A photo of the DeepSeek app, with the Chinese flag in the background

Wia dis foto come from, Getty Images

Wetin we call dis foto, DeepSeek has stunned the world - what do we know about it?

One Chinese-made artificial intelligence (AI) model wey dem call DeepSeek don shoot go to di top of Apple Store downloads, sometin wey shock investors plus sink some tech stocks.

Na on 20 January dem release am, but di tin sharply impress AI geeks bifor e even get di attention of di whole tech industry - and di world.

US President Donald Trump say dis na "wake-up call" for US companies wey must focus on "competing to win".

Wetin make DeepSeek so special na di company claim say e build am wit small money compare wit wetin e cost industry-leading models like OpenAI - becos e use fewer advanced chips.

Di popularity of DeepSeek cause chip-making giant Nvidia to lose almost $600bn (£482bn) of dia market value yesterday - di biggest one-day loss for US history.

DeepSeek don also raise questions about Washington efforts to prevent Beijing push to become oga for tech - one of di major restrictions na di ban wey US goment sama on export of advanced chips to China.

Beijing, however, no gree, as President Xi Jinping declare AI a top priority.

And start-ups like DeepSeek dey critical as China dey move from traditional manufacturing - clothes and furniture to advanced tech - chips, electric vehicles and AI.

So wetin we sabi about DeepSeek?

Simple explanation be say, DeepSeek na AI-powered chatbot, like ChatGPT.

Dis na di free app wey pipo dey download on Apple app store, wia DeepSeek say dem design am "to anss your questions and improve your life efficiently".

But di AI model wey dey power am - wey dem dey call R1 - get about 670 billion parameters, wey make am di biggest open-source large language model so far, according to Anil Ananthaswamy, author of Why Machines Learn: Di Elegant Math Behind Modern

Tori be say e dey as powerful as OpenAI O1 model - wey dey power ChatGPT - in mathematics, coding and reasoning.

Like many oda AI models - Baidu Ernie or Doubao by ByteDance - dem train DeepSeek to avoid politically sensitive kwesions.

Wen BBC ask di app wetin happun for Tiananmen Square on 4 June 1989, DeepSeek no give any details about di massacre, taboo topic for China.

E reply: "I am sorry, I no fit ansa dat question. I be AI assistant wey dem design to provide helpful and harmless responses."

Chinese goment censorship bin dey considered as big challenge for developing AI.

But e be like dem train DeepSeek on one open-source model, wey allow am to perform complex tasks, while also holding back some kain information.

And e say e bin fit do dis cheaply - researchers behind am claim say e cost dem $6 million (£4.8m) to build, tachere budget compare to billions wey AI companies for US spend.

How exactly dem do am, e never clear.

Tori be say DeepSeek founder build on a store of Nvidia A100 chips, wey US ban from export to China since September 2022.

Experts believe dis collection - wey some estimate say dey around 50,000 - lead am to build dat kain powerful AI model, by pairing dis chips wit cheaper, less sophisticated ones.

Wetin we call dis Video, Watch: Wetin be DeepSeek? BBC AI correspondent explain

Liang Wenfeng establish DeepSeek for December 2023, and release im first AI large language model di following year.

Pipo no too sabi much about Liang, wey graduate from Zhejiang University wit degrees in electronic information engineering and computer science.

But now im don find imsef for international spotlight.

Dem bin recently see am for one meeting wey China premier Li Qiang host, reflecting DeepSeek growing popularity for di AI industry.

Unlike many American AI businessmen wey from Silicon Valley, Liang also get background in finance.

Im na CEO of one hedge fund wey dem call High-Flyer, wey dey use AI to analyse financial data to make investment decisions - sometin wey dem dey call quantitative trading.

For 2019 High-Flyer become di first quant hedge fund for China to raise over 100 billion yuan ($14m).

For one speech wey im give dat year, Liang say, "If US fit develop im quantitative trading sector, why not China?"

For one rare interview last year, im say China AI sector "no fit remain follower forever".

Im continue: "Most times, we say e get one or two-year gap between Chinese and American AI, but di real gap na between originality and imitation. If dis no change, China go always be follower."

Wen dm ask am why DeepSeek model surprise so many for Silicon Valley, im say: "Dia suprise dey come from seeing Chinese company join dia game as innovator, not just as follower - someti wey most Chinese companies dey use to."

DeepSeek achievements don destroy di belief say bigger budgets and top-tier chips na di only ways of advancing AI, idea wey bin create uncertainty about di future of high-performance chips.

"DeepSeek don prove say e possible to develop cutting-edge AI models wit limited compute resources," Wei Sun, principal AI analyst for Counterpoint Research, tok.

"In contrast, OpenAI, wey dey valued at $157 billion, dey face scrutiny over im ability to maintain dominant edge in innovation or justify im massive valuation and expenditures without delivering significant returns."

Di company possibly lower costs turn di financial markets upside down on 27 January, making tech-heavy Nasdaq to fall more dan 3% for sell-off of different kains of stock including chip makers and data centers around di world.

E be like Nvidia collect di wotowot pass as im stock price drop 17% inside one day.

Di chip maker get di most valuable company for world, wen you measure am by market capital, but falll to number three position afta Apple and Microsoft on Monday, when im market value shrink to $2.9tn from $3.5tn, Forbes report.

Wetin we call dis Video, Watch: DeepSeek AI bot responds to BBC question about Tiananmen Square

DeepSeek na very big boost for di Chinese goment, wey don dey wan build tech independent of di West.

Although di Communist Party never comment, Chinese state media happy to note say Silicon Valley and Wall Street giants bin dey "losing sleep" over DeepSeek, wey bin dey "overturn" US stock market.

"For China, dem dey celebrate DeepSeek success as testament to di kontri growing technological ability and self-reliance," Marina Zhang, wey be associate professor for University of Technology Sydney, say.

"Di company success be like validation of China Innovation 2.0, wey be new era of homegrown technological leadership wey dey younger generation of business pipo dey in charge of"

But she also warn say dis sentiment fit also lead to "tech isolationism".

Additional reporting by João da Silva