What’s DeepSeek and why is everybody speaking about it?

It’s practically unimaginable to flee the web hype surrounding DeepSeek, a comparatively new and unknown AI chatbot, proper now. In only a few days, it has not solely dethroned ChatGPT’s dominance in benchmarks but in addition develop into essentially the most downloaded app on iOS and Android. What’s much more spectacular is that the AI was developed by a small Chinese language startup with a tiny funds and comparatively outdated {hardware}.

So what precisely is DeepSeek, why has it garnered a lot consideration, and must you think about using it over ChatGPT or Gemini? I dug into the Web’s new darling AI and took it for a test-drive to search out out.

What’s DeepSeek?

On the floor, DeepSeek is solely an AI chatbot that you should utilize totally free. In these phrases, it’s functionally the identical as OpenAI’s ChatGPT or Google’s Gemini. Nevertheless, not like its opponents, DeepSeek doesn’t impose any limits or cost a month-to-month subscription to make use of its most succesful fashions. It’s additionally open supply, which means that you may obtain and run DeepSeek regionally — so long as you’ve gotten {hardware} that’s highly effective sufficient.

DeepSeek’s newest fashions don’t simply come near matching the competitors; they typically surpass it in a number of areas. The most recent DeepSeek-V3 mannequin scores higher in a number of coding, math, and Chinese language language benchmarks than OpenAI’s GPT-4o and Anthropic’s Claude-3.5. The latter have been considered the gold normal in AI efficiency till DeepSeek dethroned them virtually in a single day.

DeepSeek has been round for a while now, but it surely wasn’t till January 20, 2025 that it started gaining mainstream recognition. That day noticed the discharge of DeepSeek-R1, the corporate’s first AI mannequin able to reasoning like a human.

At the same time as a reasonably common AI person, I’ve to confess that the DeepSeek-R1 mannequin appears magical in the way it appears to assume earlier than responding to my prompts (pictured above). Whereas we’ve already had AI able to reasoning since OpenAI launched GPT-o1 in September 2024, it has remained out of attain for most individuals. That is still true even at this time — you’ll have to fork over $20 monthly for a ChatGPT Plus subscription to entry the GPT-o1 reasoning mannequin.

For builders seeking to combine AI fashions into their very own apps, DeepSeek is about 20 to 30 occasions cheaper as in comparison with ChatGPT’s underlying mannequin. All of those components mixed make DeepSeek a formidable contender within the AI race, regardless that it appeared out of just about nowhere.

Whereas DeepSeek technically outperforms different AI chatbots, I don’t assume it’s the breath of contemporary air that the web is at the moment making it out to be. For one, the mannequin continues to be simply as susceptible to hallucinations as any of its rivals and I’ve discovered that it’s going to completely invent info out of skinny air when requested about one thing it’s unfamiliar with. This makes its usefulness situational at finest — and never that a lot completely different than ChatGPT in any case.

Even DeepSeek-R1, the mannequin able to human-like reasoning, solely is sensible in very restricted use-cases. Until I’m writing advanced code or fixing math issues on the common, I received’t get any higher outcomes from the reasoning mannequin than the usual DeepSeek-V3 mannequin. The latter doesn’t let me observe the AI debate itself but it surely’s considerably sooner and practically as adept at answering questions and different general-purpose duties.

I can’t say there are numerous incentives to make the swap to DeepSeek proper now, whilst a daily ChatGPT and Gemini person. The latter have a way more polished ecosystem, with options like imaginative and prescient and two-way voice chat a la Gemini Stay that I take advantage of rather more steadily. DeepSeek solely helps text-based conversations for now, though that can doubtless change sooner moderately than later.

Then there’s the elephant within the room: privateness and censorship. Whereas I’ve no qualms with DeepSeek’s Chinese language origins, that’s solely as a result of I don’t have a tendency to make use of AI for any duties that require discretion or secrecy. Provided that OpenAI has to deal with information privateness laws in Europe and North America, you possibly can at the very least change a setting to stop your chats from getting used to coach future fashions. The identical can’t be mentioned for DeepSeek, until you host it your self.

In case you use the web model of DeepSeek, you’ll additionally need to cope with censorship. The chatbot at the moment avoids speaking about subjects delicate to Chinese language sentiments, but it surely may start selling biased info sooner or later. So if you happen to’re searching for essentially the most sincere chatbot on the market, DeepSeek doesn’t precisely match the invoice.

DeepSeek is owned by Chinese language entrepreneur Liang Wenfeng, who additionally created a hedge fund named Excessive-Flyer. The startup’s excellent efficiency would have gone largely unnoticed exterior of the AI world if it weren’t for its Chinese language origins and virtually shoestring funds. DeepSeek has managed to dethrone billion-dollar ventures like OpenAI whereas additionally proving that bigger investments don’t at all times lead to higher outcomes.

China has traditionally lagged behind the West within the AI race, largely because of the U.S. authorities imposing strict export controls on American corporations like Nvidia beginning in 2022. These controls banned the sale of superior AI coaching and processing {hardware} to Chinese language corporations. Furthermore, with out the help of tech giants like Microsoft and Google to pour billions of {dollars} into AI analysis and growth, it appeared unlikely that China would ever catch up.

But, we now know {that a} lean Chinese language startup managed to develop a extremely succesful AI mannequin with allegedly simply $6 million in computing energy — a fraction of the funds utilized by OpenAI or Google. DeepSeek achieved this feat utilizing older NVIDIA H800 GPUs that it managed to accumulate despite the US’ export controls. The chatbot additionally makes use of homegrown Huawei-made chips to generate responses, additional proving that China doesn’t want American {hardware} to compete within the AI race.

Why did DeepSeek have an effect on US tech inventory costs?

DeepSeek’s success has rattled each the U.S. authorities and American tech giants, however doubtless solely within the quick time period. NVIDIA’s valuation dropped by $600 billion whereas the tech stock-heavy NASDAQ 100 index dropped 4% in a single buying and selling session. Nevertheless, most of them reversed course the very subsequent day, signalling that the market continues to be optimistic about the way forward for AI. A rising tide lifts all ships, in any case.

Why do governments need to ban DeepSeek?

After TikTok, DeepSeek would possibly develop into the most recent app to attract the ire of the US authorities and even face a complete ban over nationwide safety considerations. Whereas the chatbot doesn’t acquire information immediately out of your telephone, we don’t know the way the corporate collects and shops its customers’ chats on its servers. It’s completely doable that the Chinese language authorities can acquire entry to all person information, which might have grave penalties for nationwide safety. Provided that, it’s not stunning that we’ve seen the Australian and American governments urge warning when utilizing the chatbot.

Presently, DeepSeek’s foremost benefit is in attaining cutting-edge efficiency with far much less computing energy than the competitors. Fortunately, that doesn’t have a lot affect on the way in which you and I take advantage of AI at this time, so a ban would have little affect within the quick time period.