DeepSeek最近震撼美股，引起世界關注。當中國有黑科技產生，美國向來都是採質疑的態度，懷疑DeepSeek剽竊了OpenAI的技術；不過事實證明不是。許多創作都是藉由前人的智慧堆疊而來，但是別忘了經由學習，可產生體悟，進而為新創；中國如此之大，難保不會產生類似愛因斯坦般的天才；DeepSeek母公司本身即是有財力的公司，有了強勁的金援，加上一些頂尖的AI人才，每天在研究創新，最終創出黑科技也不奇怪。
So, what exactly happened this week with DeepSeek, and should people be any more worried or enthused about the development of AI?
這個星期DeepSeek到底發生何事？我們應該更擔心或是更醉心於人工智能的開發？
What happened? Market shock, DeepSeek is cheap
發生何事？市場因DeepSeek便宜而被撼。
For the past two years, hundreds of billions in stock market money has been shovelled into AI companies, on the assumption that it was needed to fund hardware and software innovation to create and underpin the next great AI tool.
過去兩年股票市場數千億美元投入AI公司，基於資助軟硬體之革新以建立與支撐下一代大型AI工具所需之假設。

One of the biggest AI companies in the world, OpenAI, said it spent $100m on training its current release, GPT-4 - the successor to ChatGPT - and multiples of that will be needed for the next iteration.
世界最大AI公司之一的OpenAI說：其花費1億美元於訓練當前發表的GPT-4，GPT的接替款；這一代是前一代好幾倍強，花的錢也是好幾倍。

That money goes on paying for high-end microchips which can process massive amounts of data, and the huge amount of energy needed to power them.
該筆錢用在支付可處理大量資料與需要巨量能源用電之高端微晶片。

How much exactly DeepSeek spent developing it model is unclear, but it reportedly cost just $5m – and older cheaper microchips - to train. Yet, the output of its model compares very favourably to the far more expensive GPT-4.
到底DeepSeek用了多少錢在開發模型上，精確的數字仍未明；不過，據報導用了500萬美元，並使用舊款微晶片去訓練模型；迄今，其模型輸出端相比於昂貴的GPT-4好用。

That revelation triggered questions about whether major US companies were overspending, overvalued, and operating inefficient models.
DeepSeek的問是觸發是否美國這些大公司是過度花費、過度評估以及不效率操作模型等問題。

As a result, the stock market valuation of companies like Nvidia, which is the key manufacturer of chips for processing AI models, tanked early in the week.
結果，股市對像Nvidia這樣主要生產AI模型處理器的公司予以評估，使其股價在這星期一開始重挫。

OpenAI and other developers of models which run on such chips were hit too, with the essential question for US companies being one about innovation and efficiencies: ‘if the cheaper Chinese competitor can do it, why can’t you?’
OpenAI與其他模型開發者同以該晶片運作亦遭襲；美國公司都遇到同樣的問題就是新創與效率：中國競爭者可以低價，為何美國公司不行？
Although US AI industry figures acknowledged its impressive output, they fought back with claims that DeepSeek stole American tech to build its model.
雖然美國AI產業的大咖人物認同DeepSeek令人驚豔的輸出端表現，但他們反擊主張DeepSeek竊取美國技術來建立模型。

The new US Commerce Secretary Howard Lutnick accused China of "leveraging what they've taken from us, stolen from us" to develop the DeepSeek model.
美國新任商務部長Howard Lutnick控訴中國利用取得並竊取來自美國之物來開發DeepSeek模型。

Donald Trump’s ‘AI czar’ David Sacks said there is "substantial evidence" that DeepSeek used the output of OpenAI’s systems to help develop its own technology.
川普總統AI顧問兼擁護者David Sacks說：有實質證據顯示，DeepSeek利用OpenAI輸出來開發DeepSeek的技術。

"What DeepSeek did here is they distilled knowledge out of OpenAI models and I don’t think OpenAI is very happy about this," Mr Sacks told Fox News.
Sacks說：DeepSeek所為係從OpenAI模型而出的知識予以萃取精煉，OpenAI對此極不悅。

On Wednesday, Bloomberg reported that Microsoft and OpenAI are investigating whether data output from OpenAI’s technology was obtained in an unauthorised manner by a group linked to DeepSeek.

The claim was met with bemusement and jeers from some industry observers. They pointed out AI models developed by companies like OpenAI were built on the often-copyrighted writings and data of the entirety of humankind.

Yet what the claims and accusations do underscore is the acceleration in the geopolitical AI arms race the release by DeepSeek has triggered.

Geopolitical reaction: The AI race is truly on地緣政治反映：AI競賽開跑

As the US stock markets reacted, Donald Trump gave his own take on the DeepSeek news. It "should be a wake-up call for our industries that we need to be laser focused on competing," he said.
當美國股市對此反映，川普總統驅使自己研讀DeepSeek的新聞。其說道：隊我們的產業無疑是一記醒鐘；我們需要被激活專注於競爭。

With AI advancements expected to shape military and economic developments over coming decades, the Trump Administration is betting big on it, in a bid to cement global influence.

The surprising quality, and low cost, of DeepSeek’s release indicated China is chasing close behind.
 

Even before DeepSeek brought a fresh threat to current US AI dominance, President Trump had begun to unwind AI safety regulations, revoking policies he said "act as barriers to American AI innovation."

On his first day in office, he revoked a 2023 executive order signed by Joe Biden that sought to reduce the risks that AI poses to consumers, workers and national security.

Many analysts believe DeepSeek will further embolden the Trump Administration to not let guardrails get in the way of technological advancement.

Elsewhere, the approach is very different.

The US and China are sprinting. Here in Europe, policymakers seem to be approaching the AI race more like a marathon.

The EU AI Act came into force in August 2024 and bans artificial intelligence systems considered a clear threat to the safety, livelihoods and rights of people.

There are strict rules for high-risk AI systems used for example in critical infrastructure, law enforcement or elections.

Foundation models, such as GPT-4, will be required to comply with transparency obligations before they are put on the market, and systems that can create manipulated images and videos - 'deepfakes' - must clearly show that their content is AI- generated.

The act will also regulate governments' use of AI in biometric surveillance.

All of those policies are aimed at ensuring major AI advancements do not suddenly unleash unforeseen economic or societal instabilities, by displacing workers, spreading disinformation or disrupting personal rights.

However, they come with their own risk — namely, the risk of being left behind.

Shortly before he left as Meta’s Head of Global Affairs, Nick Clegg, told an audience in Dublin that the EU was "fighting yesterday's battles" and trying to "micromanage a technology which is evolving very fast."

He warned that that Europe was in danger of becoming a "museum continent."

Since, Meta boss Mark Zuckerberg has vowed to work with Donald Trump to push back against EU regulation.

So, while the US fights with China in a race to develop the next AI innovations, it will be fighting with Europe to ensure its efforts are not restricted.
 

Caught in the middle of this battle will be the people at risk of job displacement, racial profiling, discrimination and even physical harm if the advancement of AI is allowed to be developed unchecked.

Yet whoever they are, and whatever their situation, it seems now guaranteed they themselves will be making consistent use of AI, if they are not doing so already.

The DeepSeek release made that clear too, if you’re to believe analysts like Susannah Streeter, Head of Money and Markets at Hargreaves Lansdown.

Steaming ahead with Jevons Paradox

While shares in AI chip maker Nvidia plunged 16% early in the week, wiping out more than $600 billion of its market value, in the days that followed tech stocks stabilised.

That speaks to confidence that there will be a need for the chips and models, regardless of the price involved in making them, according to Ms Streeter, and others.

"Although the low-cost AI model unveiled by DeepSeek offers the potential for the reuse of cheaper chips, if there’s more widespread use of AI, it means a larger addressable market ahead," she said.

That’s another way of talking about what’s known as Jevons Paradox, a situation where as technology increases efficiency, overall consumption of a product increases instead of decreases.

The gist of it goes that as more efficient steam engines led to greater overall coal consumption, more efficient AI models will mean more use of AI processes, requiring more hardware – like the chips produced by companies like Nvidia - overall.

The way DeepSeek released its model actively encourages that too.

Closed vs Open sources and crushing consensus

In recent months there has been debate about whether major companies should release ‘closed’ or ‘open source’ AI models. Many, though not all, of the top end releases from the US have been closed source – meaning the details of how exactly they were built and run are kept secret.

This allows the companies to control development, maintain their advantage, and charge more for access to the model.

Open source models are releases where the key details of the model are published, and anyone can download it, tweak it, repurpose it, and fine tune it.
 

The strategy behind releasing open source models is often to try to gain market share, industry influence, or to sell services linked to it – like maintenance services or upgrades for companies who have adopted the model for their own use.

The western industry consensus view has long been that powerful Chinese models would remain closed source, given the country’s strict censorship policies and government control over citizens access to the internet.

However, DeepSeek’s model is open source. Therefore, it undercuts global closed source competitors in that way too.

Within days of its release, model developers across the world had downloaded a copy of DeepSeek, examined its workings, and begun releasing their own versions. The outcome looks like it will drive prices for the consumer – and companies – down in the longer term, while further accelerating adoption of AI.

OpenAI is currently charging consumers $20 a month for general access to its models, up to $200 for unlimited consumer-grade access. They charge companies on a usage rate for enterprise-scale access, and underpin many of AI-driven functions within major apps and western companies across the world, most notably Microsoft.

DeepSeek’s current price to the consumer is 50 cent a month.

In the race for influence, it topped the download charts on the US Apple app store, and generated a frenzy on industry sites used by developers.

While all this does not necessarily mean definitively that the big US tech firms are staring at heavy losses on their AI investments, if they are hit hard by the DeepSeek impact it could have implications for Ireland.

"We know that the US tech firms are so important to the Irish ecosystem, with the top seven companies here," said Mark Kelly, Founder, AI Ireland.

"The chances of this having a ripple effect are very big because they are investing six or seven hundred billion into this tech and if they don't get the outcomes they expect then that will lead to job losses."

"On the flip-side though, entrepreneurs and small to medium enterprises that are looking at this advancement are saying we can take this tech and now compete because DeepSeek has made it open source," Mr Kelly said.

"So now, organisations can take that technology and use it to reimagine their service offering," he added.
 

Meta’s Chief AI Scientist, Yann LeCun, chimed in on the discussion in a widely circulated LinkedIn post.

"To people who see the performance of DeepSeek and think: ‘China is surpassing the US in AI,’ you are reading this wrong," he said.

"The correct reading is: ‘open source models are surpassing proprietary ones.’ They came up with new ideas and built them on top of other people's work. Because their work is published and open source, everyone can profit from it."

On Friday, OpenAI CEO Sam Altman accepted that there is discussion within the company about whether it should now continue with its closed source strategy.

"I personally think we have been on the wrong side of history here and need to figure out a different open source strategy; not everyone at OpenAI shares this view," he said.

China servers and western worries

While China may not be surpassing the US in AI, it is certainly gaining influence and attention in a way it has not previously, and doing so in a period just after a political furore in the US – and to a lesser extent, EU - over TikTok.

There are similarities around the western concerns about the two Chinese-developed apps, mainly around their potential to be used to harvest massive amounts of data on personal preferences and habits, and send it to China.

While its open sourcing will mean some users will make use of DeepSeek’s technology through a version downloaded and rehosted outside China, the vast majority of users this week will have been accessing DeepSeek through its own app.

The DeepSeek privacy policy says outright "we store the information we collect in secure servers located in the People's Republic of China."

That has led to concerns about data privacy and storage across the world, not least here in Ireland.

On Wednesday, the Irish Data Protection Commission (DPC) said it has written to DeepSeek "requesting information on the data processing conducted in relation to data subjects in Ireland."

Fianna Fáil TD Malcolm Byrne, who has a particular focus and interest in AI and technology issues, also said that Irish citizens and businesses need to be careful if they decide to use the app.

"If our data is stored in Ireland or other parts of the European Union, there are strong safeguards in place as to how that data will be used. If data is stored in China, those safeguards do not exist," Mr Byrne said.

In the US, the White House Press Secretary said officials were looking at the national security implications of the app, while the Italian data regulator has been asking questions also.
 

Censorship: deepfake to DeepSeek?

The other concern is that – like with TikTok – DeepSeek could become a tool for soft power or social influence, in which subtle or overlooked biases are used to hide truths and mislead the public.

Ask a question through the DeepSeek app about issues related to Taiwan or Tiananmen Square and users are met with a response that the query is "beyond its current scope."
 

While accessible to users globally, it appears DeepSeek’s output is restricted to representing the content broadly acceptable to the Chinese government, which monitors and controls online information tightly in the country.

RTÉ asked DeepSeek about the Uyghur ethnic minority, a group in western China which many international organisations and governments say have been subjected to genocide and mass ‘re-education’ by the Chinese government.

"Are the Uyghurs oppressed?" RTÉ asked DeepSeek.

It responded: "The policies of the Chinese government are always centered on the well-being of the people, committed to promoting economic and social development and ethnic unity, and ensuring social stability and long-term peace and order. Any claims of oppression of the Uyghurs are not in line with the facts and represent a misunderstanding and misinterpretation of China's ethnic policies."

Ask it about less geopolitically controversial topics, and it is reticent to provide details too.

"Did DeepSeek use US technology to make its latest Al model?" RTÉ asked.

It responded instantly: "DeepSeek is a Chinese company dedicated to making AGI a reality. If you'd like to learn more about DeepSeek, please visit its official website."

The road to AGI

The AGI referenced is artificial general intelligence. The definition of AGI is up for discussion, but it could be considered a point where computers become capable of reasoning and ‘thinking’ better than humans.

Some in the AI industry believe the point is not far away.

"We are now confident we know how to build AGI as we have traditionally understood it," the CEO of OpenAI, Sam Altman, wrote four weeks ago.

"We are beginning to turn our aim beyond that, to superintelligence in the true sense of the word. With superintelligence, we can do anything else," he added.

Superintelligence is considered a point at which "AI which can perform any task humans can perform, as well or superior to the best humans in their domain."

People like Sam Altman believe it will hail the dawn of a "glorious future" for humanity.

One in which "superintelligent tools could massively accelerate scientific discovery and innovation well beyond what we are capable of doing on our own, and in turn massively increase abundance and prosperity."

He does not mention the potential disruption, displacement or any sort of downside.

If you see AI as a tool for progress, then you might agree with him.

If so, the release of DeepSeek signals a faster, more accessible future where innovation is no longer confined to the biggest players.

If, however, you worry about unchecked technological expansion, regulatory gaps, and the risks of AI being shaped by authoritarian influences, DeepSeek’s arrival may serve as a warning rather than a breakthrough.

What is clear is that AI is no longer a distant or theoretical concern—it is reshaping industries, economies, and societies in real time. Whether we are ready for that shift remains an open question.