This morning you woke up, checked the stock market, and saw it was down quite a bit. Oh look Bitcoin bros, Bitcoin is down too. Hint: it’s not a hedge.
The Never Trumpers will blame Trump. They have no meaning in life and every single sentence uttered by someone is a confirmation bias chance to compare it to Hitler. Here is a beautiful satire by my friend Dan Proft. Racheal Bitecofer has over 200k followers on X, and is a “political analyst”. Instead of “followers” it should say “sheep”.
Future offerings from Bitecofer:
JD Vance: I like dogs.
Bitecofer: Hitler had a German Shepherd named, "Blondi."
JD Vance: I want more highly-trained combat pilots in our Air Force.
Bitecofer: You know who was a highly-trained combat pilot? Hermann Göring!
JD Vance: We need to enforce the law.
Bitecofer: Heinrich Himmler wanted to enforce the law too! One must obtain a PhD in political science to become a person as silly as Bitecofer.
Some people will think it is the American government yet again running into the debt ceiling. It’s not. Hakeem Jeffries will be front and center today with some grade school stock market analysis. Some will think that it’s just a normal cycle. The Technical analysis chart people will tell you that “the stock market hit some resistance it has been trying to break and failed. It will retrace to here.”
They are all wrong.
Yesterday, the Chinese released a revolutionary new AI engine that blows the doors off everything everyone has been working on in America. Remember Sam Altman and the guys who announced Stargate to great fanfare last week? A lot of that is now ancient history.
You are getting the chance to see creative destruction in real-time. In America, we don’t allow that to happen enough. We protect industry by stopping mergers and acquisitions, subsidies, tax breaks, taxes, and tariffs. We give some favored businesses under market loans and change labor laws. It slows creative destruction but doesn’t stop it.
Marc Andreessen called yesterday a “Sputnik moment” in AI. It truly was. Chamath Palihapitia was amazed by the Chinese release.
It is concerning because the next war isn’t going to start with gunboats hammering your shoreline with ordinance. It’s going to start with satellites taking out all electric communication and the electrical grid. Then millions of drones are going to swoop in and destroy targets. Artificial intelligence will power the battle.
It’s also concerning because American businesses are going to harness AI for their own devices. American businesses need access to the best most efficient AI to create opportunities for people and solve their problems.
The AI race in Silicon Valley had two certainties.
The first was that they were using $NVIDIA chips to build on, which sent NVIDIA stock soaring. All the big boys came to play: Google, Microsoft, Meta. But many startups came to play, too. The most famous is Sam Altman’s ChatGPT, which had a valuation of $157 billion until yesterday. It’s not worthless, but it is worth billions and billions less. Elon Musk’s X platform had Grok. Musk and Altman are suing each other over the corporate structure of ChatGPT. That suit might be a moot point.
The second certainty was that billions of dollars were invested by startup investors. Venture capital, much of it undisciplined, threw billions of dollars into startups. Many startups that weren’t actually using AI just put an AI wrapper on their business so they could raise at a higher valuation. Some VCs who didn’t know anything about AI went on panels using AI buzzwords and put money into companies. We have not seen the result of that yet, but I can guarantee you this vintage of specialized AI funds won’t have spectacular returns.
It was sort of like the crypto craze between 2016-2021. But different.
The other lesson in all this is for startup founders. Be efficient with your capital stack. Raise just enough capital to get to the next milestone. Don’t let your ego get involved in your raise. I just had a wonderful exit and that firm only raised capital from investors three times. It can be done, but you need discipline.
The other concern I have is that when you use their LLM, you are giving data to the Chinese. I realize that all kinds of platforms mine all kinds of data from me every minute of every day. But, while I hate that my privacy is constantly being invaded I am more comfortable with Western corporations doing it than Communist ones.
It also brings the broader point that where TikTok was owned by the CCP and gathered data, along with influencing the hearts and minds of the West, DeepSeek is at a whole new level.
The answer is not to ban it. The answer is to better it. Compete, compete, compete.
I read a very long analysis of the new Chinese DeepSeek R1 LLM yesterday. Jeffrey Emanuel wrote a rigorous article that will help you understand what all the fuss is about. Here is one reaction to the Chinese DeepSeek R1.
the r1 drop should be setting off alarm bells in the entire western capitalist apparatus. for the first time in a while, i’m genuinely afraid the u.s. is losing its dominance—its influence, its capacity to innovate, & its ability to outcompete on a global scale. the core issue isn’t just regulatory capture or bureaucratic inertia. the real problem is also the managerial class: a bloated, risk-averse, overpaid, over glorified layer that’s systematically dismantled the mechanisms that once made the u.s. competitive. meta is a great example of this (look at what happened with llama)—that’s with a founder running it, now imagine google where there is a professional manager running it & 5k vp’s underneath. meanwhile, china is surging ahead. it’s not just drones or now a world-class ai model; it’s the entire ecosystem. robots are building other robots at scale. their tech is cheap, effective, & ubiquitous. the world’s supply chain—its manufacturing core—is de facto chinese. china isn’t just catching up; it’s scaling the future. if the west continues down this path—short-term thinking, managerial inefficiency, & a refusal to prioritize strategic innovation—it risks becoming irrelevant in the face of a system designed for precision & long-term dominance. this isn’t just a challenge; it’s an existential threat to western economic & technological leadership.
DeepSeek uses high-level probability to predict what they will need to use to figure out a problem. In order to truly understand how their LLM works, you have to know a bit about how others work. The end result is the Chinese model uses 5% of the GPUs compared to the model Meta built.
The other thing that is mildly fascinating is the decentralized nature of not only the code but the team that built the code. China is communist. Centralization is paramount to everything they do. It’s hard-wired into how they think about solving societal problems. To see them do this in a decentralized manner is amazing.
We don’t know if they are telling the truth about how much involvement the CCP had, or how much money was used to build the engine. But, anyone who is in AI can see that it is novel, a great leap forward, and was built a helluva lot cheaper than anything to date.
One other thing to note, the sanctions on all sorts of technology the US placed on China didn’t stop them one bit.
Don’t be scared. Welcome the competition!
Is the US losing the war in AI??” I don’t think so. DeepSeek had a few big breakthroughs, we have had hundreds of small breakthroughs. If we adopt DeepSeek’s architecture, our models will be better. Because we have more compute and more data.
Many techies on X yesterday said, “It’s on.” Remember, Joe Biden and the Democrats wanted to regulate AI. That was the straw that broke many in Silicon Valley’s back and caused them to vote for Trump.
We should not regulate AI. Let the competition bloom and boom.
https://x.com/morganb/status/1883686162709295541 good thread on this.....
This is a gigantic overreaction and the sky is not falling. Calm down, all of us.
The big takeaway is that DeepSeek did their work without the same number of Nvidia chips as the Americans, but we only have their word on that and I wouldn't believe China as it relates to whether water is wet. Color me wildly skeptical.
This speaks to efficiency, if correct and a different software thrust.
The Americans have massively superior compute, data, and scale whilst they now also have an inkling as to how the Chinese did their work. Much of the DeepSeek work is open source, but not all of it.
This gap will be quickly closed.
If the Americans adopt some or all of the Chinese approach and couple that with access to chips and compute, data, and scale then the outcome is predictable.
This may indeed be a Sputnik moment, but let's recall how that ended with America spurred to compete and untimately overcome any shortlived Russian advantage.
In many ways there is a different scale issue at work that is similar to nuclear weapons. Is there any real advantage of one adversary having a 250kt weapon when its enemy has a 150kt weapon?
No, it then becomes weapon reliability, the delivery system and market penetration. The US needs to have a coherent and war time footing policy on its chip manufacturing because this -- in the implementation phase -- becomes a war on chips.
Chips -- designing them, making them, obtaining them, using them -- is splitting the atom as it relates to nuclear weapons. The Chinese are wildly vulnerable as to interdicting the flow of chips in the world.
One of the reasons the Chinese have done this with a smaller base of chips is because they are not supposed to be able to access Nvidia chips. [The Chinese have another advantage -- total focus via a single company.]
The Holy Grail is AGI. Keep the main thing the main thing. The Americans should get to AGI before the rest of the world. Everyone will eventually get there.
Instead of regulating American AI, we should be spurring them on. Turn them loose.
This kick in the balls will be very good for American tech.
JLM
www.themusingsofthebigredcar.com