Home > Article > Technology peripherals > Robin Li talks about AI in Europe: I am more interested in super applications than GPT-5
The biggest difference between Zhonggu AI and the West is application. There are hundreds of basic models in China, but people are increasingly discussing what are the super applications in the AI era.
Applications are driving the rapid development of AI. Progress in the application field can promote innovation in basic models and help accelerate the transition from the Internet era to the artificial intelligence era.
At the main forum of the "European AI Spring Festival Gala", Robin Li made his latest appearance. As the only Chinese entrepreneur invited to the conference, he once again "spoke" for China's AI.
The full name of this conference being held in Paris, France is "Viva Technology", which is one of the largest technology events in Europe.
In addition to Robin Li, speakers participating this year include Musk, Turing Award winners Yoshua Bengio and Yann LeCun, Anthropic CEO Dario Amodei, Mistral AI CEO Arthur Mensch, etc. It can be said that it brings together the most cutting-edge AI technology power from different perspectives around the world.
Let’s take a look at the specific content shared by Robin Li.
In this conversation with Maurice Levy, Chairman of the Supervisory Board of Publicis Groupe, Robin Li "APP" was mentioned 22 times.
The reason is that when it comes to the choice of large model development route, "I think China is taking a more application-driven approach".
In the mobile Internet era, there are a large number of super applications with hundreds of millions or even billions of daily active users. But now, we have not seen such native applications of AI.
What is the most likely form of super application? Whether it is a chatbot or Copilot is not yet determined.
The unknown space of technological change is full of challenges but also lurks huge opportunities. Therefore, “In China, although there are hundreds of basic models, everyone, whether it is a start-up company or a major Internet company, is working hard to find PMF (product market fit) and is committed to exploring how to best utilize the capabilities of generative AI. An application form that can be used by billions of people."
Robin Li also revealed that since its launch last year, the cumulative number of Wen Xinyiyan users has reached 200 million, and tens of millions of users are using Baidu every day. AI functions for applications such as search and Baidu Library.
Super applications have not yet been formed, but changes are already underway. "New AI applications will gradually take shape" and "the future will be very different from what we see today."
It is worth mentioning that Robin Li emphasized that focusing on applications does not mean ignoring the development of basic models. On the contrary, AI applications and the development of basic large models complement each other.
The progress of applications can promote the innovation of basic models and help accelerate the transition from the Internet era to the artificial intelligence era.
He gave examples of applications such as Baidu search and live e-commerce digital people: after the application side obtains real user needs and problems, it can continuously feed back to the model, thereby promoting the technical iteration of the basic model.
At the same time, the birth of super applications will also promote the generation of large amounts of data and provide more abundant training data for large models.
Some people say that as the model parameters become larger and larger, we will soon use up all the existing data, but I think that when super applications are born, data will be generated in large quantities, so I do not do not worry.
During the conversation, Robin Li also mentioned the positioning of Baidu’s basic model:
Our most important market is in China, so In Chinese, we must do better than any model in the world. Now, the Chinese capability of Wenxin Large Model 4.0 has surpassed GPT-4.
And this is also an opportunity for countries around the world to develop their own models: local companies have the opportunity to make the basic models for their own languages better.
Let’s talk about Viva Technology (VivaTech for short). This is one of the most watched technological innovation events in Europe. In previous conferences, the President of France has attended, and the chairman and CEO of technology giants such as Google, Microsoft, IBM, and Tesla are regular guests.
今年的VivaTech在法國巴黎凡爾賽門展覽中心舉行,有超過2500家新創公司、2000名國際投資者、25個行業和350多家公司和組織參與。
在大會開始之前,VivaTech官方就已經預告了李彥宏的出席。而邀請李彥宏的原因是,他是「長期的AI支持者」和「中國AI頭號玩家」。
大模型掀起的風暴中,中國無疑是最受關注的技術變革風暴眼之一,正如莫里斯·列維所說,「人們普遍認為人工智慧領域的兩大主導力量是美國和中國」。
從百模大戰到應用探索的生成式AI第二階段,中國AI界一直在不斷發出強而有力的聲音。
在全球各大模式評測榜單上,TOP10中國產大模型的身影並不鮮見。
例如,在今年3月清華大學基礎模型研究中心發布的《SuperBench大模型綜合能力評測報告》中,文心一言4.0在數學能力上的得分追平Claude 3,並在閱讀理解能力上超過GPT-4 Turbo、Claude-3拿下榜首。
中文能力方面,百度文心一言4.0、智譜GLM-4、通義千問、Kimi等國產大模型更是顯現出優勢,有超越GPT-4、Claude 3的表現。
此外,在基於人類對比評價的大模型競技場上,也不斷有國產大模型刷新榜單記錄,躋身世界前列。
這個過程中,有目共睹的是,李彥宏和他領導下的百度,始終站定在一線推動著國產大模型的快速更新迭代和落地應用。
去年,李彥宏也和馬斯克、黃仁勳、山姆·奧特曼等人一起被《時代》周刊評為全球AI領袖,評語是:
李彥宏是中國最傑出的未來主義者,長期投入AI發展的浪潮。
目前,文心大模型API日均調用量已突破2億,相較於23年Q4財報披露的超5000萬次,日調用量環比增長300%,成為國內應用最廣泛的AI基礎模型。
與三星、榮耀達成合作後,今年一季度,小米、OPPO、vivo等手機廠商均接入文心大模型API,聯想、蔚來汽車也在季度內與百度達成合作。
傳統產業,也已經吹響了大模型重構一切的號角。百度智能雲已攜手上海山源電子科技股份有限公司,基於文心大模型,依托百度智能雲千帆大模型平台能力共同打造了礦山行業智能應用“電博士及數位調度員”。
透過注入供電專業相關知識、經驗、規則、規範等,「電博士」具備煤礦供電的專業知識和技能,能實現陪伴式的煤礦供電輔助運維。
而在百度內部,繼去年「百度要把全部產品用大模型重構一遍」之後,今年5月,李彥宏再次發聲:
#大模型我們走在最前面,我們要去勇闖無人區,需要去冒前人沒有冒過的風險。
十個創新,可能九個最後都是以失敗告終的,但我們不會放棄,我們會去做第11個、第12個…我相信終有一天,我們會向世界證明,技術是有價值的。
#事實上,百度用大模型重構現有業務的進程一年來不斷有新消息傳出,核心業務亦涵蓋在內:
搜尋方面,基於文心大模型,百度搜尋整合極致滿足、推薦激發與多輪互動三大特色。在搜尋框輸入問題,搜尋引擎可以不只是給出鏈接,而是直接產生最好的答案,解決用戶不會搜、搜不準的問題。
廣告行銷方面,百度推出了AI原生的新一代行銷平台“輕舸”和大模型商業引擎“揚楫”,顛覆網路廣告投放模式,廣告主只需自由地用語言表達需求,就能搞定行銷全流程。
還有百度文庫,基於大模型,更是革了自己的命,從“內容工具”變身“生產力工具”,不僅可以查找現成的內容,還可以撰寫文章、製作PPT,直接生成創作新的內容。
如李彥宏所說,真正的大模型超級應用尚未被發明,但在這種破釜沉舟用大模型重建一切的試驗之中,通往AGI的「破繭時刻」正越來越近。
事實上,李彥宏這次在歐洲AI春晚的亮相,側面反映出了國際市場對中國AI的矚目。
以百度為代表,中國AI已經展現出技術累積、應用成果,並且還有巨大的場景機會。隨著大模型在應用端的探索不斷深入,相信中國AI話語權仍會持續增強。
當被問及對人工智慧技術的擔心時,李彥宏的回答是:
我的擔心是人工智慧技術的進步速度不夠快。每個人都對過去幾年AI技術的快速發展感到震驚,但對我來說,它太慢了。
李彥宏感慨,很多人認為AGI(通用人工智慧)可能還要兩年或五年的時間,但要真正抵達理想中的AGI時代,「我認為還要十年以上」。
The above is the detailed content of Robin Li talks about AI in Europe: I am more interested in super applications than GPT-5. For more information, please follow other related articles on the PHP Chinese website!