The Open-Source Revolution How DeepSeek Changed Everything

The Open-Source Revolution

How DeepSeek Changed Everything

 

 

 

 In my Previous Two Blogs, “The Dragon is Out” I discussed the cost of developing smaller AI models by using data from DeepSeek which can be Locally downloaded on a normal Laptop. Then in my blog” DeepSeek Vs ChatGpt” I discussed how Deepseek beats ChatGpt in writing and thinking skills. And how it is beneficial for students who use these models to help them in their studies.

 

Now in this blog let's find out how Liang Wenfeng and his team managed to pull out this model and that too at such a low price

Brief Background.

39-year-old Liang Wenfeng is a Chinese entrepreneur and businessman. He was born in the small village of Mililing, Tanba Located in South China.

Liang holds a master's Degree in information and communication engineering.

 During his Studies in 2008, he formed a team with his classmates to accumulate data related to financial markets. His team worked in Quantitative trading (known as Mathematical finance or Quantitative finance) using Machine Learning ( A field of Study using artificial intelligence).

After Completing his Studies Liang started experimenting with ways to apply AI to various fields. Most of his ventures failed but he was successful when he applied AI to financial models.

By Integrating artificial intelligence with Quantitvae Trading in 2013 He founded “Hangzhou Yakebi Investment Management Co Ltd” In 2015 He co-founded “Hangzhou Huanfang Technology Co. Ltd.”  and in 2016 he co-founded Ningbo High-Flyer Quantitative Investment Management Partnership (Limited Partnership).

USA Bans and Restricts China to Have Advanced Chips.

In October 2022 the Biden Administration introduced the export control stopping China from getting U.S semiconductors and technology. U.S. further tightened the restrictions in 2023 baning advanced chips and supercomputer components.

This was the Same Period when Liang Company High-Flyer announced a new research body. This Research team's role was to explore the essence of artificial general intelligence. However, like other Liang Companies, this one would not be used to perform stock trading. It was named as DeepSeek.

The Dragon is Out.

During 2022 before the US imposed a ban on China Liang started buying thousands of Nvidia GPUs. This was the time when he bought 10,000 Nvidia A100 GPUs (A100 was announced and released on May 14, 2020, by Nvidia).

By having A100 DeepSeek researchers were able to develop it as an LLM (Large Language Model like ChatGpt) model. DeepSeek didn't hire high-profile names from the tech industry instead they picked up locally educated people and trained them.

on 20 January 2025, DeepSeek released DeepSeek-R1. What disrupted the US and Tech Giants Nvidia was that R1 is open source which means you can study how it was made and tweak its code to create smaller models to perform specific tasks, Unlike Open AI which is rather a Close Source we can't see it codes, DeepSeek also published a detailed technical paper explaining its architecture and training methodology which anyone can study.

 The model was built using just 2,048 Nvidia H800 GPUs at a cost of $5.6 million Compare this to the OpenAI billion-dollar budget.

You can download the DeepSeek model entirely on a say simple Macbook Mini use it offline and play and tweak with its code it's entirely free. Whereas the OpenAI model charges millions of dollars for their API services.

By 27 January, DeepSeek surpassed ChatGPT to become the #1 free app on the U.S. iOS App Store. U.S. stocks plummeted, as more than $1 trillion was erased in market capitalization amid panic over DeepSeek.

U.S president who was teaming up with Sam Altman to show the world their superiority is now in shock. The emergence of DeepSeek sends shivers to President Trump's ambition to control the world economy with the stick of AI.

What makes DeepSeek researchers different from OpenAI.

AI researchers would never have produced something like DeepSeek. The DeepSeek team members are not traditional AI researchers; they come from a quantitative trading background, where they used to optimize down to nanoseconds. Due to GPU restrictions, in some cases, they bypassed CUDA (Compute Unified Device Architecture platform to accelerate computing). They used something closer to assembly language, which dramatically increased performance. Their original intention might be to publish a paper to enhance their resumes. Little did they know, they ended up creating a 'Ford Model T moment.' The Model T is generally regarded as the first mass-affordable automobile, making car travel accessible to middle-class Americans.

DeepSeek may not be seen as so much a revolution in AI, but it is a revolution in the way that it has put AI into everyone's hands.

Contribution by: Omar Hayat.