Chinese AI DeepSeek tops iPhone app chart, triggers $1 trillion panic, and sparks fears about free speech

DeepSeek AI logo on an iPhone screen pictured lying on a laptop keyboard

DeepSeek AI is now the most-downloaded free app in the United States and the UK

REUTERS
Aaron Brown

By Aaron Brown


Published: 28/01/2025

- 10:26

Updated: 28/01/2025

- 10:26

All products and promotions are independently selected by our experts. To help us provide free impartial advice, we will earn an affiliate commission if you buy something. Click here to learn more

Built by a relatively unknown startup from China, DeepSeek has had an impact on all the metrics that matter — the cost of training a new AI model, access to the latest chips, and availability

  • Latest DeepSeek AI model has caused panic in stock markets worldwide
  • Reports claim it has wiped $1 trillion (£801m) off company valuations
  • Chinese start-up claims it spent just $6 million (£4.8m) to build and train DeepSeek
  • That compares to the $100 million spent by OpenAI to create ChatGPT
  • Other companies have pledged eye-watering sums to AI technologies
  • Facebook parent company Meta has vowed to spend $65 billion this year
  • As a Chinese company, DeepSeek is subject to strict censorship
  • Asking questions about Tiananmen Square doesn't elicit a response
  • Screenshots show the DeepSeek chatbot erasing answers in real-time

The launch of a low-cost Artificial Intelligence (AI) model from China has sent tremors around the globe — wiping $1 trillion off stock markets and raising fears around free speech. DeepSeek is a new chatbot that can purportedly match the performance of industry leader ChatGPT, despite being developed at a fraction of the cost.

The DeepSeek chatbot currently offers two Large Language Models (LLMs):

  • DeepSeek-V3
  • DeepThink (R1)

DeepSeek-V3 functions in much the same way as the standard ChatGPT chatbot, offering speedy responses to general knowledge queries, generating never-before-seen text based on your written prompts, rewriting your emails or documents to adjust the tone, and summarising mammoth documents and tables. Apple recently baked many of these same capabilities into the iPhone 16 and 16 Pro with Apple Intelligence.

Likewise, Google Gemini is also capable of manipulating large swathes of text, summarising PDFs into bullet points, and writing everything from poetry to cover letters based on a short written prompt. It's included in the latest Android phones, like those released by Samsung earlier this month.

Meanwhile, DeepThink (R1) is comparable to the released-launched ChatGPT-o1 model, which is currently only available if you subscribe to the OpenAI assistant to the tune of either $20 (£16) or $200 (£160). This model boasts many of the same capabilities, but answers are presented in a step-by-step process — offering an insight into how the LLM is thinking about the question and why it has surfaced its final answer.

The team behind DeepThink (R1) has boasted of "performance on par with" the premium ChatGPT-o1 model when tackling complex tasks such as maths, wiring computer code, and natural language reasoning.

But unlike ChatGPT-o1, DeepSeek-R1 is available to use free of charge.

DeepSeek has made all of its AI models open-source model, enabling them to be accessed by anyone. Unsurprisingly, this has triggered a flood of downloads, with the DeepSeek – AI Assistant app becoming the most-downloaded free app on iPhone in the UK, United States of America, and several other countries.

Given that OpenAI was founded as a non-profit — with $45 million (£36 million) in funding from X and Tesla owner Elon Musk — to democratise AI for the masses, there's a certain irony that its original mission has been usurped by a little-known Chinese startup.

Nvidia senior research manager Jim Fan posted on X: “We are living in a timeline where a non-US company is keeping the original mission of OpenAI alive — truly open, frontier research that empowers all. It makes no sense. The most entertaining outcome is the most likely.”

The emergence of this new AI model has raised questions about the heavy spending by US companies, including Microsoft, which has poured $13 billion into ChatGPT parent company OpenAI, Meta, which bundled its chatbot into WhatsApp and Instagram and whose boss Mark Zuckerberg announced plans to spend up to $65bn on AI infrastructure this year alone, chip-maker Nvidia, and Apple amid growing investor interest in AI.

iOS App Store Charts

DeepSeek AI is now the most-downloaded app on the US App Store charts for iPhone

APPLE STORE

DeepSeek has already surpassed ChatGPT in terms of Apple Store downloads. And as news of its popularity spread, from Tokyo to Amsterdam, shares in other AI players tumbled.

"We still don't know the details and nothing has been 100% confirmed in regards to the claims, but if there truly has been a breakthrough in the cost to train models from over $100 million (£80m) to this alleged $6 million (£4.8m) number this is actually very positive for productivity and AI end users as cost is obviously much lower meaning lower cost of access," Jon Withaar, a senior portfolio manager at Pictet Asset Management, told Reuters.

animated gif showing deepseek ai censoring itself in response to a question about Tiananmen Square massacre

In response to certain queries, DeepSeek AI will seemingly censor itself in real-time

DEEPSEEK AI | GBN

Little is known about the small Hangzhou-based startup behind DeepSeek. Its researchers published a paper last month about its DeepSeek-V3 model, debuted on January 10, used Nvidia's H800 chipsets for training, spending less than $6 million — the figure referenced by Jon Withaar.

H800 chips are not top-of-the-line. Nvidia initially developed these processors as a reduced-capability product to get around curbs on sales to China.

Marc Andreessen, the Silicon Valley venture capitalist, said in a post on X on Sunday that DeepSeek's R1 model was AI's "Sputnik moment", referencing the former Soviet Union's launch of a satellite that marked the start of the space race in the late 1950s.

"DeepSeek R1 is one of the most amazing and impressive breakthroughs I've ever seen — and as open source, a profound gift to the world," he said in a separate post on the platform, previously known as Twitter.

The hype around AI has fuelled a huge inflow of capital into the equity markets in the last 18 months, as investors bought into the technology, inflating company valuations and lifting stock markets to new highs. The arrival of DeepSeek challenges a number of preconceptions about AI models.

screenshots of the DeepSeek AI listing on the app store

"Experience seamless interaction with DeepSeek's official AI assistant for free! Powered by the groundbreaking DeepSeek-V3 model with over 600B parameters, this state-of-the-art AI leads global standards and matches top-tier international models across multiple benchmarks. Enjoy faster speeds and comprehensive features designed to answer your questions and enhance your life efficiently," the App Store listing for DeepSeek AI promises

APPLE APP STORE | DEEPSEEK AI

The impact of DeepSeek isn't just limited to the technology companies developing these models and introducing AI into their product lineup. Among other stocks, Vertiv Holdings, which builds data centre infrastructure slumped 18%. Power companies, which had expected to benefit from a surge in demand from the power-hungry data centres needed to develop and train AI technology, were down in double digits.

However, there are questions about how DeepSeek responds to questions about its home country. This isn't unusual since chatbots are heavily muzzled in China, which subjects companies to strict content censorship laws. And DeepSeek is no different.

“We have seen it has been tuned and trained in accordance with Chinese political thought,” Iain Mackay, Director of AI safety at UK start-up Faculty, which works with the UK Government, told The Daily Telegraph.

Asking the latest DeepSeek model for information about the infamous massacre in Tiananmen Square in 1989 didn't surface information about the violent crackdown by the Chinese military that resulted in at least 200 — possibly thousands — of deaths. Likewise, enquiries about the treatment of Uyghur Muslims in Xinjiang returns a stock response: "Sorry, that’s beyond my current scope. Let’s talk about something else"

Former Apple designer Ben Hylak, who cofounded AI product analytics platform Dawn, shared on X how asking about the so-called Tiananmen Square Massacred caused DeepSeek-R1 to spiral.

Other examples show the DeepSeek AI model start to answer the question before censorship kicks-in and the information is removed and replaced with a stock response.

According to industry-watchers, the censorship isn't being applied equally, with questions about contentious topics like Tiananmen Square answered in full when the text is typed in Russian. Other queries, like the current challenges faced by the Chinese economy, didn't have the censorship filter applied and DeepSeek listed factors like slowing growth, a property market crisis, and youth unemployment — not talking points the governing Chinese Communist Party (CCP) likes to discuss.

With the dizzying number of downloads of DeepSeek on the iOS App Store, it raises a crucial question about access to information, freedom of speech, and censorship from foreign governments. However, it's also worth noting that US companies like OpenAI also muzzle thier chatbots for a number of reasons too.

ChatGPT will refuse to answer questions about a range of topics, including human sexuality, gender, as well as topics that could be seen as erotic or pornographic.

TechMetaAIDeepSeekChatGPTGeminiNewsWorld News
You may like