I watched the entire Keynote of the AWS Summit New York City 2024 and focused my attention on generative AI and the features released (and to be released) for AWS.
You can watch the entire keynote here, but it's over 1 hour and 30 minutes long, so if you want a summary of what they spoke about, continue reading below.
AWS Generative AI Infrastructure
"AWS launched 326 generative AI features into general availability since 2023." "Generative AI will be woven into all of our applications." "The foundation for AI is in the cloud."
AWS Infrastructure includes:
Latest Generation NVIDIA GPUs
Cluster Scaleout (EC2 UltraClusters)
Compute Capacity (EC2 Capacity Blocks)
Filesystems (Amazon FSx for Lustre)
Networking (EFA)
Security & Virtualization (Nitro)
DSKs (Neuron)
Custom Accelerator (Chip) for Training (Trainium2)
Custom Accelerator (Chip) for Inference (Inferentia2)
Amazon and NVIDIA partnered to create an AI Supercomputer for NVIDIA Internal R&D called Project Ceiba, which has 414 exaflops of compute power running across 20,736 GB200s linked together using EFA networking.
Amazon SageMaker
Amazon SageMaker is used for Machine Learning & AI Models
Build
Train
Tune
Deploy
SageMaker Studio is an IDE for your machine learning workflows and data. New in SageMaker Studio is Amazon Q, which will allow you to build ML models using natural language.
Amazon Bedrock
Offers a wide selection of models to use for AI including models from A121labs, Amazon, Anthrop\c, Cohere, Meta, Mistral AI_ and Stability AI.
You can also import your own model, combine it with any other model, tune them and import them into Amazon Bedrock.
Model Customization with Fine-Tuning
Customized for specific needs
Customized for task performance
Customized for domains and formats
Each model that Amazon Bedrock offers can be fine tuned to the specific needs listed above.
You can "Fine-Tune" Anthropic's Claude 3 (supposedly only in AWS) with your private company data and keep it private.
Model Evaluation Features
Model evaluations lets you quickly evaluate and compare model's abilities to perform a task.
Building multi-modal and multi-model applications is the "super power" of generative AI.
Retrieval-Augmented Generation
Specialize and personalize models that are confidential with private data, fresh data or real-time data.
Knowledge Bases for Amazon Bedrock allows you to fully manage RAG (retrieval augmented generation). It's fully managed support for end-to-end RAG workflow.
Securely connect Foundation Models (FMs) and agents to data sources.
Retrieve relevant data and augmented prompts
Provide source attribution
They introduced expanded data connectors for Knowledge Base including
Salesforce
Confluence
SharePoint
Custom Web Source
CSV and PDF Data
Guardrails for Amazon Bedrock implements safeguards customized to your application requirements and responsible AI policies via:
Word filters
Topic filters
Harmful content filters
PII (Personal Identifiable Information) filters
Security
Prompt Injection
Contextual Grounding Checks are guardrails to detect and block hallucinations and this new feature was introduced today for general availability.
The Guardrails API allows you to safeguard applications built using foundation models available outside of Amazon Bedrock.
Agents for Amazon Bedrock executes multistep tasks across company systems and data sources.
Automatic Prompt Creation
Native RAG Support
Orchestrate & Execute Multi-step Tasks
Traces and Explainability
Memory retention for Agents (a new feature announced today) allows agents for Amazon Bedrock to remember and learn from instructions over time.
Code interpretation for Agents (another new feature announced today) allows agents for Amazon Bedrock to generate and execute code, analyze data and generate graphs.
Amazon Q
Generative AI powered assistant for software development and leveraging internal data.
Amazon Q for Developers has code suggestions, security scanning, and agents for code assistant, feature development and code transformation.
Amazon Q Developer Customization, which is generally available as of today, allows for the customized suggestions tailored to developer's needs based on internal code and best practices. The chat is aware of your internal code base and how it works.
Generate with personalized coding styles
Understand private documentation
Mastery of internal software packages
Amazon Q for Business delivers quick, accurate and relevant answers to your business questions, securely and privately. It connects with over 40 popular data sources including S3, Salesforce, Google Drive, Microsoft 365, ServiceNow, Gmail, Slack, Atlassian, and Zendesk. It respects existing access controls and will only return information you're authorized to see based on your role.
Amazon Q Apps are a form of personalize software that allows you to easily and quickly build simple tools to use for yourself or for your team that are lightweight and streamline repetitive tasks that you have to do every single day. You did this using natural language (via a chat interface). You can then publish the app in the Amazon Q Apps library.
AWS App Studio
The fastest and easiest way to build applications using AI (a text prompt).
It allows you to:
Build feature-rich apps in minutes, not days.
Deploy scalable apps without operational overhead.
Track usage patterns, control user and data access and set guardrails to maintain compliance.
Free to build.
It's focused on people with technical abilities, but not software abilities, like IT managers, data engineers and enterprise architects to build apps that solve business problems quickly.
That's it for now. Lots of new cool AI powered features announce for AWS!
Exciting times ahead!
Comments