DeepSeek CEO Liang Wenfeng's recorded speech to investors has been circulating online for a whole day, In the afternoon, it was gradually taken down from domestic public platforms. I saw it was a group friend's document, but I still enjoyed it very much! In the 42 page manuscript, AGI, open source, commercialization, computing power, domestic chips are discussed, as well as organizational management and talent. There is a lot of information, but I think he repeatedly talked about only one thing: how to improve the probability of ultimately achieving AGI. For example, what left a deep impression on me was that DeepSeek did not choose to attack everywhere, but to constantly subtract. Don't rush to grab users, don't create the next super app; Not for the sake of income, eat everything from the C-end and B-end; Video generation, 3D, and world model reheating, also not following; The model can make more money with closed source, but still choose open source; The price can be set higher, but it only earns profits that one considers reasonable. It's not because he doesn't want to make money. On the contrary, it is because he knows which money can be earned and which money has been earned now, which will actually slow down himself. one ️⃣ Restraint is a strategy! There is a saying that left a deep impression on me: restraint is a strategy. The business competition that many people understand is to compete for users, entry points, and revenue. Whatever you can get, you get it first. But Liang Wenfeng's understanding is completely opposite. AI is big enough that no company can truly monopolize it. Since the watermelon behind is big enough, there is no need to stop for the sesame in front of us. C-end users can do it, and B-end revenue can also be earned, but these are just by-products on the road to AGI and should not be turned into the company's goals. This is actually the most philosophical aspect of the entire conversation: The more you want, the harder it may be to get in the end; You are willing to take the initiative to take less, which makes it easier to reach the end. Open source follows the same logic. Many people believe that open source is giving up the moat, but Liang Wenfeng believes that as long as DeepSeek always has advantages in cost and efficiency, open source will not harm business, but can attract talent, establish an ecosystem, and reduce enemies. He even judged that if a company wants to take away too much benefit from the AI era, it will ultimately be defeated by someone who is willing to take less. Taking too little, the company cannot survive; If you take too much, you will be challenged by new competitors. The only thing that can last for a long time is the 'reasonable profit'. So his restraint is not Buddhist, let alone without desires or pursuits. But rather a strong sense of purpose: Profit can be reduced, popular businesses can be avoided, and short-term users can be avoided, but there are only two things that cannot be lost: the AGI mainline and the stability of the core team. This is where restraint is truly sharp. He doesn't have no desires, but he puts all his desires on one thing. two ️⃣ Technical Roadmap Technically, the route provided by Liang Wenfeng is also very clear: After the language model is the thought chain, after the thought chain is the agent, and what really needs to be solved after the agent is "continuous learning". The current AI capabilities are already very strong, but it still requires humans to provide complete context. It cannot enter the company for two months like a new employee, understand the people, relationships, rules, and work habits here, and then continue to grow. Once AI has the ability to continuously learn, it can begin to assist humans in researching the next generation of AI, forming a cycle of "AI accelerating AI". Further on, it is self iteration and embodied intelligence. Of course, he also honestly admitted: Continuous learning has not yet been truly solved, and the whole world is still exploring. This kind of honesty is also rare. While believing that AGI will definitely happen, acknowledging that the specific path is still full of unknowns. It's not about drawing a definite timeline for investors just for financing. three ️⃣ The gap between China and the United States His judgment on the gap in AI between China and the United States is also very direct: The biggest gap is not in talent, but in computing power and resources. In the case where there is still an order of magnitude gap in computing power, China cannot surpass it comprehensively. It can only catch up or even lead in some directions with higher efficiency, lower costs, and clearer choices. He judged that the real gap in future big models would not be any mysterious technology that can never be replicated, but three things: Cost, time, and user experience. Whoever can produce it earlier, provide it at a lower cost, and make their product more comfortable to use, will be able to stay. four ️⃣ long-termism The biggest inspiration for me from this exchange is not about which AI company to buy, but about re understanding the four words' long termism '. True long termism is not just about saying that one can see far ahead. But when short-term benefits are really in front of you, do you have the ability to refuse. Investment is the same. Every hot topic wants to participate, every market trend wants to make a profit, every rise is afraid of being missed, and in the end, funds, attention, and judgment are all dispersed. A person's true ability circle may not only know what they are good at, but also know: What money doesn't belong to oneself in the first place. But Liang Wenfeng's model cannot be simply romanticized. DeepSeek dares to restrain itself because it has technical efficiency, a team, funding, and believes that the market in the future is large enough. Open source, no KPI, and allowing employees to explore freely, not every company can succeed by copying. Liang Wenfeng himself admits that as the company expands, some departments still need clearer organizational structures; DeepSeek must also rely on APIs and commercial revenue to survive. So what is truly worth learning is not the superficial "not making money", "not working overtime", and "not setting KPIs". But first think clearly: What is something that one absolutely cannot lose, and what can be voluntarily given up. For DeepSeek, what cannot be lost is AGI and the core team. For ordinary people, it may be health, family, cash flow, judgment, and the direction they are truly willing to invest in the long term. Today everyone was discussing how to have more, but Liang Wenfeng spent nearly four hours explaining why he couldn't have so much. I think this may be the most powerful part of the entire communication: It's not about seizing every opportunity to be powerful, but facing countless opportunities and still knowing where you really want to go. Many people believe that persistence is the most difficult aspect of long termism. Actually, the more difficult thing may always be giving up.