Low overhead, high competency, ownership, independence, and speed.

It makes me wonder if products can be built this way too…

What is impressive about Deepseek though is the flat hierarchy. The company is structured as an open lab where all researchers can collaborate. All resources (GPUs) are shared and if someone hits gold, the team supports them.

This is such a refreshing departure from the typical eng structures.

Instead of commercialization, Deepseek focuses on research to leapfrog the current frontier models and plant a Chinese flag in the AI field.

In the interview, Liang (Deepseek CEO) stresses that this strategy is the only way China can get ahead of the competition and make something nascent.

3. This was a big surprise to the world because Chinese tech companies were considered unimaginative, and relies on the US for breakthroughs.

This is where the leader of Deepseek stepped up and said “no”.

He lead a carelessly brilliant team to break stigmas and focus on innovation. Massive chad

2. Their innovation comes from China’s constrained GPU market. Because Nvidia can’t sell H100 (top line) GPUs to Chinese companies, Deepseek had to experiment imaginatively with their training algorithms.

By working smarter and not harder, they unlocked a training method that mocks the Americans.

1. Deepseek caught up with Open AI and other frontier model builders on a TENTH of the cost and in a ridiculously short time.

According to leading researcher Andrej Karpathy, their 6M dollar budget was a “joke” compared to the billions OpenAI is spending on training costs.

🎩

Deepseek: The Quiet Giant Leading China’s AI Race Annotated translation of its CEO's deepest interview

This is a great article about one of the most exciting tech ventures in recent years.

A long admiration thread:

www.chinatalk.media/p/deepseek-c...