Last month, "elsewhere" reported on DeepSeek's financing story. The most discussed topic was undoubtedly the legendary 4-hour investor meeting.

In the past month, Liang Wenfeng's various quotes have circulated widely, and we have collected some of them from various sources.

During this time, Liang Wenfeng repeatedly said "no": He is not a genius, he will not earn unreasonable profits, he will not pursue user numbers, he will not close the source code, he will not do 3D/video generation/world modeling, and he will not create the next super app.

He said that restraint is a strategy used to increase the probability of successfully implementing AGI. In the limited materials we have gathered, the following keywords frequently appear: model, cost, AGI, time, open source, etc. Most of the time, Liang Wenfeng's tone is reserved, and his language is simple and straightforward. Only when discussing a few issues he cares about does he reveal a hint of sharpness: "As long as we can maintain the stability of the team, I will definitely be able to succeed in AGI. It's that simple." Below are 52 quotes we collected. Some phrases may differ slightly from the original text while retaining their original meaning. DeepSeek has only one main theme: 1. Now is not the time to maximize profits by developing a product. On the road to AGI, we need to pass through the product stage, but we don't need to spend too much time and energy developing C-end or B-end products. Working on a relatively lower-level technology from a high technical standpoint is like a dimensional reduction attack. The product is a byproduct of the AGI journey. 2. Many things are not part of our main focus, such as 3D and video generation; and world models, which are not directly related to the upper limit of intelligence. 3. Multimodal is important for products and for C-end users. But it's just a component, not the main focus or intelligence itself. 4. The illusion of a large model can certainly be solved, but it's a long-term problem. Internally, we classify the illusion as a product issue, which we will address, but it's not our priority. 5. At this stage, the most important thing is still the Coding Agent. From a domestic perspective, the most reasonable approach should be to focus entirely on developing general-purpose agents, with lower priority given to agents in finance, healthcare, and other sectors. 6. If the AI ​​era produces many trillion-dollar companies, it would be great if DeepSeek were one of them. First achieve continuous learning, then AI self-iteration, with the ultimate goal of embodied intelligence. 7. AI doesn't lack taste and intuition; what it lacks is the ability to learn continuously. 8. Humans can learn continuously, but for the same task, AI needs all the context. This is almost impossible, so AI cannot replace employees. Therefore, the next generation of models must have the ability to learn continuously to be considered a next-generation model. 9. We hope the next generation of models can help our own development. To put it simply, the primary goal of our models is not for everyone to use them well, but for us to use them well. This is the fastest way to achieve AGI. 10. Currently, no good method has been found worldwide because "learning" is composed of many things. 11. DeepSeek's long-term vision is AGI. If we compare the implementation path to climbing stairs, last year's stairs were CoT (Chain-of-Thought), and this year's stairs are Agents. The problem to solve after Agents is continuous learning. 12. After achieving continuous learning, we may reach an asymptotic singularity: the model can do everything humans can do, including developing more cutting-edge AI models. That is, AI can accelerate AI research. Only after this step is completed can we achieve embodied intelligence. 13. The ultimate goal of intelligence may always be embodied. Because for a normal person, their need is not for computers, but for human power.

We are still a long way from fully commercializing.

14. We only aim for a reasonable profit, not a profit-maximizing price.

15. Initially, we were worried about high demand for one of our models, so we set the price relatively high. Later, we lowered it to a quarter of the original price, which caused a lot of cheers in the company group. This is because this is the purpose of our hard work in making this model so that everyone can use it fully.

16. Low cost is a result; our models are constantly being designed to be more cost-effective. We also hope that the cost is affordable, especially in the context of scarce computing power.

Another reason: the lower the cost, the larger the model we can train. When computing power is limited, high computational efficiency allows us to train larger models. Large companies can solve this by adding resources; we prioritize cost efficiency.

17. From the outside, it might seem like we chose a difficult model. But actually, we did it very easily. Lowering prices is definitely not good news for our competitors; they certainly won't be cheering. I don't think selling APIs is that attractive. I only need a few people to maintain the API; there's no customer service or sales needed, users will come on their own. 18. We've always been commercializing, just not commercializing as our goal. DeepSeek is still a long way from completely transitioning to commercialization. 19. I don't even need to worry about securing a position within it. If there's such a large business opportunity, there will always be a way. DeepSeek is a product of its time, a reflection of reality, not a result of imitation. Open source is a sweet point for a company of our size. 20. Restraint is a strategy: give up some things to gain more others. Open source is about making concessions: internally, employees feel a sense of accomplishment, and the company has cohesion; it also benefits society, making peers and ordinary people happy. I have no doubt that AGI will have enormous commercial value. Based on this, my priority is not to gain more market share, but to increase the probability of success. 21. Open source is beneficial for making AI commercially viable. This may sound counterintuitive. Historically, a software company might have a market worth billions of dollars a year, but after open source, that disappears. However, AI is a huge undertaking; ultimately, it may account for 10% of human society's GDP. If we try to monopolize this benefit, we will inevitably be abandoned by history. This is an objective law, a historical perspective. 22. The open-source model we provide is the same as the model we deploy ourselves. We won't open-source a inferior model and then use a better one for our own deployment. 23. I'm not worried about others deploying our model to compete with us. Not every company has the willingness and ability to achieve this goal. Startups are too small to do this; large companies find it difficult to organize. This is a sweet point for companies our size. 24. Open source has no impact on our business model, provided we "only earn a certain profit." If you want to earn a hundred times the profit, open source will indeed affect you. 25. We don't want to become competitors with any large or small internet company. Under this premise, we are very willing to assist and help anyone, even Alibaba, Zhipu, and Lunar Dark Side, to do better. 26. In the future, we need to rewrite the narrative of AI between China and the US: using a fraction of the computing power, we can narrow the gap even further, to 6 months or 3 months. 27. The gap between China and the US in AI is mainly in resources. We believe that scaling will definitely yield better results with larger scales. Training such a large model is not because we think such a large model is enough, but because we only have so many resources. 28. There is almost no gap in talent—it's the same group of people. There is no shortage of talent in China. Talent shortages are temporary; historically, there has never been a long-term shortage of a certain type of person. 29. In the competition among model manufacturers, cost is the top priority. Anthropic's current surpassing of OpenAI is not a long-term phenomenon, but a temporary one. OpenAI and Google will likely take turns rising in the future. 30. There are too many companies in China making models, each doing the same thing, resulting in fragmented resources. Eventually, they will converge, but it will take time. If each company only takes a reasonable profit, there won't be so many people making large models: perhaps two large companies and two small companies would be enough. 31. I absolutely do not believe that large model companies can take away most of the profits (in the AI ​​industry). 32. The final difference in the competition for large models will be reflected in three aspects: cost, time, and user experience. Cost is the first priority—at what cost can you provide the same quality service? Second is time; a few months earlier or later makes a difference. User experience will create some stickiness and barriers, but it's not fundamental.

We have no intention of becoming the next super app

33. We don't want to become the next super app. Become the next ByteDance? The next Tencent? We have absolutely no such thoughts.

34. We don't compete for this because there are still big watermelons behind, and what's in front is probably just a sesame seed. Of course, this sesame seed might be bigger, but I don't think it's that big.

35. Last year everyone was scrambling for chatbots and C-end traffic; this year it's about scrambling for B-end revenue. But we don't think it's important. What the company is really concerned about is the AGI roadmap and how to achieve the next technological breakthrough. It's strange; what we want most is hard to get, while what we don't care about as much is easier to obtain.

36. Last year's Spring Festival hit wasn't part of our plan.

Maintaining team stability is key. 37. There's only one thing we can't compromise on: maintaining team stability. This is also a very significant risk we face. Of course, this risk has been largely mitigated with this round of financing. 38. Many things we do are aimed at maintaining team stability. We don't want to become competitors with any large or small internet companies; we hope to empower and help them. We don't want to make enemies. This will also improve our own environment. 39. Some people think our organization is top-down, some think it's bottom-up; I think both are correct. Top-down is called "doing the real work," and generally, we hope that "real work" doesn't take up half of an employee's time. The other half of their time is bottom-up, unstructured; they can research whatever they want, explore independently, and explore whatever they deem important, without any preconditions. 40. We generally don't work overtime much. The first reason is that research requires a relatively relaxed environment. The second reason is that we are very focused. Many products are not perfect, but we haven't fixed them; it's a kind of restrained culture. 41. Organizations are dynamic, not static. As the company grows, there may be some adjustments. It won't completely become a traditional hierarchical system, but there may be some necessary structures. However, the vision-driven approach will not change. 42. When we started this company, our initial intention wasn't to make a lot of money or go public. The first few dozen people never thought that way. If they had, they wouldn't have come. We are doing this with a great deal of goodwill towards the world, believing it will be useful to humanity. 43. "Achieving XX KPIs" is not our approach; we are a vision-driven organization. This has its advantages and disadvantages. We will find ways to leverage our strengths and mitigate our weaknesses in the future, but this is our defining characteristic. 44. Our vision isn't even written down; it's woven into our methods and our attitude towards the world. Everyone in our company may have a different understanding of this vision, but we are aligned on the overall direction. 45. About twenty years ago, my biggest management idol was Jack Welch (former CEO of GE). Looking back now, much of what he said may be incorrect, but he was right about one thing: the most important thing for a company is its vision. Vision isn't a slogan on the wall; it's not about what you say, but what you do.

Restraint Makes Us More Capable of AGI

46. AGI is the most profitable. As for other things, if we have the energy, we will do them; if we don't, we won't. Restraint is part of our vision.

47. AI is a huge project, and the potential profits are enormous. Even a small share of the profits would be substantial if we could succeed. The more restrained you are, the more likely you are to succeed.

48. I think this aligns with intuition—at least with my intuition. Besides our vision, we don't have many other advantages.

49. When we founded this company two years ago, we didn't have much money, many credit cards, little brand recognition, or much influence. We were just a group of very ordinary people. I prefer the narrative of "a group of ordinary people doing extraordinary things," rather than "a group of geniuses doing extraordinary things."

50. Open source is also part of restraint. In pricing, we certainly don't start with maximizing company revenue or profit. In the short term, higher prices mean more revenue; but in the long term, it's hard to say. For me, restraint is a strategy. 51. Open source and low prices make employees feel a great sense of accomplishment, fostering organizational cohesion, benefiting society, and making other peers and ordinary people happy. Therefore, this restraint, in the long run, increases our probability of succeeding with AGI. 52. If your vision is to take more, you've already lost; you may face even greater difficulties. That's just how the world is.