Chen Xiaojian: Five key factors to build the universalization of Amazon Cloud Technology’s generative AI

Mr. Chen Xiaojian, General Manager of Amazon Cloud Technology Greater China Product Department

In the current "Battle of Hundreds of Models", generative artificial intelligence, represented by large models, is rapidly becoming the leader in the development of artificial intelligence at an unprecedented speed. However, generative AI doesn’t just mean big models. In late October, Amazon Cloud Technology held a Generative Artificial Intelligence Builders Conference. Chen Xiaojian, General Manager of the Product Department of Amazon Cloud Technology Greater China, spoke on the theme of "Empowering the new era of generative artificial intelligence and helping data and artificial intelligence benefit everyone." mentioned in his speech that the entire generative artificial intelligence application is like an iceberg floating on the sea. Only the corner exposed to the sea can be seen by most people. This corner is the basic model, and acceleration chips and databases are also needed at the bottom of the iceberg. , data analysis and data security services and other support services beyond the basic model. So, what kind of operational and service improvements can generative artificial intelligence bring to innovative companies? What contribution can Amazon Cloud Technology make in the field of generative artificial intelligence? The answer is already available at this conference of generative artificial intelligence builders, which brings together the upstream and downstream industries.

This conference has three major sub-venues, respectively discussing hot topics in the field of generative AI such as application base, data strategy and AI native. At the same time, in order to help more developers experience the charm of development in the generative AI era, the conference also set up a developer forum and power training camp. These activities are designed to provide developers with practical guidance and support, and can help developers better understand and apply generative AI technology. In addition, representative customers of Amazon Cloud Technology’s generative AI innovation, such as Siemens Group, Kingsoft Office Software, and Covestro China, also shared their innovation cases at this conference, providing valuable experience and inspiration to participants.

As the organizer of the conference, Amazon Cloud Technology is currently focusing on five key areas: application scenarios, tools and infrastructure, data foundation, AI native application construction and generative AI services to help enterprises and developers fully explore generative AI services. The potential of AI. Amazon CEO Andy Jassy once said: "Our goal is to allow anyone to obtain the same advanced infrastructure and cost as large enterprises to realize their own innovation." Currently, Amazon Cloud Technology provides a complete end-to-end generative system The AI ​​technology stack includes the bottom acceleration layer such as acceleration chips and storage optimization, the middle layer model building tools and services, and the top layer of generative AI-related applications. Innovation continues at every level to meet the different needs of customers. Amazon Cloud Technology hopes to popularize generative AI technology through related products and services, and provide support for more enterprises and individual developers to accelerate innovation

On the issue of how to realize the universalization of generative AI, Chen Xiaojian elaborated in detail on the five key factors that Amazon Cloud Technology focuses on.

First, choose the appropriate application scenario and start with typical scenarios to innovate the business model. Chen Xiaojian believes that enhancing customer experience, improving employee productivity and creativity, and optimizing business processes are the three main aspects of business value that application scenarios bring to enterprises. During his speech, he cited the Generative AI Productivity Frontier Technology Report released by McKinsey Consulting in June 2023. The report pointed out that about 3/4 of the economic benefits brought by generative AI technology come from marketing and sales, products and R&D There are four main functions: , software engineering and customer operations, which are precisely the main force of generative AI application direction. To this end, Salesforce has integrated Amazon Bedrock and Amazon Titan into its generative AI products, allowing customers to easily and securely use their data on Salesforce Data Cloud to build generative AI applications.

As a case study, Haier Innovation Design Center uses generative AI to achieve efficiency improvements in four aspects: graphical drawings, graph-generated drawings, quantitative drawings and full-scenario drawings. After going online, the application of automated design systems shortens the operation cycle of related businesses. 20%. At the same time, Mutong Technology, a leading company in the domestic gaming field, is also using Amazon Bedrock to optimize business processes related to game development.

Re-written as follows: Second, by leveraging purpose-built generative AI tools and infrastructure, cost-effective generative AI applications can be quickly built. In this area, the Amazon Bedrock service provided by Amazon Cloud Technology is combined with Amazon SageMaker Jumpstart to help customers with different needs for basic models easily and safely choose the basic model that suits them. Currently, Amazon Bedrock provides a wide range of basic models to choose from, including leading third-party providers such as Meta, Anthropic, Cohere, AI21 Labs, and Stability AI, as well as the Amazon Titan model family developed by Amazon Cloud Technology. Using Amazon Bedrock Agent, a managed agent that requires no coding, you can automatically break down and orchestrate tasks, connect to relevant data sources through APIs, and connect to Amazon Lambda on the backend to execute tasks. In addition, Amazon SageMaker JumpStart provides access to more open source models from industry and academia, and provides a deeply customized environment and evaluation capabilities

Amazon Cloud Technology provides a wide range of high-performance, low-cost training solutions that are highly flexible and cost-effective. For example, Amazon Cloud Technology's Amazon EC2 P5 instance is equipped with the latest Intex GPU chip H100 Tensor Core, which is 6 times faster than the previous generation and reduces training costs by 40%. In addition, Amazon Cloud Technology's Amazon EC2 Inf2 instance uses Amazon Cloud Technology's self-developed machine learning inference chip Amazon Inferentia2, which is 40% more cost-effective than other similar EC2 instances. Similarly, Amazon EC2 Trn1 instance of Amazon Cloud Technology uses Amazon Trainium, the self-developed machine learning training chip of Amazon Cloud Technology. Compared with similar instances, the training cost is saved by 50%

The rewritten content is as follows: First, lay a solid foundation for data and use privatized data to establish differentiated competitive advantages. Amazon Cloud Technology provides comprehensive data services, from storage, query, and analysis of data to the utilization of business intelligence, machine learning, and generative artificial intelligence, as well as easy integration and management of data, and effective security policies to manage the application and opening of data. Serve. For example, for application scenarios such as user personal information, session information management, and private domain knowledge base in the field of generative artificial intelligence, Amazon Cloud Technology has added vector database functions to Amazon OpenSearch Service, Amazon Aurora PostgreSQL, and Amazon RDS for PostgreSQL. In terms of data integration, Amazon Cloud Technology has proposed the concept of "zero ETL" and launched Aurora Zero ETL for Redshift Integration, which allows real-time generated business data to be synchronized from Amazon Aurora to the data warehouse Amazon Redshift without the need for ETL tools to achieve near real-time Big data aggregation analysis. In terms of data governance, Amazon Cloud Technology provides a new data governance service, Amazon DataZone, to reduce the heavy workload of internal members when accessing data and using analysis tools

At the meeting, the Siemens Dayu team shared the intelligent chat robot "Xiaoyu" they built with the support of Amazon cloud technology. The robot interacts through artificial intelligence generation. The most attractive part is the adoption of the "RAG architecture vector database" design: the core knowledge base is built in vector form and can store large-scale vector data. In addition, the RAG architecture greatly expands the usability of large models, allowing new parts to be processed using the same model without adjustments. In addition, because Amazon Cloud Technology provides a series of core technologies, including vector databases and generative artificial intelligence, the guidance completion rate of the entire solution is as high as 80%

Through cloud native services, the construction of AI applications can be accelerated, thereby promoting agile business innovation. Chen Xiaojian believes that in today's era of generative artificial intelligence, more customers need a native architecture, and proposed five details for this purpose. The first is a design framework with microservices and event-driven architecture as the core, which handles the dependencies between each functional module in a loosely coupled manner. Secondly, give priority to the use of Serverless architecture to reduce the operation, maintenance and deployment burden of the infrastructure, so that you can focus more on business logic and innovation. Third, put data decision-making first, regard data capabilities as the core competitive advantage of applications, and incorporate them into the design concept of generative artificial intelligence applications. Fourth, focus on security measures and adopt impact control methods to reduce the scope of potential risks, while placing security compliance and data protection in an important position. Finally, to avoid reinventing the wheel, in addition to focusing on the technology itself, we must continue to invest in modern application governance concepts such as DevOps, infrastructure as a service, and automation, promote the sharing of application assets and practices within the enterprise, and build an efficient and agile builder culture

Finally, by using out-of-the-box generative AI services, you can eliminate repetitive work and focus on innovation. To this end, Amazon Cloud Technology provides Amazon CodeWhisperer, an artificial intelligence programming partner, which can provide programming code suggestions in real time, fundamentally improving developer productivity. Compared to developers who did not use the tool, developers who used CodeWhisperer completed tasks 57% faster and had a 27% higher success rate. In addition, Amazon Whisperer has launched custom features that can generate better code suggestions. It allows customers to securely customize CodeWhisperer's code recommendations using private code libraries that can cover internal APIs, databases, best practices, architectural patterns, and more. At the same time, Amazon Cloud Technology also combines the Amazon Quicksight Q function with the large language model function provided by Amazon Bedrock to provide generative BI functions for Amazon QuickSight

Currently, Amazon Cloud Technology has helped more than a thousand small and medium-sized enterprises and startups quickly realize generative AI innovation through out-of-the-box generative AI services and tools, and has empowered more than 100,000 Chinese developers. .

"Developing generative AI applications is a systematic project full of challenges, and it is not a simple splicing of products and services." Chen Xiaojian finally added, "In addition to Amazon's own resources, we also need to establish a strong partner ecosystem. Work with them to solve various technical problems in the construction of generative AI applications and accelerate the commercialization of applications. In addition to providing cloud services, we also have solution architects, product technical experts, artificial intelligence laboratories, data laboratories, rapid We have multiple resources such as prototype team, professional service team, training and certification department to jointly help customers succeed. At the same time, we also work with ecological partners and start-up circles to build a complete generative AI system to further promote the application of generative AI technology. .”

The above is the detailed content of Chen Xiaojian: Five key factors to build the universalization of Amazon Cloud Technology’s generative AI. For more information, please follow other related articles on the PHP Chinese website!