"There are hundreds of basic models, which is a huge waste of social resources, especially when our computing power is still limited." Robin Lee, Baidu's founder, chairman and CEO, said in a conversation on December 16.
Robin Lee said that the term artificial intelligence has been put forward for more than 70 years. After several waves and ups and downs, many people are a little tired, but the emergence of the big model made him very excited and mobilized almost all the resources of the company to do related things.
Why is the technology wave of this big model different from previous times, and what are the differences? Robin Lee said: "It is universal. We call it 'intelligent emergence'. It has learned what we have not taught. With this feature, when you have a set of basic technologies that can be done very well and are very advanced, it can quickly make valuable applications in a variety of scenarios. This is something that AI has never done in the past 70 years, so it is a completely different opportunity."
Robin Lee reiterated that his view has not changed, that is, the main opportunity in the future is the AI application based on the model, but in the past year, the main excitement of the media, society and the public is still based on the basic model, and has not shifted to the AI native application. Therefore, in recent public speeches and internal company speeches, he has been constantly emphasizing that it is necessary to focus on the native application of AI, and only when the application is developed can the model have value.
The surging journalists found that if you look back at Robin Lee's views on the big model in recent months, most of them focus on "application". For example, at the Xili Lake Forum, he pointed out that there are too many large models in China, but there are too few native applications developed based on these models; At the 2023 World Internet Conference Wuzhen Summit Internet Entrepreneurs Forum, he stated that only with millions of AI applications can big models be considered successful.
In the above conversation, Robin Lee once again stressed that there is only one key point in the development of the big model, namely, application; There are three key factors for developing good applications: firstly, there are relevant industrial policies to encourage the development of AI native applications based on large models; Secondly, existing enterprises utilize large models to have a positive impact on their core business key indicators; The third is super applications. When and in which field can they appear? More startup companies need to work harder and make various attempts.
Another noteworthy feature is that Google has released the highly anticipated new AI model Gemini, and there are also rumors in the industry that OpenAI's GPT4.5 will also be released. Has the gap between domestic technology companies and foreign companies narrowed or widened in the past period of time?
In this regard, Robin Lee said that the market environment between China and the United States is different, especially at the application level, and the development direction may also be different. The United States is more advanced in enterprise level software, while China will be more advanced in the field of To C, "If we go back to the technology gap between China and the United States, my view is always that technology should still serve applications. When you say that this technology is good or that technology is bad, what do you mean? Today, our PPT generation capability of Baidu Wenku is the best in the world, which is based on ERNIE Bot's big model. At this time, I don't think it's worth mentioning how far we are from the world's most advanced level Is there a gap. But if you talk about enterprise level applications, the market in China is too small, and we are not prepared to do any optimization for these things, it may indeed be outdated
Robin Lee once again stressed that the progressiveness of model technology depends more on what the model does in what application scenarios
Robin Lee said that the statement that the big model "vigorously works miracles" more refers to the process of exploration of the big model from 0 to 1. Moving forward, whoever has higher efficiency wins.
In the past few months, Baidu has conducted AI original biochemical reconstruction and access to its various product lines, while some external companies have begun to access ERNIE Bot.
Robin Lee said that competing with the word "access" is actually competing with muscle memory. Access is the simplest, but also the lowest value, "What is the relationship between your business and the big model? Can the big model help you increase your DAU, retention rate, user experience, revenue, and profits? These are the key indicators of your business."
On November 21st, Baidu released its third quarter financial report, which showed that Baidu's revenue reached 34.447 billion yuan, a year-on-year increase of 6%; The net profit attributable to Baidu (non GAAP) reached 7.3 billion yuan, a year-on-year increase of 23%, and both revenue and net profit exceeded market expectations. Robin Lee said at the financial report conference that the reconstructed advertising system of Wenxin model will bring hundreds of millions of yuan of incremental revenue in the fourth quarter. The ban on the chip industry in the United States has limited impact on Baidu in the short term, as Baidu has a large reserve of AI chips that can allow it to continuously update its content model in the next one or two years.