DeepSeek's Breakthrough Sparks Global AI Debate 'wake-up call' for US tech firms, Trump says

[Social Media]

 

The emergence of Chinese tech firm DeepSeek and its groundbreaking AI model has ignited heated discussions across the global tech industry, with US President Donald Trump describing it as "a wake-up call" for American companies. DeepSeek's advancements have sent shockwaves through Wall Street, triggering a sharp decline in shares of major US tech firms, including Nvidia, which saw its market value drop by nearly $600 billion (£482 billion).  

 

What has truly unsettled the industry is DeepSeek's claim that its R1 model was developed at a fraction of the cost compared to its American counterparts. This has raised pressing questions about the future of US dominance in AI and the sustainability of the massive investments currently being made by American firms. DeepSeek’s cost-effective approach has introduced a new paradigm, potentially reshaping how AI innovation is pursued globally.  

 

In just a week since its launch, DeepSeek has already become the most downloaded free app in the United States, underscoring its immediate impact. Despite the disruption, President Trump has offered a cautiously optimistic perspective. "If you could do it cheaper, if you could do it [for] less [and] get to the same end result. I think that's a good thing for us," he said while addressing reporters aboard Air Force One. Trump also expressed confidence that the US would maintain its leadership in the AI sector, stating that he was not concerned about China's technological advancements.  

 

DeepSeek's AI success is powered by its open-source DeepSeek-V3 model, which, according to its researchers, was trained for approximately $6 million (£4.2 million). This figure is significantly lower than the billions spent by rival developers in the United States. However, these cost claims have been met with skepticism by other players in the AI field, sparking further debate about the company’s methods and technological processes.  

 

The rise of DeepSeek comes amidst heightened tensions between the US and China over technology, with Washington imposing restrictions on the sale of advanced AI-related chip technology to China. In response, Chinese developers have turned to innovative strategies, including greater collaboration and new approaches to AI development. These efforts have yielded models requiring far less computing power than previous iterations, drastically reducing development costs and creating a potential industry upheaval.  

 

The market reaction has been dramatic. While the FTSE 100 index of major UK companies showed resilience, rising 0.46% on Tuesday, US markets experienced volatility. Futures on the Nasdaq, a tech-heavy index, showed slight recovery with a 0.1% increase, and Nvidia shares ticked up in after-hours trading. However, Japanese AI-related firms such as Advantest, Softbank, and Tokyo Electron faced sharp declines, pushing Japan’s Nikkei 225 index down by 1.4%.  

 

The founder of DeepSeek, Liang Wenfeng, has become a focal point of attention. Liang, a 40-year-old information and electronic engineering graduate from Hangzhou, China, launched the company in 2023 with backing from a hedge fund he also founded. Recently, Liang attended a meeting with industry leaders and Chinese Premier Li Qiang, further cementing his influence in the tech world.  

 

In a July 2024 interview with *The China Academy*, Liang expressed surprise at the global reaction to his earlier AI model. "We didn't expect pricing to be such a sensitive issue," he said. "We were simply following our own pace, calculating costs, and setting prices accordingly."  

 

As the world watches DeepSeek’s rise, the implications for AI development, global competition, and market dynamics are profound. While the US remains confident in its dominance, the rapid evolution of cost-efficient AI models may redefine the rules of the game in the tech industry.

 

 

Based on a report by BBC 2025-01-29

 

 

[scubascuba3]

Seems to be giving better answers than Gemini, I'm using both and comparing 

[sikishrory]

Great that it's open source but they need to address the privacy policy concerns and add an opt-out option.

Until then it is only safe to run locally (ollama) or via a third party that processes data in a better way.

 

Particularly:

Where We Store Your Information
"The personal information we collect from you may be stored on a server located outside of the country where you live. We store the information we collect in secure servers located in the People's Republic of China."

 

and

 

"The Services are supported by certain entities within our corporate group. These entities process Information You Provide"

[thaibeachlovers]

9 hours ago, Social Media said:

with US President Donald Trump describing it as "a wake-up call" for American companies.

Oh dear oh dear. Seems those pesky Chinese have upset the west's rich that were hoping for a big payoff.

Never mind, they'll be wearing grooves in the footpath to the W H to implore Trump to ban it.

 

[thaibeachlovers]

4 minutes ago, sikishrory said:

Great that it's open source but they need to address the privacy policy concerns and add an opt-out option.

Until then it is only safe to run locally (ollama) or via a third party that processes data in a better way.

 

Particularly:

Where We Store Your Information
"The personal information we collect from you may be stored on a server located outside of the country where you live. We store the information we collect in secure servers located in the People's Republic of China."

 

and

 

"The Services are supported by certain entities within our corporate group. These entities process Information You Provide"

If you use facethingy or suchlike don't they already steal your info?

IMO anyone that assumes anything they put on the internet is secure needs to wake up and smell the coffee.

[Paul52]

7 hours ago, scubascuba3 said:

Seems to be giving better answers than Gemini, I'm using both and comparing 

Gemini is a very low bar for comparing AI models.  ChatGPT is the leader.  DeepSeek is actually a line for line copy of OpenAI’s ChatGPT.  When DeepSeek was asked what AI model it is (frontier model) it said it was created by OpenAI based on  the GPT-4 architecture. DeepSeek also just suffered a large-scale cyberattack raising questions about how much work the Chinese put into making it secure and limiting vulnerability to it being manipulated by malicious bad actors 

[Snackbar]

$6M is the management accounting cost of distillation.

 

Training model costs $Bs

[Paul52]

7 minutes ago, thaibeachlovers said:

Oh dear oh dear. Seems those pesky Chinese have upset the west's rich that were hoping for a big payoff.

Never mind, they'll be wearing grooves in the footpath to the W H to implore Trump to ban it.

 

I think it’s the Chinese who are feeling a bit sore today.  DeepSeek just suffered a large scale cyberattack.  It’s not just a ChatGPT lookalike, it actually is ChatGPT (a line for line copy), but doesn’t have the security of OpenAI’s ChatGPT.  The US markets have bounced back. Panic over.

[John Drake]

I was able to use DS for about four hours yesterday afternoon. Since then, going until just a few moments before I posted this, all I get is "The Server is Busy. Please try again later." Maybe building one of these things on the cheap isn't the groundbreaking opportunity people think it is. BTW, DS will tell you it is only trained on content through October 2023, whereas the other AIs are up to date as of this very minute.

 

[thaibeachlovers]

6 minutes ago, John Drake said:

I was able to use DS for about four hours yesterday afternoon. Since then, going until just a few moments before I posted this, all I get is "The Server is Busy. Please try again later." Maybe building one of these things on the cheap isn't the groundbreaking opportunity people think it is. BTW, DS will tell you it is only trained on content through October 2023, whereas the other AIs are up to date as of this very minute.

 

When an AI creates a cure for cancer I'll agree it's worthwhile. Till then it's just higher tech that is making the 1% richer.

As for myself I already have enough of my info stolen to not be using something even worse.

[John Drake]

2 minutes ago, thaibeachlovers said:

When an AI creates a cure for cancer I'll agree it's worthwhile. Till then it's just higher tech that is making the 1% richer.

As for myself I already have enough of my info stolen to not be using something even worse.

 

I have to say that I was stunned at its effectiveness. I took my own published articles and asked DS the same questions I asked myself and I received an outline that very closely approximated my work. It even gave me some new ideas. Unfortunately, I can't get back in to DS in order to continue.

[candide]

Deepseek is just the tip of the iceberg. China has been investing in AI for years. There are several other AI initiatives in China.

If Trump were not ignorant he would be aware of it...

 

Below a 'pre-Deepseek' report on AI in China (August 2024)

https://itif.org/publications/2024/08/26/how-innovative-is-china-in-ai/

 

Quote:

 

The reality is the United States’ efforts to hold back China’s AI progress are unlikely to succeed. China is advancing rapidly in AI research and application, challenging the United States’ dominance in this critical field. Despite hurdles, China’s robust academic foundations, innovative methodologies, and increasing foreign investment are propelling it toward becoming a leading AI powerhouse. The question is not whether the United States can contain China in AI, but whether it can keep ahead

 

[John Drake]

20 minutes ago, John Drake said:

 

I have to say that I was stunned at its effectiveness. I took my own published articles and asked DS the same questions I asked myself and I received an outline that very closely approximated my work. It even gave me some new ideas. Unfortunately, I can't get back in to DS in order to continue.

 

I do wonder if DS/ChatGPT are operating off my abstracts. I know they have access to them. I've verified that before.

[proton]

why does deepseek want email and phone numbers?

[candide]

17 minutes ago, proton said:

why does deepseek want email and phone numbers?

Same as the other online services. In addition, I think they particularly need to distinguish between Chinese numbers and other numbers, as the Chinese may be subject to specific restrictions.

 

Let's see if they can pass the test of compliance to EU laws, which are the most protective of privacy.

 

Don't worry, all data centers in China are guarded and supervised  by the Chinese security forces. What could go wrong? (Sarcasm alert)

[James105]

1 hour ago, Paul52 said:

I think it’s the Chinese who are feeling a bit sore today.  DeepSeek just suffered a large scale cyberattack.  It’s not just a ChatGPT lookalike, it actually is ChatGPT (a line for line copy), but doesn’t have the security of OpenAI’s ChatGPT.  The US markets have bounced back. Panic over.

 

Is that so?  I was quite surprised that they came up with something of their own but guess not.  I guess the old mantra of USA innovates, China imitates and Europe (EU) regulates is still valid then.   

[John Drake]

29 minutes ago, proton said:

why does deepseek want email and phone numbers?

 

Good question. The Chinese can get all that and more already through Thai Immigration, I'd think.

[scottiejohn]

1 hour ago, John Drake said:

 

I have to say that I was stunned at its effectiveness. I took my own published articles and asked DS the same questions I asked myself and I received an outline that very closely approximated my work. It even gave me some new ideas. Unfortunately, I can't get back in to DS in order to continue.

Have you tried doing the same with the other major US AI's?

[John Drake]

5 minutes ago, scottiejohn said:

Have you tried doing the same with the other major US AI's?

 

Yes. ChatGPT was just as thorough. Grok was even better.

 

[scottiejohn]

3 minutes ago, John Drake said:

 

Yes. ChatGPT was just as thorough. Grok was even better.

 

Thanks!

I thought that would be the case!

[John Drake]

4 minutes ago, scottiejohn said:

Thanks!

I thought that would be the case!

 

DeepSeek will tell you that its answers are generalist in nature, and it cannot provide specialist level sourcing. It also informs you that it has only been trained through content available as of October 2023. So, it's more than a year behind all the others, which provide content updates almost until the very day you ask the question. I would pay money for Grok.

[jingjai9]

Maybe the pot is calling the kettle black - Elon Musk is always honking his horn how the federal government is not cost efficient and so on. If Deepseek can develop the AI so much cheaper, perhaps old Elon, Mark and the rest of the tech geniuses should check their own recipes. 

[John Drake]

Alibaba out today with a new AI called Qwen. Slower than all the others, including DeepSeek. Trained content also cuts off at October 2023 like DS. And looks exactly like ChatGPT.