What's striking about Liang isn't the technology, it's the posture. Every previous framing of Chinese tech strategy has been catch-up logic — close the gap, localize the supply chain, achieve parity. Liang's bet is the opposite: skip the commercialization phase entirely and go straight for the frontier, treating today's products as irrelevant compared to whoever reaches general intelligence first. That's not an incremental strategic adjustment. It's a company deciding the entire "catching up" framework no longer applies to it — and open-sourcing the models isn't generosity, it's a wager that speed matters more than moat.
Why Deepseek Free Code while West's ones not? My view is simple (even if wrong/limited): Millennial Confucian Ethics embedded in many (the majority?) of chinese individuals makes it far more easier for one of their genius (Liang) to share their knowledge than from the typical greedy, selfish westerner.
I am very impressed/surprised by the extent to which this model of learning instantiates Deutschean epistemics - specifically a Neo-Popperian model wherein current learning paradigms don’t represent continuous and generalizable explanation refinement.
There is no such thing as a 545% profit margin. Margin has to be between 0% and 100%. The chart shows a *theoretical* profit margin of more like 84%.
About the fact that the revenue is theoretical revenue: the authors even note (via Google Translate):
"Of course, we don't actually have that much revenue, because V3 is priced lower, paid services only account for a portion, and there are also discounts at night."
If you believe, as I do, that the cost of everything will trend toward zero within a couple of decades thanks to AI, everything in this piece makes sense.
As the author’s surname suggests, the article very much reflects the perspective of an ordinary Chinese person — someone with no hands-on experience running a hedge fund or building an LLM from scratch. In reality, no serious fund manager or LLM researcher would rate DeepSeek above Doubao, which is developed by Douyin (the Chinese parent company of TikTok) and is entirely home-grown, with no distillation involved.
What's striking about Liang isn't the technology, it's the posture. Every previous framing of Chinese tech strategy has been catch-up logic — close the gap, localize the supply chain, achieve parity. Liang's bet is the opposite: skip the commercialization phase entirely and go straight for the frontier, treating today's products as irrelevant compared to whoever reaches general intelligence first. That's not an incremental strategic adjustment. It's a company deciding the entire "catching up" framework no longer applies to it — and open-sourcing the models isn't generosity, it's a wager that speed matters more than moat.
Why Deepseek Free Code while West's ones not? My view is simple (even if wrong/limited): Millennial Confucian Ethics embedded in many (the majority?) of chinese individuals makes it far more easier for one of their genius (Liang) to share their knowledge than from the typical greedy, selfish westerner.
Yes, I hate all those westerners who have championed democracy and freedom. Terrible people.
He sure sounds pretty AGI pilled
I am very impressed/surprised by the extent to which this model of learning instantiates Deutschean epistemics - specifically a Neo-Popperian model wherein current learning paradigms don’t represent continuous and generalizable explanation refinement.
There is no such thing as a 545% profit margin. Margin has to be between 0% and 100%. The chart shows a *theoretical* profit margin of more like 84%.
About the fact that the revenue is theoretical revenue: the authors even note (via Google Translate):
"Of course, we don't actually have that much revenue, because V3 is priced lower, paid services only account for a portion, and there are also discounts at night."
If you believe, as I do, that the cost of everything will trend toward zero within a couple of decades thanks to AI, everything in this piece makes sense.
As the author’s surname suggests, the article very much reflects the perspective of an ordinary Chinese person — someone with no hands-on experience running a hedge fund or building an LLM from scratch. In reality, no serious fund manager or LLM researcher would rate DeepSeek above Doubao, which is developed by Douyin (the Chinese parent company of TikTok) and is entirely home-grown, with no distillation involved.