[EDRM Editor’s Note: EDRM is proud to publish Ralph Losey’s advocacy and analysis. The opinions and positions are Ralph Losey’s copyrighted work. All images in this article were created by Ralph Losey using his Visual Muse GPT.]
DeepSeek, a startup AI company owned by a Chinese hedge fund, which is in turn owned by a young AI whiz-kid, Liang Wenfeng, claims that its newly released V-3 software-R1 was trained inexpensively and without using NVIDIA’s high-end chips, the ones that cannot be exported to China. Wenfeng also claims his software is as good as the best-in-class AI software trained on the best NVIDIA chips. These claims were believed for reasons I will explain. This possibility that the claims might be true caused a run on Wall Street, the likes of which have never been seen before. The Trillion Dollar market crash included a loss in value of Nvidia of $593 billion, a new one-day record for any company, ever.
Introduction
One reason DeepSeek’s claims triggered a crash is that DeepSeek’s software is open-source and can be copied freely. Further, the training costs for the software are claimed to be 100 times less expensive and require less data and energy. The market assumed that if the claims were true, it would be an industry-disruptive breakthrough. It would also disrupt the political balance of world power. The day after the crash, former Google CEO Eric Schmidt said that the rise of DeepSeek marks a “turning point” for the global AI race and argued for more open-source products. That is one reason many are skeptical of the economic savings claims by an unknown Chinese startup. Still, the market panicked because many were quickly convinced of the overall quality of DeepSeek’s new R1 software itself. Seeing is believing.
There is no proof provided of the cost savings, but published test results and hands-on trials by tens of thousands of people worldwide trying the free smartphone software confirmed the quality claims. Here is the paper, in English, that DeepSeek also released explaining the programming innovations. DeepSeek-R1: Incentivizing Reasoning Capability in LLMs via Reinforcement Learning (Arxiv, 1/22/25). The top scientists of the major AI labs, all of whom immediately tested the software, also generally confirmed the quality claims. Rumors began flying that they were all in crisis mode, especially Meta, the only other company who had gone open source.
Sam Altman immediately responded to the release of R1 and market crash by tweeting on X:
Deepseek’s r1 is an impressive model, particularly around what they’re able to deliver for the price. We will obviously deliver much better models and also it’s legit invigorating to have a new competitor! we will pull up some releases.
Sam Altman, (X/@sama, 1/27/25).
My hands-on tests of DeepSeek show that it is at least “close” to the quality of ChatGPT. I will share the results in my next article, where I will closely study the one new software feature that so far only DeepSeek provides, called DeepThink. Many others are testing DeepSeek and reaching the same conclusion. DeepSeek is already the number one downloaded app on both Apple and Android phones.
Legal professionals, please note the software has no privacy protections. Your prompts will be used for training. DeepSeek should not be used for legal work that in any way involves confidential information. Also see, Wiz Research Uncovers Exposed DeepSeek Database Leaking Sensitive Information, Including Chat History (1/29/25).
Claims of Breakthroughs
This kind of market reaction is hard to believe when you consider DeepSeek’s owner, Liang Wenfeng, is a forty-year-old Chinese hedge-fund owner (High-Flyer) in Hangzhou, China. Wenfeng is a thin, bespectacled engineer, a billionaire with a genius for stock trading using AI. This allowed him to assemble and privately pay for a team of China’s top AI software developers.
His company, DeekSeek, is also virtually unknown, as its R1 software is its first consumer product. In spite of the fact that Wenfeng is basically an AI stock trader and his company is unknown, many stock traders and analysts in the U.S. were persuaded to believe Wenfeng’s claims to have achieved an AI development cost breakthrough by using:
- older, cheaper, non-trade-restricted models of NVIDIA chips; and,
- new innovative training methods.
Wenfeng also claims DeepSeek’s new software is better in at least one respect than existing models because it can describe the chain of thought and assumptions made by the AI to respond to each user request. They call this new feature “Deep-Think.” My next article will go into that next. Spoiler alert: it does appear to work as claimed, and I predict the U.S. companies will come out with their own versions of this Deep-Think feature soon.
Political Impact of DeepSeek
Are all the claims of DeepSeek real and to be believed as the market has done? Or are they just clever AI-driven propaganda? Especially the unsubstantiated claim that DeepSeek has invented a way to train cheaply on older chips? Could this be part of some kind of master strategy of the People’s Republic of China to counter the prior actions of President Biden, so far continued by President Trump, to restrict chip sales to China? Presidents Biden and now Trump have pledged to maintain U.S. superiority in AI. President Trump now appears ready to go further by imposing new trade tariffs and AI restrictions on China.
China’s DeepSeek claims, but has not proven, that many companies all over the world can now create an equal or better model at far less costs than ever before, that it can be done using older, non-trade-restricted computer chips and more advanced data training methods. If so, then U.S. dominance in AI may end. The fear of industry disruption by Chinese code innovation is the basic cause of the U.S. market panic.
Was the market panic propaganda manipulated by China? How will the Trump Administration react now? How will the leading established AI companies? The rumor mill claims the AI companies are all in a panic mode now and knee-deep in emergency meetings about DeepSeek.
There is one piece of evidence to support speculation of the involvement of the Chinese government in market manipulation. Liang Wenfeng was seen meeting with Chinese Premier Li Qiang on January 20, 2025. The market sell-off was just a week later and was obviously very good news for the Chinese government leaders. It was also a slap in the face to U.S. leaders. Only a week earlier, President Trump announced a $500 Billion Dollar initiative for AI development in the U.S. Then a Chinese upstart comes out of nowhere and punches away a Trillion Dollars out of our market by claims that American chips are no longer needed. The Trump Administration appears to be gearing up to fight back once it figures out what happened and what to do. See e.g., Trump Commerce pick slams China: ‘Stop using our tools to compete’ (The Hill, 1/29/25) (confirmation testimony of the nominated Commerce Secretary, Howard Lutnick, blames trade-secret theft for DeepSeek’s success).
The night after the stock market crash, President Trump appeared before reporters at his home in Del Largo and told them the release of DeepSeek AI from a Chinese company should be a “wake-up call for our industries that we need to be laser-focused on competing to win.” He stated the development could be positive for the United States: “If it comes in cheaper, that’s going to benefit us too.” He was expecting new AI systems by U.S. companies as soon as next week that “will top” DeepSeek’s model. Then, of course, he said in usual fashion: “We’re going to dominate! We’ll dominate everything!‘ China’s DeepSeek disrupts American plans for AI dominance (Washington Post, 1/28/25). The market did recover slightly the next day, but not much, and Nvidia briefly dropped below $120 per share again today, January 30, as I am writing this. Still, I do expect a bounce back.
Was the market panic propaganda manipulated? How will the Trump Administration react now? How will the new Trump SEC react? (Market manipulation is illegal. Is the SEC investigating?) How will the leading established AI companies now react? The rumor mill claims the AI companies are all in a panic mode now and knee-deep in emergency meetings about DeepSeek.
CEO of Anthropic, Dario Amodei, Argues for Continued Restrictions on China
The CEO of Anthropic, Dario Amodei, has already written an essay reacting to DeepSplash. On DeepSeek and Export Controls (January 29, 2025). Below is his image and the opening paragraphs of his blog.
Here, I won’t focus on whether DeepSeek is or isn’t a threat to US AI companies like Anthropic (although I do believe many of the claims about their threat to US AI leadership are greatly overstated)1. Instead, I’ll focus on whether DeepSeek’s releases undermine the case for those export control policies on chips. I don’t think they do. In fact, I think they make export control policies even more existentially important than they were a week ago2.
Export controls serve a vital purpose: keeping democratic nations at the forefront of AI development. To be clear, they’re not a way to duck the competition between the US and China. In the end, AI companies in the US and other democracies must have better models than those in China if we want to prevail. But we shouldn’t hand the Chinese Communist Party technological advantages when we don’t have to.
Dario Amodei, On DeepSeek and Export Controls (darioamodei.com, 1/29/25).
Amodei argues that the recent advancements of DeepSeek do not undermine U.S. export control policies on AI chips but instead reinforce their necessity to maintain a technological lead. He sets forth key AI development dynamics, including scaling laws, efficiency improvements, and paradigm shifts, to put DeepSeek’s recent progress into perspective. Interestingly, he does not contest their claims. He only says, I think correctly, that DeepSeek’s improvements are part of a predictable trend and are not breakthroughs. Dario Amodei contends that well-enforced export controls are critical in shaping a future where the U.S. retains dominance in AI. He argues that this is necessary to prevent China from amassing the millions of chips needed to create future AI systems that could shift global power balances.
DeepSeek’s New DeepThink Feature in R1
I have tested DeepSeek’s new Deep-Think feature of R1. Although it is not a big breakthrough, it is an excellent new feature that none of the other AI software has yet developed. I will report on this in detail soon. It is a true innovation, unlike the cost and training advances that appear, at first glance, to be the result of trade secrets and copyright violations.
I predict that new Deep-Think-type features will soon be included in U.S. AI companies. In this manner, competition from DeepSeek will have a good impact, even if the cost claims later prove to be more political smoke and mirrors and litigation triggers than legitimate tech breakthroughs that magically trivialize NVIDIA’s inventions.
Conclusion
As the old Chinese adage supposedly goes, may you be cursed to live in interesting times. The pace of change in AI improvement is incredible. I suppose this is what exponential change looks like. These are not only interesting times, they are very dangerous, especially in the relations between the U.S. and China. We certainly do not want control of superintelligent AI to fall into the hands of any dictator, anywhere. I will continue to follow this closely, not only from the point of view of law and technology, but also a political, economic, and military perspective.
Ralph Losey Copyright 2025 – All Rights Reserved
Assisted by GAI and LLM Technologies per EDRM GAI and LLM Policy.
