% from where it was in February, based on daily visits," David Carr from Similarweb told TechCrunch. Still, it's nowhere near ChatGPT, which zoomed past 500 million weekly active users in March.

A disruptive approach

If DeepSeek has a business model, it's a bit of a mystery. They price their products and services way below market rates and give some away for free. They're also not taking investor money, despite a lot of VC interest.

DeepSeek says efficiency breakthroughs have kept their costs super low. Some experts aren't buying their numbers, though.

Whatever the truth, developers are loving DeepSeek's models. They're not open source in the traditional sense, but they come with permissive licenses that let you use them commercially. Clem Delangue, CEO of Hugging Face, one of the platforms hosting DeepSeek's models, says developers have made over 500 "derivative" models of R1, which have been downloaded 2.5 million times in total.

DeepSeek's rise against bigger, more established players has been called "upending AI" and "over-hyped." They're at least partly to blame for Nvidia's stock dropping 18% in January and for getting a public shout-out from OpenAI CEO Sam Altman. In March, U.S. Commerce department bureaus told their staff to ban DeepSeek on government devices, according to Reuters.

Microsoft announced that DeepSeek is available on its Azure AI Foundry service, which brings AI services for businesses together under one roof. When asked about DeepSeek's impact on Meta's AI spending during its first-quarter earnings call, CEO Mark Zuckerberg said AI infrastructure spending will keep being a "strategic advantage" for Meta. In March, OpenAI called DeepSeek "state-subsidized" and "state-controlled," suggesting the U.S. government should consider banning their models.

During Nvidia's fourth-quarter earnings call, CEO Jensen Huang praised DeepSeek's "excellent innovation," saying that reasoning models like theirs are great for Nvidia because they need more compute power.

At the same time, some companies and entire countries, including South Korea and New York state, are banning DeepSeek from government devices.

As for what's next for DeepSeek, new and improved models are a sure bet. But the U.S. government seems to be getting nervous about what it sees as harmful foreign influence. In March, The Wall Street Journal reported that the U.S. will likely ban DeepSeek on government devices.

*This story was originally published January 28, 2025, and will be updated regularly.*