Li Yanhong’s latest speech in Dubai: The essence of innovation is lower costs (attached record)

Photo source: Company official

Blue Whale News, February 11 (Reporter Zhu Junxi)In the past half of a month, DeepSeek has become the most eye-catching topic in the AI industry at home and abroad. On February 11, at the World Governments Summit 2025 Summit held in Dubai, United Arab Emirates, Baidu founder Robin Li was asked if DeepSeek’s appearance was unexpected. He said that innovation cannot be planned. What can be done is to create an environment conducive to innovation.

After DeepSeek caught up with OpenAI at low costs, it triggered widespread doubts on Wall Street about AI infrastructure investment. In response, Li Yanhong’s statement is quite similar to the responses of giants such as Google, Microsoft, and Meta. He said that when technology develops so fast, you can’t stop investing. You must invest to ensure you are at the forefront of this technological innovation or revolution. rdquo; While it is possible to find a low-cost shortcut, billions of dollars could be spent exploring different paths before doing so.

In a conversation with United Arab Emirates AI Minister Omar Sultan Orama, Li also revealed the implementation application of Baidu’s autonomous driving radish fast. He said that technology is advancing very rapidly and has proved that autonomous driving is currently 10 times safer than human driving. Under complex road conditions in China, the actual accident rate of radish fast is only 1/14 of that of human drivers.

The following is a partial transcript of the conversation, the content has been adjusted by Blue Whale News to ensure the original intention.

Large model costs are reduced by more than 90% annually

Orama: Baidu is one of the pillar companies of artificial intelligence. I am very happy to invite Robin Li to discuss his views. Let’s talk about DeepSeek, the elephant in the room. In your opinion, is its appearance expected?

Li Yanhong:I believe that innovation cannot be planned. You don’t know when and where innovation will come. All you can do is create an environment conducive to innovation.

We live in a very exciting era. In the past, when we talked about Moore’s Law, performance doubled and costs halved every 18 months; but today, when we talk about large-language models, we can say that reasoning costs can be reduced by more than 90% every 12 months. This is much faster than the computer revolution we have experienced in the past few decades.

Big language models are a very huge field. In China, we must innovate in reasoning and training to reduce costs. Fortunately, over the past year, we have seen significant progress.

Baidu’s technical background is a search engine, which is naturally close to a big language model, so we launched Wenxinyiyan in March 2023. We are also the first listed company to launch a ChatGPT-like application. Google later launched Bard and renamed it Gemini. As we know today. This is a very exciting time, we see innovation everywhere, and we have to adapt to this rapidly changing innovation.

Train better next-generation models

Orama: A few weeks ago, when DeepSeek became a topic of conversation, the share prices of many large chip makers and many large exchanges around the world fell sharply. Because there have been jaw-dropping billions of dollars invested in reasoning data centers and training these artificial intelligence systems and models, how do you view the future of data centers and AI infrastructure?

Li Yanhong:I have been thinking about this issue for the past month or so. I think from a fundamental perspective, the most important theme is still that technology is advancing very rapidly, costs are reduced by about 90% every year, and performance is getting better and better. When technology develops so fast, you can’t stop investing. You must invest to ensure you are at the forefront of this technological innovation or revolution. We still need to continue to invest in chips, data centers and cloud infrastructure to create better and smarter next-generation models.

To do this, you need to use more computing power to try different paths. Maybe at some point you will find a shortcut, say, to train a model for just $6 million, but before then, you may have spent billions of dollars exploring which path is the right way to spend that $6 million.

Olama: Isn’t this a gambler’s dilemma? I’ve already spent $100 million, and now I need to recoup some of the costs, so keep investing. So, who will win? So, will we one day get enough rewards to prove it’s worth it?

Li Yanhong:I am optimistic about the future of artificial intelligence. I think even at the current level, the big language model has created a lot of value in a variety of scenarios. We have hundreds of thousands of customers using large models to improve efficiency in areas such as recruitment, e-commerce, healthcare and even energy and electricity.

We have seen many such use cases. In the past, they might have spent less than $10,000 to achieve certain goals, compared with about $1000 after using a large language model. So, this has created value for them.

But you’re right, I think we do need to focus on value creation at the application level. If you invest hundreds of billions of dollars as an infrastructure layer and can’t develop applications that deliver more than ten times the return, then this is unsustainable. Although we have seen a variety of use cases in a variety of different scenarios, these may be more concentrated in the ToB (enterprise-oriented) domain.

In the ToC (consumer-facing) space, we haven’t seen so-called super apps, like in the mobile Internet era or social media, with hundreds of millions of daily active users and people spending two hours a day on an app. We have not seen such an opportunity yet. But I believe that over time, someone will find a solution. Even now, some people can develop applications with strong user stickiness, such as ChatGPT, whose daily active users/monthly active users ratio is 16%. And if you look at Facebook or WeChat, the proportion is close to 90%. Today, the average ChatBot user usage time is about 10 minutes, compared with 100 or 120 minutes for social media.

So, we are still a long way from reaching that level of application. I think the entire world is currently anxiously looking for such super apps.

Forced to innovate

Olama: You just said that China is working hard to innovate to reduce costs. When we look at the development of China, it seems that this is true for every industry. Taking automobiles as an example, it is also about reducing costs and manufacturing new products. I think this is a way of technological innovation in China. What do you think?

Li Yanhong:If you look back over the past few hundred years, most innovations have been related to cost reduction, not just in artificial intelligence, or even just in the IT industry. If you can reduce costs by a certain amount, a certain percentage, then that means your productivity has increased by the same percentage. I think this is almost the essence of innovation. Today, innovation is much faster than before.

For China companies, we may be the first to feel the pain caused by these high costs. Let me give an example with Robotaxi. In the United States or more developed markets, online ride-hailing prices are much higher than in China. So, even a Robotaxi car, say, costs $100,000, can still make money by driving it alone. But in China, the price of hailing a car is much lower, so you have to come up with a much lower cost technology to achieve driverless driving. Therefore, to a certain extent, we are forced to innovate to reduce costs.

Autonomous driving is 10 times safer than human driving

Orama: Let’s talk about Apollo GO. Do you think China’s autonomous driving ecosystem is more advanced than the United States? How do you compare the two products?

Li Yanhong:There are many technical paths for autonomous driving. For example, Tesla does not use any sensors other than cameras, adopts a pure vision solution, relying on neural networks and algorithms to gradually transition from assisted driving to fully autonomous driving, and finally to completely unmanned driving. Our self-driving taxi is called Carrot Run. We used a different method. From the first day, we used a fully unmanned taxi to operate. Each of these methods has its own advantages.

OLAMA: To be honest, I think maybe at next year’s World Government Summit, all teams will be autonomous, backed by Carrot Run.

Li Yanhong:Yes, I hope so. You can see that road conditions in China are actually quite complex and challenging, right? Sometimes you will be jammed by a car that comes from nowhere, and motorcycles will shuttle through the traffic; sometimes you have to pass the bus in front of you. This is not an imaginary or some futuristic scene. This is something that happens every day in some cities in China.

Orama: If you want to deploy it in a city around the world, how long will it take from deciding to deploy it to actually deploy it and seeing it become a reality?

Li Yanhong:Technically, it will take about two weeks.

Orama: Autonomous driving is a good application of artificial intelligence. It heralds a positive future where people can enjoy driving, have fun, and improve productivity. Is there anything you are worried about?

Li Yanhong:Technology is advancing very rapidly. Although we are satisfied with the achievements we have made so far, think about the situation that will change a lot in six months or two years. Everyone is competing to each other, and the competition is fierce. This innovation also carries risks.

Although we have proven that Robotaxi is much safer than a human driving, it is at least 10 times safer. More than 1 million people die in traffic accidents every year around the world. But with Robotaxi, mortality rates can be greatly reduced. In addition, if you look at the accident rate from our actual records, our accident rate is only 1/14 of that of human drivers.

This is still a new industry and a new field, and people’s tolerance for accidents is very low. We attach great importance to safety issues. As of today, we have been operating on a relatively large scale for two or three years without any serious accidents.