re:Invent 2023 AI/ML Session Summaries
Get up to speed on all the GenAI, AI, and ML focused 300 and 400 level sessions from re:Invent 2023!
Adam Selipsky delivered a keynote brimming with thrilling and groundbreaking announcements at reInvent 2023! Explore the latest services and updates set to significantly scale your business and technology!
Get up to speed on all the GenAI, AI, and ML focused 300 and 400 level sessions from re:Invent 2023!
Get up to speed on all the storage focused 300 and 400 level sessions from re:Invent 2023!
Get up to speed on all the serverless focused 300 and 400 level sessions from re:Invent 2023!
Brian is an AWS Community Hero, Alexa Champion, runs the Boston AWS User Group, has ten US patents and a bunch of certifications. He's also part of the New Voices mentorship program where Heros teach traditionally underrepresented engineers how to give presentations. He is a private pilot, a rescue scuba diver and got his Masters in Cognitive Psychology working with bottlenosed dolphins.
View Brian's articlesAdam Selipsky presented the second Keynote of the week, providing an overview of where AWS is in the cloud landscape, especially compared to some of the other cloud providers. He made a number of service announcements and provided context for the services.
S3 Express One Zone speeds up storage retrieval, which can save money by reducing computing runtime. It can deliver up to 10x better performance than the S3 Standard storage class while handling hundreds of thousands of requests per second with consistent single-digit millisecond latency.
Combined with Graviton 4 and Trainium 2, compute is also faster. Graviton 4 is available now in the R8G instance type, with many more instance types coming online “soon”. Graviton4 has 50% more cores, and 75% more memory bandwidth while increasing overall compute performance by 30%. Short-term capacity blocks lets you reserve scarce GPU resources without overspending. Graviton 4, combined with Elastic Fabric Adaptor, allows building super compute clusters with performance on the ExaFlop scale.
Trainium2 is designed to deliver up to 4x faster training than first-generation Trainium chips and will be able to be deployed in EC2 UltraClusters of up to 100,000 chips, making it possible to train foundation models (FMs) and large language models (LLMs) in a fraction of the time, while improving energy efficiency by up to 2x
AWS and NVIDIA also announced an expansion of their strategic collaboration to deliver the most advanced infrastructure, software, and services to power customers’ generative artificial intelligence (AI) innovations.
This continued partnership ensures that AWS will continue to have the broadest and deepest set of compute offerings across Intel, AMD, Nvidia, Qualcomm, and their own Annapurna ARM based silicon.
Customers can do more for less time and less money. This opens the possibility of tackling problems that might have been out of reach previously. Express One Zone files live in a single Availability Zone, which you can select. This lets you co-locate your data with your compute resources for even faster response times. You can easily transfer files from “normal” S3 buckets to Express Zone buckets. Graviton 4 continues the trend of higher performance at lower energy cost. This lets you scale while reducing the carbon footprint of applications as per the new Sustainability Pillar of the Well Architected Framework.
Bedrock provides choice in GenAI by by allowing customers to write against a common interface allowing easy movement between models. Events of the last two weeks have shown how important this flexibility is. In these early days of LLMs there will be winners, losers, and new models practically every week. Combined with Agents and Knowledge Bases, it becomes easier for customers to develop their own GenAI models. In addition to Bedrock's existing industry-leading compliance regimes (GDPR, HIPPA, SOC), AWS announced Guardrails for Bedrock which lets customers restrict the data that is returned.
Dario Amodei, CEO of Anthropic, an AI safety and research company in which Amazon is investing up to $4 billion, explained to the audience how the large language models (LLMs) supporting his company's AI assistant, Claude, have been trained on AWS infrastructure.
GenAI can be scary, and these announcements provide assurance that AWS will make it safer. GuardRails for Bedrock lets a company specify information that can and can not be returned by the LLM. This enhances your security posture and helps prevent the exfiltration of sensitive information. One of the new models is Claude 2.1, which has been shown to have significantly fewer hallucinations. Agents allow you to build applications that integrate LLMs with actions. You can create an agent that calls a Lambda function to perform actions like sending an email, creating a Slack message or processing a Jira ticket. This expands the LLM application from just creating a text response to performing an action based on the intelligence of the model.
Amazon Q brings a GenAI experience to many aspects of the AWS ecosystem, allowing for faster development and better BI. When a system gets an error, Q can help troubleshoot with access to underlying AWS resources. In a similar vein, DataZone now has a GenAI interface to create data descriptions and suggest ways to use the data. This greatly reduces the effort required to use the data in DataZone. Another simplification is the addition of Zero-ETL to RDS PostgreSQL, RDS MySQL, DynamoDB, and OpenSearch.
Q brings the power of GenAI to data analysis. Significantly, it can make use of data from both AWS as well as customer-supplied data. And it does this in a safe and secure manner. Q has a Feature Development Tool that helps you organize the multiple steps involved in adding a feature to an application. It creates a plan for you, including ticketing, code development, and testing. It currently integrates with CodeCatalyst, and IDE integration is coming soon. Q also integrates with QuickSight and AWS Connect.
Zero-ETL is a concept that AWS started introducing last year, and as expected, they have expanded it to four new data sources. Reducing the effort involved in using your company’s data makes that data more actionable and cost-effective.
AWS is years and generations ahead of the other public cloud providers. AWS is on generation 3, 4, or 5 of resources that other providers have yet to create a first generation. They continue to emphasize cost reduction and security while lowering the bar for development, thus increasing companies' ROI for new and existing products.
We expected an emphasis on GenAI, and there certainly was quite a bit of emphasis there, but there were many other areas where AWS created significant advancements.
AWS showed the interactions between compute and storage enhancements. Better and faster storage options have implications for the compute side systems. While all of the services are getting faster and (generally) less expensive, they continue to get enhanced security. AWS continued to enhance Zero-ETL by adding four new data sources. This is another example of simplifying the process of making data actionable.
GenAI, of course, pervaded a wide range of services, ranging from the obvious cases to specialized cases like using GenAI to create CloudWatch Log queries. Q is a large new service or actually a set of services. The ability to ask natural language questions about AWS documentation is significant because, similar to CodeWhisperer, this allows a developer to stay focused on their IDE or console. Avoiding a costly context switch to performing a Google or StackOverflow query can greatly encourage developer “flow” and, thus productivity.
Q also has a code migration capability. AWS talked about a small team migrating 1000 Java applications to a new version of Java in just two days. This is remarkable, and avoiding version lag is a Good Thing. Coming soon Q will gain the ability to automatically migrate Windows code to Linux code. This will enable more applications to avoid the high cost of Windows licenses.
AWS also talked about public Give Back via Project Kuiper, which provides high-speed satellite networking in underserved areas, as well as free Cloud Training to democratize the cloud.
AWS is, of course, a business, but they continue to put the customer first. Faster everything, more secure everything, leveraging their data with your data. As they add new services, they continue to expand and enhance the existing ones. They have found a way to create win-win scenarios. The continued evolution of Graviton is a great example of this. Their innovation in silicon is a win for customers as it lets them work faster. They save money, and because of the extra usage this generates, AWS makes more money. And since Gravition is energy efficient, the Earth wins.
This dedication to winning together creates an environment where AWS and customers can learn from each other. This creates the kind of synergy that moves us all forward and makes working with AWS a very safe choice.