AWS is all in on Gen AI

On Tuesday, AWS kicked off its annual conference in Las Vegas and AWS CEO Adam Selipsky's keynote focused exclusively on Generative AI with his messaging targeting executives and decision makers. On Wednesday, Swami Sivasubramanian, Vice President of AWS Data, carried the AI torch even further and got more technical with announcements on how AWS is embedding AI technology in all of its services while targeting the builders.

AWS has always had the advantage of being a first mover on just about any cloud technology but with AI, this is not the case. So it was extremely important for AWS to tell the world its AI story and how they feel they are ahead of the competition. Selipsky made it his mission to portray AWS as the leader in AI in the cloud and he even took some jabs at the competitors.

Whether he was referencing a competitor's region outage in France ("imagine if you had a region supported by a single datacenter") or referencing a competitor's dependency on a single foundational model (FM) provider - "events in the last 10 days has made that very clear" referencing Open AI's board's firing of Sam Altman - Selipisky held back no punches.

But the real messaging was the vision and development of new and improved cloud services that provide platform capabilities for builders to quickly bring new AI solutions to market. Selipsky walked us through their Gen AI stack, layer by layer.

He worked his way up from the bottom (infrastructure) to the top (applications) and everything in between. He opened with announcements about performance and cost improvements of compute and storage services and shared the stage with NVDIA founder and CEO Jensen Haung to discuss their deep partnership with providing massive compute cluster capabilities to lower the cost of entry for customers looking to use Gen AI.

From there he moved up to the tool layer where he discussed a robust set of capabilities for building Gen AI foundational models with a fully managed service that provides both AWS trained models with AWS Titan and popular third party models such as Anthropic, who shared the stage as well. AWS feels that having an open model approach is a competitive advantage over their competition who are focused on their own models as a single provider (Open AI and Bard).

Amazon Bedrock is a fully managed service (serverless) that provides all of the underlying infrastructure including massive clusters on demand, integrates with all of the security and governance frameworks and services that all AWS services inherit and provides capabilities to automate functions to process tasks. It also streamlines and can even eliminate the need to build custom, complex ETL processes. In short, Bedrock takes the rocket science out of standing up secure and compliant infrastructure and functional models so customers can focus on the upper layer, the applications.

A big blocker for the adoption of AI inside of large companies is privacy and security concerns. AWS announced Guardrails for Bedrock, a service for enforcing responsible AI, implementing customer specific governance and redacting PII information. In addition, AWS announced that customer data is not used by AWS and cannot be used to train their models.

At the application layer, AWS strives to accelerate the time to market for builders who are bringing AI capabilities to their customers. Two key services in this layer is the newly announced Amazon Q, a secure and compliant AI chatbot integrated throughout the AWS service catalog, and CodeWhisperer which accelerates the development process for builders.

Many companies don't allow or greatly restrict the usage of AI chatbots. Q was built to provide a secure and compliant chatbot that can be tailored to answer questions about a company's knowledgebase, code, operational processes, training resources and many other areas of importance. If you have ever used chatGPT, you know how powerful chatbots are, but they lack context of your business and your company. Q allows companies to create that missing context so the chatbot can be tailored to provide unique access and insights into your company's data and policies.

I wrote a post a while back about how to preload chatGPT with personal coding preferences to tailor the AI bot's responses to my needs. CodeWhisperer takes this 100 times further. CodeWhisperer not only can answer questions about AWS best practices, but it can inspect your code and make architecture and code recommendations. It can also be used for troubleshooting, building test cases and much more. This is Amazon's answer to Github's co-pilot.

The following image is a nice high level summary of the AWS Gen AI Stack and its services.

Wednesday's keynote with Swami took the day 1 announcements and went much deeper to show the builders how these services actually work. One interesting announcement was that AWS has now integrated vector search throughout all of its database and AI services.

Use cases for vector search include image searches, document searches, music retrieval, product recommendations, video searches, location-based searches, fraud detection, and anomaly detection (per AWS website).

Swami also announced a new database service called Neptune. Neptune is a fully managed graph database that provides the capabilities to perform customer 360 analysis, detect fraud detection patterns, make predictions using machine learning and much more. Amazon Quicksight senior technical product manager Shannon Kalisky showed a great demo using Quicksight on top of Neptune and several other data services. In her demo they addressed the problem of delayed flights by quickly building an application that leverages public data like baggage, weather, flight traffic, etc. to quickly reroute and rebook travelers. This was all done without writing a lick of code but instead configuring several data services. To top it off, she then created a quick executive report that used templates to traverse the data and show the cost benefit of the application.

I mention this because a few years back I managed a prototype team called Cloud Garage that tackled the same the problem. Back then a lot of these services did not exist. We had a small team that knocked out something similar in two 2-week sprints. This effort could now be completed in a day or two by one person. Pretty amazing!

I won't go through all the announcements from Wednesday because there were way too many to write up but a few to highlight are:

• AWS Clean Rooms, provides a way for customers and their partners to more easily and securely collaborate and analyze their collective datasets without sharing the underlying data
• Sagemaker Hyperpod, reduces infrastructure provisioning time and costs for training foundational models.
• Amazon Q for SQL, now Q can not only do text to code but text to SQL
• Titan Image Generator - A MidJourney and Dall-e like service for creating rich images which includes an invisible watermark for identifying real versus fake images.

Summary

The entire conference including Werner's keynote today was focused on Gen AI. If you still think Gen AI is a fad, think again. Gen AI will change the way we work and Amazon's investments in this space is a sign that AWS is all in.