Deepseek R1 - TikTok and What All The Buzz Really Says About Us

We want what we want when we want it AND we are willing to sacrifice both privacy and security to get it!

(warning this is more of a paper than a blog post)

Digital 'Hungry Ghosts'

I received this text from a good friend and fellow pseudo-techy just over an hour before Wall Street opened yesterday. Admittedly, I consume AI and tech podcasts like most people binge-watch Netflix series - like 5 to 7 podcasts a day, so Deepseek R1's release wasn't breaking news to me.

DeepSeek-R1 was officially released on January 20, 2025. The company unveiled this new open-source AI model, which focuses on reasoning tasks and is designed to compete with OpenAI's o1 and other 'reasoning' models. In the podcast world, the event did gain significant attention, however, it wasn't until this morning when the stock market reacted that 'breaking news' hit X and every other platform.

This morning Nvidia (the real-world version of Skynet)'s stock opened 11 points down and lost another 5.8 by closing. What's all the 'buzz'? Let's break it down at face value:

While American tech giants like Google, Microsoft, Anthropic, and OpenAI have been spending exorbitant amounts on training(1) their AI models—with costs potentially reaching billions of dollars by 2027—Deepseek, a Chinese-owned company, claims to have developed R1 at just one-thirtieth the cost of o1, OpenAI's flagship reasoning model(2).

If the Chinese company is telling the truth that could be highly disruptive to the perceived American supremacy in AI and the sustainability of the current approach of using increasingly larger and more costly data centers for training. If you don't follow and understand the AI market this a pretty complex issue but let's look at the big picture, what it says about where we are and more importantly what it says about us and let's start with some definitions:

(1) AI Model Training- Training an AI model is like teaching a very complex, digital brain. It involves: Data Collection, Processing, Pattern Recognition, Trial and Error, Iteration Fine-tuning, and Computational Power.

This entire process requires enormous amounts of computing power, which is why it's so expensive.

The goal is to create an AI that can generalize from its training data to handle new, unseen situations accurately.

*The more data and computing power used, generally the more capable the resulting AI - which is why it's often so costly.

(2) Reasoning Model: A reasoning model is an advanced AI system that can think critically, solve complex problems, and make logical decisions. These models are designed to mimic human-like reasoning processes, making them more effective at handling complex tasks and providing more calculated and accurate responses.

(3) Open Source vs Closed Source: Open-source AI models are like public libraries of code where anyone can read, use, and modify the technology freely. Closed-source models are like private books locked in a company's vault - you can use the final product, but you can't see how it was created or change its inner workings. Open-source models promote collaboration and transparency, while closed-source models protect proprietary technology and intellectual property. Think of it as the difference between community-developed software versus a product created and controlled by a single company.

*Open Source: Meta, IBM, Hugging Face

*Closed Source: ChatGPT, Microsoft, Amazon, Google, Anthropic

The Meat and Potatoes

A GPU is a special computer chip that's really good at processing lots of information at once, especially for graphics and complex calculations. It's crucial for things like video games, but it's also become essential for AI.

Nvidia is the big dog in the GPU world. They make the best chips for AI, and they've got a stranglehold on the market - we're talking about 90% of the AI chip market. This means most companies doing serious AI work are using Nvidia's tech.

Now, China may not play nice with America but they have a very aggressive tech and ecommerce ecosystem. From electric cars to lights out warehouses, and robotic docks China is our only true competition in tech and AI. However, the U.S. government has imposed strict export controls on advanced AI chips to China, including Nvidia's most powerful GPUs like the H100 and upcoming Blackwell series.

This ban aims to limit China's access to cutting-edge AI technology for stated national security reasons. In my opinion, this is a bad move: necessity is the mother of invention. Based on the most recent data from 2020, China graduates significantly more engineers annually than the United States:

1. China awarded approximately 1.38 million engineering bachelor's degrees in 2020.

2. The United States awarded about 197,000 comparable degrees (144,000 in engineering and 54,000 in computer science) in the same year.

This means China graduates about 1.18 million more engineers annually than the United States or roughly 7 times as many

In response to the bans, China is investing heavily in developing its own GPUs and AI chips to reduce reliance on American technology and it's got the talent, leadership, and government support to do it. Chinese companies like Huawei are emerging as potential alternatives to Nvidia. Additionally, Chinese firms like Deepseek are exploring innovative approaches, such as developing more efficient code and smaller, specialized AI models to compensate for the lack of access to the most advanced chips.

Add China's historic tendency to 'copycat' American technology to the mix and you end up with 'distilled' open-source models like R1.

But, the R1 dilemma is more than a question of China vs the US, it's really about open source vs closed source, and here are my predictions to how all of this will play out:

American Businesses:

1. Enterprise-level companies (think Coca-Cola, GM, Nabisco) will opt for more expensive 'out of the box' foundation model products that provide safety, security, technical support, and the best user experience to ensure mass adoption. Think Microsoft, Google, and Apple where the technology will be 'backed into' their suite of goods. Those who want really leverage state of art with the same support and functionality will go with an OpenAI or Anthropic at a premium price.

   1. OpenAI CEO Sam Altman has already spoken of a $2,000-per-month subscription possibility. (that's per seat) Don't swallow too hard: McKinsey will pay it and then charge companies millions to teach them how to build their global strategies around them.

2. Everybody else will tap into big open-source foundation models like Meta's Llama series and then either:

   1. Use a 3rd party AIaaS (AI as a Service) API platform to store, clean, and operationalize their data with custom AI products and services or

   2. Hire AI software engineers to build smaller, on-premise, proprietary models customized to their data and needs.

      1. Note option (b) will require repeated fine-tuning as well as a sufficient data lake house so start cleaning your data now.

         

3. For the majority of American companies, all of this will be provided by AI and tech companies that are either US-based or housed in a nation we consider an ally.

   
   

Global Economy

1. China, unfortunately, may very well dominate the markets in 3rd world and less affluent countries like South American and African nations.

2. The U.S. and it's allies will prevail in the European and more wealth markets and that brings us to why:

   
   
   
   

   

   Safety and Security

   
   
   
   

What I believe the real takeaway from the past two weeks has been is this:

Americans want what they want when they want it and they are willing to sacrifice safety and security and privacy to get it - to a point.

TikTok - the US government declared TikTok unsafe for the American market based on suspected spyware, data collection, and mass propaganda. ByteDance (TikTok's parent Chinese company) tried to appeal to US Supreme Court on the basis of a 1st Amendment violation. The Supreme Court ruled that ByteDance had no standing because foreign agencies are not afforded protections under the US Constitution.

US consumers, led by Gen Z, flocked to Xiaohongshu or the 'Little Red Book' as 100% Chinese app. It wasn't all kids, many American adult TikTok fans jumped on a foreign app they little nothing about, almost as a middle finger to the US government.... soooo .... Brittney Griner's experience with a Communist regime taught us nothing about how the real world works?

Xiaohongshu or the 'Little Red Book'

Now China releases an AI model that has identified itself as both ChatGPT and Claude during testing and deployment and we aren't put off enough by the fact they have distilled a model using American technology we rush to download a foreign-adversary's top-of-the-line AI tech to our computers with no concerns over mal or spyware because its cheap and trending.

If America truly wants to come out on top in the AI race here's what needs to happen:

1. Self-regulation and common sense - don't play with dynamite just because the other kids are.

2. Yes, our best and brightest need to kick it into overdrive and innovate like tech-gladiators BUT those efforts will be for nothing if the American public doesn't aggressively adopt AI technology as well, not as consumers - as creators!

I'll leave you with this thought. On X (and in many of my posts) you'll find taglines like:

"AI won't take your job but someone who knows how to use AI may."

Here's the truth: If the American masses don't start spending more energy leveraging AI and technology to empower themselves to be more creative, productive, and effective; if we continue to look to these tools as a means of entertaining ourselves and consuming digital content like a bunch of 'hungry ghosts': we are in big trouble.

I'll leave you with my favorite Obama quote: "Change will not come if we wait for some other person or some other time. We are the ones we've been waiting for. We are the change that we seek."

Jason (human in the loop) Padgett