【观点】邬贺铨院士：大模型上云落地，能够加速形成新质生产力

12月19日，在“势起AI 智启新界”大模型创新应用与安全治理大会上，中国工程院院士、中国互联网协会咨询委员会主任邬贺铨发表《上云落地大模型，数字经济谱新篇》主旨演讲。

他指出，大模型上云落地，能够加速形成新质生产力，推动数字经济的提质换档。在他看来，广东是制造业大省，希望广东的基础大模型能够跟行业更好地结合。

一、ChatGPT还不算是通用人工智能大模型

从1997年IBM“深蓝”，到2016年AlphaGo下围棋战胜李世石，在演讲中，邬贺铨深入浅出地讲述了AI的演进过程。“直到2022年年底，ChatGPT出现，有了智能涌现，但还不算通用大模型。”在他看来，作为通用大模型应该是知识与数据的结合，并加上类脑的智能。

 “AI模型参数越多训练结果越精细，参数量呈指数级增长，AI模型参数平均2年增25倍，Transformer模型2年增275倍”，邬贺铨认为，通用大模型主要体现三个方面：从单模型单任务到多任务，从输入文字到文、图、视听等多模态数据并融合，对每一任务仅动用模型中相关部分解算无须全启动。

 在邬贺铨看来，现在都说大模型，但大部分都是基于语料训练的大模型。目前中国大多基础大模型，是互联网公司、IT公司开发。“目前来看，垂直行业的大企业还没有一个声称自己做大模型”，他说，尽管互联网企业做了很多大模型，实际尚未到行业应用，因为不具有行业知识。 CXO UNION-CXO联盟（cxounion.cn）

二、寄语广东把基础大模型与行业更好结合

在邬贺铨看来，随着AI技术兴起，由数据驱动、平台支撑、网络协同的经济活动单元所构成的新经济形态上升为新质生产力，推动数字经济提质换挡。

“广东是制造业大省，因此我希望广东的基础大模型能够跟行业更好结合”，在接受南都采访时，邬贺铨认为广东大模型发展走在全国前列，希望能够助力行业高质量发展，体现广东大模型的研究开发贡献。

邬贺铨说，随着人工智能智能化、大模型，更多企业正在搭建服务于社会的平台，他多次以广东企业举例。例如广州一家企业的云设计创意工具平台为服装企业提供服装设计建模工具软件和生产全流程数字化管理，设计、打版效率分别提升30%和100%。深圳某企业的云在线制造服务平台，一方面承接需求方发布产品订货需求，另一方面连接供应方，仅2021年上半年撮合交易采购100多亿元，增长480%。 CXO UNION-CXO联盟（cxounion.cn）

 “善于应用数据的企业成为经济增长的主要引擎”，邬贺铨认为，30年来全球市值前十强从金融到IT再到能源，近十年来数字化科技创新和平台企业为主导，现在位居前列的基本都是数智化企业。数字技术推动社会经济的进步，未来会越来越依靠数据，需要我们更好发挥数据作为新型生产要素的作用。

三、把大模型装进手机，影响将超越iPhone

在演讲中，邬贺铨描摹大模型应用将给人们生活、工作带来的深刻变化。他提到，通过模型压缩和定制AI芯片提升算力，将具有百亿参数大模型的推理能力嵌入手机，可实现提供低成本、低时延、高安全的智能助手服务。“比如同声传译，视频会话中能匹配口型，实现手语、盲文的实时翻译。将来你和外国朋友打电话，你听不懂，只要手机上安装中文键，对方说英文，你听到是中文”，他说，随之而来的还有歌曲创作、智能识图、照片扩展、眼动追踪、个性服务等，将大大丰富人们的数智生活。 CXO UNION-CXO联盟（cxounion.cn）

他说，大模型落地催生新一轮技术创新与产业变革。大模型是自2007年iPhone商用15年来互联网的新浪潮，其影响将超过当年的iPhone问世，嵌入大模型的智能终端在动漫与视频创意、游戏和虚拟世界、语音合成与转换、视觉和图像处理、聊天机器人、个性化推荐、医疗辅助、自动化内容生成、人脸识别等领域广泛应用。具有智能涌现的轻量化大模型也在工业领域得到广泛应用，例如机器人、无人机、AGV、工业模组、边缘计算等，应用场景包括产品的创意设计、机器视觉的智能质检、语音交互的远程操控。IDC预测，到2026年中国市场中近50%的终端设备的处理器将带有AI引擎技术，终端产业将迎来一次激动人心的井喷式发展。

翻译：

Academician Wu Hequan: The big model can accelerate the formation of new quality productivity

On December 19, at the “Potential AI Zhiqi New World” large model innovation application and security governance conference, Wu Hequan, academician of the Chinese Academy of Engineering and director of the Advisory Committee of the Internet Society of China, delivered a keynote speech on “Shanghai Cloud Landing Large Model, New Chapter of the Digital economy Spectrum”.

He pointed out that the cloud landing of large models can accelerate the formation of new quality productivity and promote the quality shift of the digital economy. In his view, Guangdong is a large manufacturing province, and he hopes that the basic model of Guangdong can be better combined with the industry. CXO UNION-CXO联盟（cxounion.cn）

First, ChatGPT is not yet a grand model of general artificial intelligence

From IBM’s “Deep Blue” in 1997 to AlphaGo’s victory over Lee Sedol in 2016, Wu Hequan told the evolution of AI in a simple way in his speech. “It’s not until the end of 2022 that ChatGPT comes along and there’s intelligence emerging, but it’s not yet a universal grand model.” In his view, as a universal grand model, it should be a combination of knowledge and data, plus brain-like intelligence.

“The more AI model parameters, the more refined the training results, and the number of parameters increases exponentially. The average AI model parameters increase by 25 times in 2 years, and the Transformer model increases by 275 times in 2 years.” Wu Hequan believes that the general large model mainly reflects three aspects: From single model single task to multi-task, from input text to multi-modal data such as text, picture, audio-visual and so on, and fusion, only the relevant part of the model is used for each task without full start-up. CXO UNION-CXO联盟（cxounion.cn）

In Wu Hequan’s view, large models are now said, but most of them are large models based on corpus training. At present, most of the basic large models in China are developed by Internet companies and IT companies. “At present, no large enterprise in the vertical industry has claimed to make a large model,” he said, although Internet companies have made a lot of large models, the actual application has not been applied to the industry, because it does not have industry knowledge.

Second, a message to Guangdong to better combine the basic model with the industry

In Wu Hequan’s view, with the rise of AI technology, the new economic form composed of economic activity units driven by data, supported by platforms and coordinated by networks has risen to new quality productivity, promoting the quality and shift of the digital economy.

“Guangdong is a large manufacturing province, so I hope that Guangdong’s basic large model can be better combined with the industry”, in an interview with Nandu, Wu Hequan believes that Guangdong’s large model development is in the forefront of the country, hoping to help the high-quality development of the industry, reflecting the contribution of Guangdong’s large model research and development.

Wu Hequan said that with the intelligent and large model of artificial intelligence, more enterprises are building a platform to serve society, and he repeatedly cited Guangdong enterprises as an example. For example, the cloud design creative tool platform of a Guangzhou enterprise provides clothing design modeling tool software and digital management of the entire production process for clothing enterprises, and the design and printing efficiency are increased by 30% and 100% respectively. The cloud online manufacturing service platform of an enterprise in Shenzhen, on the one hand, to undertake the demand side to issue product orders, on the other hand, to connect the supply side, and only in the first half of 2021, the transaction procurement of more than 10 billion yuan, an increase of 480%.

“Enterprises that are good at applying data have become the main engine of economic growth”, Wu Hequan believes that in the past 30 years, the top ten global market value from finance to IT to energy, digital technology innovation and platform enterprises have been dominated in the past decade, and the basic companies in the forefront are digital intelligent enterprises. Digital technology promotes social and economic progress, and the future will increasingly rely on data, which requires us to better play the role of data as a new factor of production. CXO UNION-CXO联盟（cxounion.cn）

Third, put the large model into the phone, the impact will exceed the iPhone

In his speech, Wu Hequan described the profound changes that the application of large models will bring to people’s lives and work. He mentioned that through model compression and customized AI chips to improve computing power, the reasoning ability of a large model with 10 billion parameters is embedded in the mobile phone, which can provide low-cost, low-delay and high-security intelligent assistant services. “For example, simultaneous interpretation, video conversations can match mouth movements to achieve real-time translation of sign language and Braille.” In the future, you and foreign friends call, you do not understand, as long as the Chinese key is installed on the mobile phone, the other party speaks English, you hear is Chinese, “he said, followed by song creation, intelligent recognition, photo expansion, eye tracking, personality services, etc., will greatly enrich people’s digital intelligence life.

He said that the landing of the large model has spawned a new round of technological innovation and industrial change. The Big model is the new wave of the Internet in the 15 years since the iPhone came out in 2007, and its impact will exceed that of the iPhone. Intelligent terminals embedded with large models are widely used in animation and video creativity, games and virtual worlds, speech synthesis and conversion, vision and image processing, chatbots, personalized recommendation, medical assistance, automated content generation, face recognition and other fields. Lightweight large models with intelligent emergence are also widely used in the industrial field, such as robots, drones, AGVs, industrial modules, edge computing, etc. Application scenarios include product creative design, intelligent quality inspection of machine vision, remote control of voice interaction. IDC predicts that by 2026, nearly 50% of terminal equipment processors in the Chinese market will have AI engine technology, and the terminal industry will usher in an exciting blowout development.

由CXO UNION-CXO联盟（cxounion.cn）转载而成，来源于南方都市报；编辑/翻译：CXO UNIONCXO联盟小U。

如需加入CXO UNION（CXO联盟）高管社群，请联系社群小伙伴哦~

免责声明: 本网站(http://www.cxounion.cn/)内容主要来自原创、合作媒体供稿和第三方投稿，凡在本网站出现的信息，均仅供参考。本网站将尽力确保所提供信息的准确性及可靠性，但不保证有关资料的准确性及可靠性，读者在使用前请进一步核实，并对任何自主决定的行为负责。本网站对有关资料所引致的错误、不确或遗漏，概不负任何法律责任。

本网站刊载的所有内容(包括但不仅限文字、图片、LOGO、音频、视频、软件、程序等) 版权归原作者所有。任何单位或个人认为本网站中的内容可能涉嫌侵犯其知识产权或存在不实内容时，请及时通知本站，予以删除。