DeepSeek (and before it became DeepSeek)

DeepSeek is the Chinese large language model (LLM) that stunned the AI world with its low training costs and open-weight approach. This episode dives into the extraordinary founder story of Liang Wenfeng, the secretive quant trader who pivoted his billion-dollar firm, High-Flyer Capital, into a top AI competitor.

How did a quantitative finance empire become the birthplace of DeepSeek R1 and V3? We unpack the innovations, the GPU hoarding strategy, the DeepSeek architecture, and the controversial pivot that positions them as a serious challenger to ChatGPT and Llama in the global AGI race.

👉 Subscribe & Follow: Ascent Podcast on Substack

In This Episode:

00:00 Intro

04:23 Liang Wenfeng's early life and education

08:16 Inception of the quant trading journey

18:47 Becoming quant king: building a billion-bollar empire

31:00 The hoarding (of GPUs) begins

39:53 Liang Wenfeng's vision for China's quant future

46:48 The 2021 challenge and fund drawdown

50:29 The pivot: from trading to AGI

57:17 Innovation under constraint

01:08:31 DeepSeek's unconventional hiring philosophy

01:12:35 Future uncertain: can DeepSeek outlast the giants?

01:20:50 Ascent with open source

Correction: at 59:24 - it should be 600 billion not 6 billion parameters

P.S. Yes, these show notes were also generated by DeepSeek!