Efficiency as a Belief: Organizational Culture and the Secret of "No Overtime"

I think low cost is first of all a result. Our models have indeed been moving toward lower cost in terms of model architecture, which relates to our vision. We still have many algorithmic methods that can further reduce costs.

Another reason to lower costs is that the lower the cost, the larger the model I can train, the larger the model I can afford. With the same compute, in a limited compute scenario, if my computational efficiency is higher, I can afford a larger model. For large companies, they might not think this way. For them, resources are expandable, they can add resources to solve it. But we prioritize cost efficiency.

The value of data-centric models-data is a broad term, but data should almost equal half of the model. Why do I think that if I want that, or suppose AI could account for 20% of GDP, and if I want to take 5% of that, it's absolutely impossible? Because I would definitely be beaten by someone else who says they only want 1%.

If my goal is to take 5% of AI, or the entire human GDP, theoretically the math works. Look at OpenAI, they seem to make the math work, theoretically it's fine. But they have a problem: they will be beaten by someone willing to take only 1%. Because that other person says, 'I do this well, but I only ask for 1% of global GDP,' and that will beat them. Then if another person comes along saying they only need 0.1%, they will beat the previous one.

Macroscopically, it doesn't matter where the percentage is taken from; there's no difference, it's all the same. Those who take more are beaten by those who take less. You don't even have to actually take more; if your vision is to take more, you will be beaten by someone whose vision is to take less. In reality, no one has taken the money yet; it's just a vision. If your vision is to take more, you lose first, and you face greater difficulties. That's how the world works.

OpenAI initially thought it could truly monopolize the world, but in reality it will face many challengers. It will encounter challenges and won't have it so easy. The US has faced challenges, and in the future it may also face challenges from China, because the Chinese are willing to take less and still provide the service.

In China, there will also be people willing to take even less. But eventually there will be a balance, because if you take too little, the company's business model breaks and it can't survive. So if you take too little, you can't survive; if you take too much, you are beaten by those who take less. So for us, we don't aim to maximize profits or set revenue-maximizing prices, but only earn a reasonable return. That's one explanation.

I believe in this. I'm not trying to find reasons for it, because there's no need. I already do it this way, and there must be a reason. The reason may not be conventional, but I think the company itself is unconventional.

Our company's management is actually two lines: one is top-down, and the other is bottom-up. Bottom-up means everyone does what they want to do on their own, no one manages them, no KPIs.

Top-down is formal: we collectively decide to do something and need the whole company to cooperate. For example, if we are going to release V4, then we need to divide the work, each person does their part. That's top-down, and we call that formal.

Generally, we hope that this formal work does not take up more than half of an employee's time. They have the other half of the time that is unscheduled; they can do whatever they want. This is a research scope that allows them to explore on their own, based on what they think is important, with no prior requirements.

As long as the company can support it-the company's compute can support what they want to do, or if they don't need compute and need very little, they don't need to coordinate at all. So this is how we organize. Some people think we are top-down, some think we are bottom-up, I think both are correct. My standard is that formal work should preferably not exceed half.

We generally don't work overtime much. There are two reasons for not working overtime. First, doing research requires a relaxed environment. If you push too hard, you can't do research. Since it requires you to have interest and think about these problems on your own, you need a relaxed environment to be able to explore. This is a need of the research culture.

Second, we are very focused. Being focused means we have very few things to do. So I don't have that many things to do, and I don't need to work overtime. This is consistent with the earlier point about restraint. Because I am restrained, many times I simply don't do things. So the things I need to do are few, and each person's workload is small. You can see that many of our products are not perfect, and we don't patch them up. That's also part of our culture.

OK, since there are many questions, I've glanced through most of them. If you have any other questions, feel free to ask. Investors, please open your mics and communicate freely. Just a reminder: Liang Wenfeng has shared some sensitive information. Please do not disclose any numbers or situations, including GPU counts, etc. Also, do not record the screen to share externally. Thank you all. If you have questions, please feel free to speak.