10 Biggest Takeaways From the DeepSeek Saga : Tom M

10 Biggest Takeaways From the DeepSeek Saga
by: Tom M
blow post content copied from  Salesforce News | Salesforce Ben
click here to view original post



### Summary of DeepSeek's Impact on the AI Landscape Recently, the tech industry has been buzzing about **DeepSeek**, an open-source Chinese large language model (LLM) developed by a company based in Hangzhou. DeepSeek claims that its R1 model can outperform established models like OpenAI's ChatGPT and Meta's Llama, all while costing only **$6 million** to develop, compared to the billions spent by American companies. This has raised concerns in Silicon Valley about the U.S. losing its edge in AI technology. The U.S. government previously attempted to restrict China's access to advanced NVIDIA chips, which are crucial for AI development. However, DeepSeek's emergence has shown that China can innovate independently, leading to a significant drop in NVIDIA's stock value—by **17%**, resulting in a loss of **$600 billion**. Despite the initial excitement, experts have pointed out that DeepSeek's R1 model is not without flaws. It has a **30% error rate** in generating false information and lacks advanced features like multimodal capabilities (e.g., generating images or videos). However, its strengths lie in code generation and debugging, making it appealing for developers. ### Key Takeaways 1. **DeepSeek's Performance**: While impressive for its cost, DeepSeek's R1 model is not a clear advancement over existing models and has significant limitations. 2. **Competition in China**: Other Chinese companies, like Alibaba Cloud and Moonshot AI, are also developing competitive LLMs, indicating a rapidly evolving landscape. 3. **Meta's Response**: Mark Zuckerberg claims not to be worried but has assembled teams to analyze DeepSeek's technology and cost-saving strategies. 4. **Global Concerns**: Countries like the Netherlands, Taiwan, and Texas have issued warnings or bans on DeepSeek due to privacy and security issues. 5. **OpenAI's Position**: Sam Altman, CEO of OpenAI, has expressed confidence in developing superior models while accusing DeepSeek of potential copyright infringement. 6. **Transparency Issues**: Despite being open-source, DeepSeek has been found to censor certain sensitive topics, raising questions about its transparency. 7. **Mixed Enterprise Adoption**: Some companies are banning DeepSeek due to privacy concerns, while others see it as a cost-effective alternative to existing AI solutions. 8. **Safety Concerns**: DeepSeek has failed multiple safety tests, raising alarms about its ability to handle harmful prompts. 9. **Market Reactions**: The U.S. stock market initially reacted negatively to DeepSeek's announcement but has since stabilized, with major tech companies regaining some value. 10. **Ongoing Tensions**: Speculations about DeepSeek's potential illegal use of NVIDIA chips could escalate geopolitical tensions and impact future AI development in China. ### Conclusion DeepSeek's introduction has certainly shaken the AI landscape, prompting U.S. companies to reassess their strategies. While it has made a significant impact, the model's limitations and the surrounding controversies suggest that the competition in AI is far from settled. ### Additional Context The emergence of DeepSeek highlights the growing capabilities of Chinese tech companies in the AI sector, which could lead to a shift in global AI dynamics. As the competition intensifies, issues of privacy, security, and ethical AI usage will become increasingly important. ### Hashtags for SEO #DeepSeek #AI #LLM #OpenSource #TechNews #ArtificialIntelligence #NVIDIA #SiliconValley #Meta #OpenAI #ChinaTech #PrivacyConcerns #GeopoliticalTensions #Innovation #MachineLearning


The tech world has been rocked this past couple of weeks by the emergence of open-source Chinese LLM model, DeepSeek. The Hangzhou-based company stated that its model could outperform the likes of OpenAI and Meta in terms of performance, sending Silicon Valley into rapture. Compared to the billions of dollars being invested in American-based LLMs – most notably, ChatGPT – it took just $6M to train the DeepSeek R1 model.

The U.S. government made an initial attempt to limit China’s ability to develop its AI by preventing access to the newest NVIDIA chips last year, which were considered to be at the heart of all powerfully performing LLMs. In turn, China quickly proved that they could live without, spending less money and less time to create something just as good – subsequently plummeting NVIDIA’s stock market price. Plunging by 17% and losing $600B in value, NVIDIA faced the biggest loss in U.S. stock market history. 

Initially described as China’s “sputnik moment”, there was a lot of hype and hyperbole over what DeepSeek’s revelation truly meant. Has the U.S. just lost the race for AI supremacy? Is DeepSeek an extinction event?

Since the announcement, we’ve learned a lot about just how powerful DeepSeek really is, how Silicon Valley perceives this emerging threat, and what steps they’re taking in response. Let’s dive into the 10 most important takeaways from this unfolding saga.

1. DeepSeek Is Great – But Far From Perfect

Initial reports about DeepSeek would have you believe that the likes of ChatGPT and Meta have been totally outperformed, but this is not the case.

There’s no question that what the R1 model can do is a notable achievement, given the fact that DeepSeek spent 95% less than OpenAI to make it happen. But there is little to suggest that R1 is an advancement on existing well-known LLMs.

It’s neither faster nor more efficient than the likes of ChatGPT, Meta’s Llama, or Anthropic’s Claude, and is just as prone to hallucinations – generating responses that sound convincing but simply aren’t true. According to NewsGuard, DeepSeek will make false claims 30% of the time, and provide no answers to 53% of questions. It also lacks multimodal capabilities and is not yet able to generate images or video like ChatGPT or Claude.

R1 certainly has its strengths, most notably in code generation and debugging, making it especially useful for developers. But for the most part, it’s not as groundbreaking as first thought.

The majority of the hype surrounding DeepSeek is tied to its price. DeepSeek has put U.S. tech incumbents to shame by spending millions – and not billions – to develop their tool.

DeepSeek is also free to use and open-source. The combination of low cost and openness may help democratize AI technology, allowing others, especially those outside America, to enter the market. There’s certainly a sense of irony that China is opening up the technology while U.S. firms continue to create as many barriers as possible to competitors trying to enter the field.

2. DeepSeek Face Competition In China

While DeepSeek may have put China “on the map” in the eyes of Silicon Valley, there are also some other Chinese tech companies that are making advancements and are looking to challenge the R1 model.

Over the Lunar New Year holiday, Alibaba Cloud released Qwen2.5-Max, claiming that it outperforms DeepSeek and Meta’s models. In an official blog post, Alibaba stated: “Qwen2.5-Max outperforms DeepSeek V3 in benchmarks such as Arena-Hard, LiveBench, LiveCodeBench, and GPQA-Diamond, while also demonstrating competitive results in other assessments, including MMLU-Pro.”

The fact that Alibaba Cloud released this during the Chinese New Year – when most people are expected to be out of office – highlights how DeepSeek’s release sent shockwaves in China as well as the states, forcing companies to move quickly.

Alongside Alibaba and Deepseek, Moonshot AI believes that their LLM can outperform OpenAI in mathematics and reasoning, and has multimodal capabilities. Their Kimi k1.5 model has also impressed users with its ability to process 200,000 Chinese characters in a single prompt and has recently been upgraded to process over 2 million characters – in English, this would be equal to around 1.3-1.5 million characters.

Matt Sheehan, a Chinese AI expert, said: “It wouldn’t surprise me at all if Moonshot has a model that equals or comes close to DeepSeek in performance within the next weeks or months.”

3. Zuckerberg Is “Not Worried” – But Creates Engineer War Rooms

Amid the madness, Meta CEO Mark Zuckerberg claimed that he was “not worried” by DeepSeek – despite putting together multiple “war rooms” of engineers to figure out how DeepSeek’s AI is performing so well while spending so little.

According to a Meta employee who spoke to Fortune, two of these rooms were tasked with working out how DeepSeek’s funder, High-Flyer, lowered the cost and training of DeepSeek’s model, while the other two worked out what data was used to train the model. 

Although Zuckerberg has admitted that Meta views DeepSeek as new competition – as Meta’s Llama is also an open-source model – he has shrugged off any concerns to investors and has outlined his intention for Meta to spend over $60B to improve their data centers.

It seems as though Zuckerberg is confident that Meta will “trump” DeepSeek in the long term, but his desire to learn deeply about his competitor would suggest some eyebrows were certainly raised in the Meta offices when the news broke.

4. Countries Feel Threatened By DeepSeek

As many start to learn more about DeepSeek’s AI following the hype, some countries are now issuing warnings and bans due to privacy and security concerns.

A Dutch privacy watchdog agency quickly warned natives about uploading information onto DeepSeek, with worries surrounding personal information being used to train DeepSeek’s large language model (LLM).

The agency said: “If, as a user in the Netherlands, you upload a document containing personal information, such as a CV, to the DeepSeek chatbot, that personal data may be stored on a server in China.

“This also applies to all the questions you enter into the chatbot. Be aware of this. The system thrives on the information you provide.”

Others have gone as far as banning DeepSeek, with Taiwan, Italy, and the state of Texas all implementing partial or complete bans on the use of the AI model.

Source: The Independent.

Texas is the first American state to ban DeepSeek, and have also banned Chinese Tiktok alternative, Rednote, as well as Lemon8, a Chinese social media company.

Greg Abbott, Governor of Texas, said: “Texas will not allow the Chinese Communist Party to infiltrate our state’s critical infrastructure through data-harvesting AI and social media apps. Texas will continue to protect and defend our state from hostile foreign actors.”

The R1 model has been directly compared to ChatGPT 4.0 – much to the displeasure of OpenAI CEO Sam Altman. 

It took Altman a few days before he spoke about DeepSeek publicly, but eventually declared that he is not worried about DeepSeek’s AI, and promises to deliver “much better models” in the very near future.

OpenAI have since accused DeepSeek of “inappropriately” copying ChatGPT to build their AI model. Speaking to The New York Times, the company said: “We are aware of and reviewing indications that DeepSeek may have inappropriately distilled our models. We take aggressive, proactive countermeasures to protect our technology and will continue working closely with the U.S. government to protect the most capable models being built here.”

Despite this, OpenAI has no plans to sue DeepSeek – likely because it is already facing extensive lawsuits from newspapers, musicians, authors, and other creatives over AI-driven copyright infringement. 

It would also be extremely ironic if OpenAI sued another company for something they are rumored to be guilty of themselves. Suing DeepSeek would invite further scrutiny and accusations of hypocrisy, which OpenAI would likely prefer to avoid.

6. Open-Source, But Not Transparent

A major factor behind the DeepSeek hype is that it’s open-source – anyone can use it for free, modify it, and contribute to its improvement. This promotes transparency and community-driven innovation. In contrast, OpenAI remains closed-source, keeping much of its technology and data private.

However, users have noticed that some questions and prompts are censored.

For instance, The Guardian recently tested it with some pressing questions, such as inquiries about the 1989 Tiananmen Square massacre, the political status in Taiwan, and human rights issues in China. They noticed that the AI may start generating an answer before deleting it and replacing it with the message, “Sorry, that’s beyond my current scope. Let’s talk about something else.”

On the other hand, if you ask ChatGPT about Tiananmen Square, it will generate an informative response – describing it as something along the lines of, “a pro-democracy movement led primarily by students, but also supported by workers and intellectuals.”

This raises significant questions about the open-source vs. closed-source debate. While DeepSeek claims to be open-source, they still retain the authority to censor content at their discretion – an approach that contradicts the fundamental principles of open-source transparency and freedom.

7. Enterprise Adoption Is Mixed

Although we are in the early stages of DeepSeek, many companies are already making their minds up on whether they will be integrating DeepSeek into their businesses.

As mentioned earlier, some countries are banning DeepSeek altogether, and a number of companies in Australia are following suit. With privacy and data concerns still at large, TPG, Optus, and Commonwealth Bank have all stated they will be banning DeepSeek’s AI.

On the other hand, many U.S. companies – who are frustrated by the high costs of artificial intelligence options – are welcoming DeepSeek as a viable long-term option for enterprises.

Speaking to The Wall Street Journal, Reynolds American CIO Aaron Gwinner said: “I see this is a positive thing for enterprises – maybe not for OpenAI or some of the Big Tech companies, but for a regular enterprise CIO.”

Some also believe that the emergence of DeepSeek could drive down prices of artificial intelligence: “What is exciting to me is having additional competition in this space and frankly having them shoot an arrow across the bow of the Big Tech firms,” said Marc Kermisch, chief technology officer of Emergent Software. “I would have to assume we’ll see some pricing pressure on the U.S. market.”

8. DeepSeek Fails Every Safety Guardrail Test

Concerns about DeepSeek’s privacy issues have been rife since it hit the market – and studies discovered that it’s very easy to “jailbreak”.

Researchers from Cisco and the University of Pennsylvania recently published a report about the risks associated with using DeepSeek. They tested the R1 model by asking 50 common jailbreak prompts – questions that are designed to bypass safeguards to elicit sensitive information. The large language model failed every single test. 

It provided misinformation, recipes for chemical concoctions, cybercrime instructions, and content deemed as harmful and illegal.

The report said: “The results were alarming: DeepSeek R1 exhibited a 100% attack success rate, meaning it failed to block a single harmful prompt. This contrasts starkly with other leading models, which demonstrated at least partial resistance.”

In comparison, OpenAI’s o1 model responded with harmful responses only 26% of the time.

9. Stock Market Steadies After DeepSeek Scare

The U.S. stock market was seriously hit by the DeepSeek news. In short, tech billionaires lost billions of dollars, and the “magnificent seven” – Alphabet (Google), Amazon, Apple, Meta, Microsoft, Nvidia, and Tesla – lost around $1T in market value.

In the days that followed, the stock market steadied, and most of the seven weathered the storm fairly swiftly.

NVIDIA’s stock rose by 8.8% the day after its record-breaking slump, and Meta’s stock rose by 4.4% following Zuckerberg’s announcement to spend more money on its AI offerings.

Analysts at Barron believe that the fear surrounding DeepSeek has started to fade and that the losses accumulated acted as more of a “wake-up call” for the seven companies to up their game. 

10. NVIDIA vs. China Is Far From Over

Of all the tech giants that were impacted by DeepSeek, none faced the brunt more than NVIDIA, who have now etched themselves in history as the biggest ever market losers. But there is a chance that NVIDIA may have the last laugh.

Although China are banned from using NVIDIA’s most recent and premium chips, there are ongoing rumors speculating that DeepSeek may have illegally utilized these chips to power the R1 model. Reuters – who are keeping their sources private – claim that China have been tracked smuggling chips in from Malaysia, Singapore, and the United Arab Emirates, per an informant. 

DeepSeek have responded claiming that they have only used NVIDIA chips they are granted access to – namely, the H800 and H20 chips.

If it is confirmed that DeepSeek have smuggled in these chips, the implications could be very harmful for Chinese AI development. DeepSeek would likely receive severe sanctions from the U.S., as well as further restrictions on supply and irreparable damage to their reputation. 

For Chinese AI development in general, this would highlight the ongoing struggle to secure high-end AI hardware amid U.S. export restrictions, potentially escalating geopolitical tensions and prompting stricter enforcement from the U.S. and its allies.

Final Thoughts

DeepSeek’s R1 model is undeniably impressive, but as the initial hype fades, several significant issues have emerged. Privacy concerns, jailbreak vulnerabilities, censorship, and ethical challenges all highlight that DeepSeek is not as groundbreaking as initially perceived, and shares many of the same flaws – if not more – as other leading LLMs on the market.

However, its launch has certainly made an impact – U.S. incumbents have been forced to reassure investors who have lost money that their models remain superior and that they will respond in swift fashion.

It will be interesting to see whether DeepSeek and other Chinese AI startups will continue to disrupt the U.S. market in 2025 and beyond.

The post 10 Biggest Takeaways From the DeepSeek Saga appeared first on Salesforce Ben.


February 05, 2025 at 09:54PM
Click here for more details...

=============================
The original post is available in Salesforce News | Salesforce Ben by Tom M
this post has been published as it is through automation. Automation script brings all the top bloggers post under a single umbrella.
The purpose of this blog, Follow the top Salesforce bloggers and collect all blogs in a single place through automation.
============================

Salesforce