The close and strong relationship between humans, data and AI is slowly unfolding in front of us. Powered by data and AI, generative AI is becoming the source of power to promote application innovation in thousands of industries and the focus of development in the field of science and technology. .
Amazon Cloud Technologies highlighted how the technology is at the top of the cloud giant's agenda at this year's re:Invent 2023 conference. During today’s keynote, Amazon Cloud CEO Adam Selipsky said: “Innovation around generative AI models is explosive.” He added: “It will reshape every application we interact with at work and at home. .We are approaching the whole concept of generative AI in a completely different way than before.
Amazon Cloud Technology is constantly reconstructing the three levels of full-stack generative AI: the bottom layer is the infrastructure layer for training and inference, the middle layer is all the tool services required to fine-tune the model, and the upper layer is the layer for building generative AI applications.
1. Infrastructure layer—computing chips and storage updates
Amazon Cloud Technology officially launched the cloud AI chip Amazon Trainium2 designed for generative AI and machine learning training at the conference. Based on the successful experience of the training chip Trainium, Amazon Trainium2’s basic model (FM) for trillions of parameters and large language model (LLM) design, which will be used in Amazon EC2 Trn2 instances. A single instance contains 16 Trainium chips. , while scaling up to 100,000 chips in the Amazon EC2 UltraCluster product. Amazon says using a cluster of 100,000 Trainium chips to train a large 300 billion-parameter AI model can cut training time from months to just weeks.
Amazon Cloud Technology has more than 5 years of experience in self-developed chips. Amazon Trainium and Amazon Inferentia chips provide the most cost-effective way to train models and run inference on the cloud. Amazon EC2 Inf2 instances based on Inferentia2 are designed to deliver high performance in Amazon EC2 at the lowest cost for customers' deep inference and generative AI applications. Compared with Inferentia, the throughput of Inferentia2 is increased by 4 times and is 1/10 lower than the former.
In addition, the fourth-generation self-developed server CPU chip Amazon Graviton4 was also released at the conference. Selipsky claims that Graviton4 offers 30 percent faster processing, 50 percent more cores, and 75 percent more memory bandwidth than the previous-generation Graviton processor, Graviton3 (but not the newer Graviton3E) running on Amazon EC2. Selipsky also emphasized, "We are now on the fourth generation of chips in just five years. Other cloud providers haven't even shipped their first server processors."
In addition to computing chips, Amazon Cloud Technology also announced a major update to its S3 object storage service: a new high-performance, low-latency tier S3 storage class, Amazon S3 Express One Zone, designed to provide single-digit latency-sensitive applications. , hundreds of thousands of data accesses per second at the millisecond level. Amazon S3 Express One Zone provides 10x faster data access, 50% lower request costs, and 60% lower compute costs than Amazon S3 Standard Edition.
2. Tool layer - service hosting Amazon Bedrock
"There won't be one model that will rule them all," Adam Selipsky said at today's conference. "You need to try different models and you need to choose the right model provider. I think the events of the last 10 days have made that very clear."
At this level, Selipsky focused on Amazon Bedrock, a fully managed generative AI service. Amazon Bedrock is a fully managed service that provides high-performance basic models for overseas business from many leading AI companies (including AI21 Labs, Anthropic, Cohere, Meta, Stability AI, and Amazon). Customers do not need to manage any infrastructure. You can use Amazon Bedrock to use simple API interfaces to access leading basic models such as Meta Llama2, Anthropic Claude, Titan, Stability AI, AI21 Labs, Cohere, etc. in a secure environment to build and expand their generative AI applications. Since this year, Amazon Bedrock has also undergone major upgrades, adding new functions such as Fine-tuning, Agents, Knowledge Bases, Guardrails, etc. to help customers build applications more efficiently, intelligently, and securely!
Launched two months ago, Amazon Bedrock has attracted more than 10,000 active customers from various industries to use it to quickly build and scale generative AI applications, simplifying development while ensuring privacy and security, greatly reducing the cost of From basic models to the threshold of building generative AI applications, customers can use the new GA Agents for Amazon Bedrock to create and deploy fully managed Agents in a few simple steps, and perform complex business tasks by dynamically calling APIs.
3. Application layer - AI assistant Amazon Q
Adam Selipsky said in his talk, “You can use Amazon Q to easily have conversations, generate content, and take actions.Amazon Q Get it Your systems, data repositories, and operational needs."
In this keynote speech, Amazon Cloud Technology also launched a preview version of Amazon Q . Amazon Q answers questions like "How do I build web applications using Amazon cloud technology?" Trained by Amazon’s knowledge accumulated over the past 17 years, Amazon Q can answer questions and provide corresponding explanations. Users can connect Amazon Q to the applications and software specified by their organization (such as Salesforce, Jira, Zendesk, Gmail, Amazon S3 storage instances, etc.) and customize the configuration accordingly. Amazon Q can index all associated data and content and "learn" all aspects of the current business, including organizational structure, core concepts, and product names.
Not only can it answer questions, Amazon Q can also act as an assistant to generate or summarize blog post content, press releases, and emails. It also provides a set of configurable plugins for common operations on the job, including automatically creating service tickets, working with specific teams in Slack, and updating dashboards in ServiceNow. To prevent errors, Amazon Q requires users to review its action recommendations before taking action and displays the results for verification.
Amazon Q is combined with the Amazon CodeWhisperer service to generate and interpret application code. In supported IDEs, such as Amazon Cloud Technologies' CodeCatalyst, Amazon Q can generate tests for customer code to measure its quality. Amazon Q can also create new software features, perform code transformations, and update drafts and documentation for code packages, repositories, and frameworks, using natural language to refine and execute plans.
As a generative artificial intelligence programming tool produced by Amazon Cloud Technology, Amazon CodeWhisperer has been trained and adjusted with billions of lines of code, and can generate real-time code from comments and existing code. Code suggestions ranging from code snippets to full functions can also be used to scan hard-to-find code vulnerabilities and check for potential security issues. In addition, CodeWhisperer has built-in Amazon Cloud9 and Amazon Lambda consoles, and can also be used in JupyterLab, Amazon SageMaker Studio, and Amazon Glue Studio Code by adding the CodeWhisperer extension.
4. Look up at the stars, the future is already here
From the perspective of individual developers, 2023 Amazon Cloud Technology re:Invent demonstrated the huge potential and innovation space of generative AI. Generative AI can provide us with unprecedented creativity and inspiration. The continuous development of generative AI technology will We individual developers bring richer tools and resources to promote great innovation and application development. By using generative AI models, we can generate realistic images, unique musical compositions, and stunning textual content, allowing us to ideas become reality. The products and new tools released by Amazon Cloud Technology at this re: Invent 2023 provide us with a more complete and flexible development environment and technology stack.
Click to view now2023 Amazon Cloud Technology re:Invent highlight moment, witness together a small step of Amazon Cloud Technology, a step forward in cloud computing Big step!
Finally, I will end with a quote from Adam Selipsky’s keynote speech at re:Invent: We continue to reconstruct, drive technological innovation through reconstruction, and reshape user service experience. Let us set out again to reconstruct, to turn imagination into reality, and to witness everything that will happen in the future.