在2024年世界人工智能大會(huì)的產(chǎn)業(yè)發(fā)展主論壇上，百度創(chuàng)始人李彥宏發(fā)表了演講。他強(qiáng)調(diào)，在人工智能時(shí)代，開(kāi)發(fā)具有高度實(shí)用性的“超級(jí)能干”應(yīng)用比追求日活躍用戶(hù)數(shù)量（DAU）達(dá)到10億的“超級(jí)應(yīng)用”更為重要。李彥宏認(rèn)為，應(yīng)該超越移動(dòng)時(shí)代的思考模式，避免陷入“超級(jí)應(yīng)用陷阱”，認(rèn)識(shí)到成功的定義不應(yīng)僅限于用戶(hù)數(shù)量。
李彥宏提出，智能體是開(kāi)發(fā)AI應(yīng)用的最簡(jiǎn)單形式，也是他最看好的AI應(yīng)用發(fā)展方向。他指出，開(kāi)源大模型在學(xué)術(shù)研究和教學(xué)領(lǐng)域有其價(jià)值，但在商業(yè)競(jìng)爭(zhēng)中，閉源模型因其效率和成本優(yōu)勢(shì)更為適用。他解釋說(shuō)，開(kāi)源模型需要更大的參數(shù)規(guī)模才能與閉源模型競(jìng)爭(zhēng)，這會(huì)導(dǎo)致更高的推理成本和更慢的反應(yīng)速度。此外，使用開(kāi)源模型可能導(dǎo)致無(wú)法共享算力和從基礎(chǔ)模型的持續(xù)升級(jí)中獲益。
他還提到，百度在小說(shuō)創(chuàng)作上從開(kāi)源模型轉(zhuǎn)向閉源模型后，小說(shuō)生成的可用率和優(yōu)質(zhì)率有了顯著提升。李彥宏認(rèn)為，大模型的價(jià)值在于其在具體應(yīng)用中的表現(xiàn)，而不僅僅是模型本身。
關(guān)于AI應(yīng)用的重要性，李彥宏舉例說(shuō)明，AI應(yīng)用在快遞行業(yè)通過(guò)大模型處理訂單，顯著提高了效率。他還提到，百度的文心快碼在代碼生成領(lǐng)域正在逐步滲透，并在百度內(nèi)部已有約30%的代碼由AI生成。
李彥宏預(yù)測(cè)，隨著基礎(chǔ)模型能力的增強(qiáng)，開(kāi)發(fā)智能體將變得越來(lái)越簡(jiǎn)單，未來(lái)將出現(xiàn)數(shù)百萬(wàn)量級(jí)的智能體，形成龐大的生態(tài)系統(tǒng)。他以百度的高考智能體為例，說(shuō)明了智能體在實(shí)際應(yīng)用中的巨大潛力。
最后，李彥宏討論了AI是否會(huì)取代人類(lèi)工作的問(wèn)題，他認(rèn)為AI目前更多是作為輔助工具，而不是替代者。AI技術(shù)的發(fā)展已經(jīng)創(chuàng)造了一些新的工作機(jī)會(huì)，如數(shù)據(jù)標(biāo)注師和提示詞工程師。他強(qiáng)調(diào)，AI永遠(yuǎn)只是工具，目的是為了滿足人類(lèi)需求、增強(qiáng)人類(lèi)能力，讓生活更美好。
At the main forum of the 2024 World Artificial Intelligence Conference on industrial development, Baidu’s founder Robin Li delivered a speech. He emphasized that in the era of artificial intelligence, developing “super capable” applications with high practicality is more important than pursuing “super applications” with a daily active user count (DAU) of 1 billion. Robin Li believes that we should transcend the thinking patterns of the mobile era, avoid falling into the “super application trap,” and recognize that the definition of success should not be limited to the number of users.
Robin Li proposed that intelligent agents are the simplest form of developing AI applications and are also the direction of AI application development he is most optimistic about. He pointed out that open-source large models have their value in academic research and teaching fields, but in commercial competition, closed-source models are more suitable due to their efficiency and cost advantages. He explained that open-source models require a larger parameter scale to compete with closed-source models, which would lead to higher inference costs and slower response speeds. In addition, using open-source models may lead to an inability to share computing power and benefit from the continuous upgrades of the basic models.
He also mentioned that after Baidu shifted from open-source models to closed-source models in novel creation, the usability and quality of the novel generation have significantly improved. Robin Li believes that the value of large models lies in their performance in specific applications, not just in the models themselves.
Regarding the importance of AI applications, Robin Li illustrated with examples that AI applications have significantly improved efficiency in the express delivery industry by processing orders through large models. He also mentioned that Baidu’s Wenxin Quick Code is gradually penetrating the field of code generation, and about 30% of the code within Baidu is already generated by AI.
Robin Li predicts that with the enhancement of the capabilities of basic models, the development of intelligent agents will become increasingly simple, and in the future, there will be millions of intelligent agents forming a huge ecosystem. He used Baidu’s college entrance examination intelligent agent as an example to illustrate the great potential of intelligent agents in practical applications.
Finally, Robin Li discussed whether AI will replace human work, and he believes that AI is currently more of an auxiliary tool rather than a substitute. The development of AI technology has created some new job opportunities, such as data annotators and prompt engineers. He emphasized that AI is and will always be just a tool, aimed at meeting human needs and enhancing human capabilities to make life better.
