李彥宏歐洲談AI：比起GPT-5，我更有興趣超級應用
- 王林原創
- 2024-06-10 22:55:23557瀏覽
中固 AI 與西方的最大差異是應用。中國有數百個基礎模型，但人們越來越多地在討論什麼是 AI 時代的超級應用。
應用程式正在驅動中的AI的快速發展。應用領域的進步可以推動基礎模型的創新，也有助於加速從網路時代到人工智慧時代的轉變。
「歐洲AI春晚」主論壇現場，李彥宏最新亮相，作為大會唯一受邀中國企業家，再一次為中國AI「代言」。
這場正在法國巴黎舉辦的大會全名為「歐洲科技創新展」（Viva Technology），是歐洲規模最大的科技盛會之一。
李彥宏之外，今年參與的演講嘉賓包括馬斯克、圖靈獎得主Yoshua Bengio和Yann LeCun，還有Anthropic執行長Dario Amodei、Mistral AI執行長Arthur Mensch等等。可以說匯聚了全世界來自不同視角的最前沿AI科技力量。
一起來看李彥宏具體分享內容。
「比起GPT-5，更重要的是超級應用」
在這場與陽獅集團監事會主席莫里斯·列維（Maurice Levy）的對談中，李彥宏提到了22次「應用」。
原因是，在大模型發展路線的選擇上，「我認為中國走的是一種更偏向於應用驅動的方式」。
行動網路時代，有大量每日活躍用戶數量達到數億、甚至十億的超級應用程式。但現在，我們尚未見到這樣的AI原生應用。
超級應用程式最可能的形態是什麼？是聊天機器人，還是Copilot，目前還不確定。
技術變革的未知空間，充滿挑戰的同時也潛藏著巨大的機會。因此，「在中國，儘管也有數百個基礎模型，但無論是新創公司還是網路大廠，每個人都在努力尋找PMF（產品市場契合度），致力於探索最能發揮生成式AI能力、能被數十億人使用的應用形態。搜尋、百度文庫等應用程式的AI功能。
超級應用尚未形成，但變化已經正在進行中，「全新的AI應用會逐漸成型」、「未來將與我們今天看到的大不相同」。
值得一提的是，李彥宏強調，關注應用並非是忽略基礎模型的發展。與之相反，
。
應用程式的進步，可以推動基礎模式的創新，也有助於加速從網路時代到人工智慧時代的轉變。他以百度搜尋、直播電商數位人等應用範例：應用端獲取真實的使用者需求和問題後，可以持續回饋給模型，並由此推進基礎模型的技術迭代。
同時，超級應用程式的誕生也將促進大量資料生成，為大模型提供更充沛的訓練資料。
有人說，隨著模型參數越來越大，我們很快就會用完所有現存數據，但我認為，當超級應用程式誕生時，數據會大量生成，所以我並不擔心。
對話中，李彥宏也提及了百度基礎模型的定位：
而這也是世界各國發展自己模型的機會：本土公司，有機會將面向自身語種的基礎模型做到更好。
「中國AI的頭號玩家」
再來說說Viva Technology（簡稱VivaTech）。這是歐洲最受關注的科技創新盛會之一，歷屆大會中，法國總統均到場出席，Google、微軟、IBM、特斯拉等科技巨頭的董事長和CEO都是常客。
This year’s VivaTech was held at the Porte de Versailles exhibition center in Paris, France, with more than 2,500 start-ups, 2,000 international investors, 25 industries and more than 350 companies and organizations participating.
Before the conference started, VivaTech officials had already announced Robin Li's attendance. The reason for inviting Robin Li is that he is a "long-term AI supporter" and "China's number one AI player."
In the storm caused by large models, China is undoubtedly one of the most watched eyes of the technological change storm. As Maurice Levy said, "People generally believe that artificial intelligence The two leading forces in the field of intelligence are the United States and China."
From the war of hundreds of models to the second stage of generative AI in application exploration, the Chinese AI community has been making strong voices.
It is not uncommon to find Chinese-made large models in the TOP10 on the evaluation lists of various large models around the world.
For example, in the "SuperBench Large Model Comprehensive Ability Evaluation Report" released by the Basic Model Research Center of Tsinghua University in March this year, Wenxinyiyan 4.0 tied Claude 3 in math ability, and in reading In terms of understanding ability, it surpassed GPT-4 Turbo and Claude-3 to take the top spot.
In terms of Chinese language ability, large domestic models such as Baidu Wenxinyiyan 4.0, Zhipu GLM-4, Tongyi Qianwen, and Kimi have shown advantages and surpassed Performance of GPT-4 and Claude 3.
In addition, in the large model arena based on human comparative evaluation, domestic large models continue to refresh the list record and rank among the top in the world.
In this process, it is obvious to all that Robin Li and Baidu under his leadership have always stood on the front line to promote the rapid iteration and implementation of domestic large models.
Last year, Robin Li was named a global AI leader by Time magazine along with Musk, Huang Renxun, Sam Altman and others. The comment was:
Robin Li is China's most outstanding futurist has long been devoted to the wave of AI development.
At present, the average daily calls of Wenxin Big Model API have exceeded 200 million. Compared with the over 50 million times disclosed in the Q4 financial report of 2023, the daily calls have increased by 300% month-on-month, becoming a domestic application. The most extensive AI basic model.
After reaching cooperation with Samsung and Honor, in the first quarter of this year, mobile phone manufacturers such as Xiaomi, OPPO, and vivo all accessed Wenxin Big Model API, and Lenovo and NIO also reached cooperation with Baidu during the quarter.
Traditional industries have also sounded the clarion call to reconstruct everything with large models. Baidu Intelligent Cloud has joined hands with Shanghai Shanyuan Electronic Technology Co., Ltd. to jointly create the intelligent application "Electronic Doctor and Digital Dispatcher" in the mining industry based on the Wenxin large model and relying on the capabilities of the Baidu Intelligent Cloud Qianfan large model platform.
By injecting relevant knowledge, experience, rules, specifications, etc. in the power supply profession, "Dr. Electricity" has the professional knowledge and skills of coal mine power supply and can realize accompanying auxiliary operation and maintenance of coal mine power supply.
Within Baidu, after "Baidu will reconstruct all products with large models" last year, in May this year, Robin Li once again spoke out:
We are at the forefront of the big model. We have to bravely venture into no man's land and take risks that no one has ever taken before.
Ten innovations, maybe nine will end in failure, but we will not give up, we will do the 11th, 12th... I believe that one day, we will show the world Prove that technology is valuable.
In fact, new news has been coming out over the past year about Baidu’s process of using large models to reconstruct existing businesses, and its core businesses are also covered:
In terms of search, based on text Based on the Xinda model, Baidu search integrates the three major characteristics of ultimate satisfaction, recommendation stimulation and multiple rounds of interaction. When you enter a question in the search box, the search engine can not only provide links, but directly generate the best answers, solving the problem of users not being able to search or searching accurately.
In terms of advertising and marketing, Baidu has launched a new generation of AI-native marketing platform "Qingge" and a large-model business engine "Yangji", subverting the Internet advertising model. Advertisers only need to freely express their needs in words. , you can handle the entire marketing process.
There is also Baidu Wenku, which is based on large models and has revolutionized its own life. It has transformed from a "content tool" to a "productivity tool". It can not only search for ready-made content, but also write articles, make PPT, and directly generate creations. New content.
As Robin Li said, the real large-model super application has not yet been invented, but in this kind of experiment of rebuilding everything with large models, the "breaking cocoon moment" leading to AGI is becoming more and more close.
In fact, Robin Li’s appearance at the European AI Spring Festival Gala reflects the international market’s attention to China’s AI.
Represented by Baidu, China’s AI has demonstrated technological accumulation and application results, and there are also huge scenario opportunities. As the exploration of large models on the application side continues to deepen, I believe that China’s AI voice will continue to increase.
One More Thing
When asked about his worries about artificial intelligence technology, Robin Li’s answer was:
My worry is the progress of artificial intelligence technology Not fast enough. Everyone is shocked by the rapid development of AI technology in the past few years, but to me, it is too slow.
Robin Li lamented that many people think that AGI (artificial general intelligence) may take two or five years, but to truly reach the ideal AGI era, "I think it will take more than ten years." ".
以上是李彥宏歐洲談AI：比起GPT-5，我更有興趣超級應用的詳細內容。更多資訊請關注PHP中文網其他相關文章！