China’s DeepSeek - A Balanced Overview

• Shares in advanced computer chip companies are plummeting.
• China's DeepSeq1Real AI model has been unveiled with remarkable capabilities.
• The competition in the AI field has shifted from companies to nations.
• The U.S. is investigating potential national security threats posed by DeepSeq.
• This rapid advancement in AI may lead to significant changes in the industry.
• The implications of this development extend to economic, technological, and geopolitical realms.

Shares in the advanced computer chip companies are plummeting. One of the worst outcomes follows a billion-dollar investment in the Chinese company. Hi, welcome to another episode of Cold Fusion.

Look at these stock market charts from January 28, 2025. What you're looking at is a bloodbath—a bloodbath in the U.S. stock market of over $1 trillion. This comes with the release of the DeepSeq1Real AI model from China. The Chinese model is as capable as the best U.S. models but is free to use, open-source, more efficient.

Most shocking of all, it reportedly costs less than 3% of ChatGPT to develop. Just two years ago, we were discussing an AI arms race between companies. Today that has evolved into an AI race between countries.

In one corner, we have the United States—they have a long history of technological dominance. On the other side, we have China, a country with a very different ideology and motives. In this race for dominance, it's not about weapons, but it's about developing systems designed to think: artificial intelligence.

This race is reminiscent of the Cold War. Some have dubbed these events as, quote, the Sputnik moment of AI. The White House states that they're looking into the national security implications of China's DeepSeq AI platform.

To top it all off, OpenAI has accused DeepSeq of stealing its IP to train their own model. It's heating up. With the U.S. pouring in half a trillion dollars into the Stargate AI project, the global race is on. This ongoing battle could be one of the biggest stories in tech this year.

As artificial intelligence becomes a matter of national security, the technology will have to move faster than it is today. What a crazy time to be alive! But what is really going on here? How did a company from nowhere accomplish this? Is this all just part of the AI hype cycle, or is it the real deal?

It seems like the whole world is playing catch-up since the release. Historically, when technology meets a national security threat from an ideological opponent, we see inventions like the computer and jet aircraft emerge from World War II competition.

This time, the U.S. was largely unchallenged in AI until January 20, 2025, with the release of 1Real Deepseek. 1Real, which is free, has performance reportedly on par with OpenAI's $200 a month model concerning tasks like language, reasoning, mathematics, and coding. The free model surpasses Anthropic's Claude and Google Gemini.

But many may not know that DeepSeek operates somewhat differently than the current state-of-the-art models; this is partly why it's so efficient. Users have been flocking to it, with DeepSeek becoming number one in Apple's App Store.

Here are some stats behind the jaw-dropping performance: the AI was built in two months and reportedly cost less than $5.6 million to build. The AI company Anthropic states that the general amount needed to develop an AI system from scratch is between $100 million to $1 billion. To this end, Meta plans to spend $65 billion on AI.

Creating something that performs so well for just $5.6 million is groundbreaking, but all may not be as it seems. There are two very important points regarding DeepSeek AI and how it's being interpreted on Wall Street. The first is: it doesn't matter if it's a Chinese government op or not.

The technological innovation of having an LLM train itself through reinforcement learning is impressive. The cost efficiency of doing inference with only 7 billion parameters instead of 700 billion parameters is remarkable. This could mean an acceleration of AI everywhere in the economy.

Being open-source means that DeepSeek's code is freely available for anyone to use and modify. This is in stark contrast to OpenAI's approach, which is pretty ironic. This situation poses a significant problem for U.S. AI companies as their costs get unbalanced.

DeepSeek, with its 671 billion parameters, can run locally on a stack of M4 Mac Pros. In contrast, American companies have poured billions of dollars into servers after the shock of this release. The concern now is that U.S. companies may have spent too much money and used too much energy, overcharging for their services.

Moving forward, it’s possible that the applications running on top of AI will be more lucrative than the models themselves. The response of the market reflects this uncertainty, leading to a sell-off.

A bright spot for U.S. companies might be that users may hesitate to give their data directly to China, especially in corporate settings. Sam Altman, CEO of the ChatGPT maker OpenAI, announced that their GPT-300 mini model will now be given away for free. Meanwhile, Mark Zuckerberg and Meta are reportedly in a state of panic.

Over in China, the impact is similar. Tech giants like ByteDance (TikTok's parent), Alibaba, and Tencent are facing pressure to cut the prices of their AI models to compete. Despite DeepSeek's low pricing, it remains profitable while its rivals face losses.

Interestingly, OpenAI stated they have evidence that DeepSeek was using outputs from ChatGPT to train its own model. They even blocked several OpenAI API accounts suspected to belong to DeepSeek, suggesting possible theft. The U.S. government's official position indicates potential IP theft has occurred.

Amid U.S. sanctions, it also seems Chinese AI developers continue acquiring top-tier NVIDIA graphics cards. Who are the minds behind DeepSeek? Not much is known about the founder and team, but the story is intriguing.

Liang Wenfang, DeepSeek's founder, comes from a finance background and co-founded a hedge fund called High Flyer. His firm successfully utilized AI to predict market trends, managing around $8 billion. After achieving initial success, Liang aimed to build human-level AI.

In 2021, he began purchasing thousands of NVIDIA GPUs just before the Biden administration limited U.S. exports of AI hardware to China. Eventually, Liang transformed his AI side project into what we know today as DeepSeek, with R1 as their latest model.

The more I read about Liang's story, the more fascinating it becomes. DeepSeek1Real was trained using reinforcement learning, meaning it learned without human assistance. Their model architecture differs from most other players, utilizing a technique called mixture of experts.

Sky News explains this well: while OpenAI's latest model, GPT-4O, attempts to be a jack-of-all-trades, DeepSeek functions like a university divided into expert departments. It allows the AI to decide which type of query to address, routing it only to the relevant part of its digital brain, thus conserving time and computing power.

This method includes distillation, where larger models train smaller ones in specific domains. The outcome showcases equivalent performance with significantly reduced computing power, which shook not only AI developers but also the financial markets.

OpenAI's method to depict chain of thought reasoning differs fundamentally, as they keep their process closed. On the other hand, DeepSeek's model is fully public, demonstrating its reasoning processes openly and trained with markedly less data.

However, the claim of developing their model for $5.6 million may be misleading. In a paper released by DeepSeek, this figure pertains only to the official training of DeepSeq v3, excluding costs for prior research and experiments. This raises questions about the veracity of that headline figure.

In recent news, DeepSeek also launched an open image model, and reports suggest a video model may follow, potentially rivaling OpenAI's Sora or Google's anticipated VO2. Presently, DeepSeek is now outpacing ChatGPT, amassing downloads on the App Store.

As January concluded, the tech industry witnessed a frenzy—especially during the Chinese New Year. First, Alibaba revealed Quin 2.5, an AI capable of creating code animations intuitively.

Despite the impressive tech, keep in mind that DeepSeek collects sensitive data, including chat histories and any inputs submitted. While OpenAI executes similar data practices, users’ information with DeepSeek goes straight to servers in the People’s Republic of China.

In terms of privacy, one upside is that DeepSeek can run locally without internet access. For complete privacy, users can run it on their machines. As I demonstrate in this video, you can ask DeepSeek to write code and it will do so efficiently.

Throughout this period, DeepSeek had to temporarily limit user registrations due to large-scale malicious attacks. This serves as a caution that their program may not be as ready for mass adoption as it appears.

What does Sam Altman think of DeepSeek's impressive model? He acknowledges their capabilities while asserting that OpenAI will continue to deliver superior models. This rivalry signals an exhilarating shift in the landscape.

Essentially, we witness the technological encapsulation of Thucydides's trap—the idea that conflict arises when a rising power challenges an existing one. In an interview, Liang expressed that Chinese companies can no longer rely on innovations developed elsewhere; they aim to expand the technological frontier.

As advancements continue at a rapid pace, we might witness breakthroughs across a variety of fields: from medical to theoretical physics. On the flip side, however, concerns over potential abuse and the human transition during this shift loom large.

While the future remains uncertain, keeping an eye on developments in AI—and more broadly—is essential. That’s where we stand with DeepSeek1Real, a technology that works efficiently, reflecting the shockwaves it's sending across global marketplaces.

The rise of consumer AI is undeniable, and despite varying opinions, its integration into daily life is inevitable.

For those eager to learn more about AI and other STEM subjects, check out Brilliant's course on artificial neural networks, crafted by experts from renowned institutions. For a full month trial, visit brilliant.org/coldfusion for a 20% discount on an annual premium subscription.

Thanks for watching. My name is Degogo, and you've been tuned into Cold Fusion. See you soon for the next episode. Cheers, and take care!