AWS re:Invent 2023 Keynote by Swami Sivasubramanian

Here I have captured key highlights from ~2 hours of AWS re:Invent 2023 Keynote by Dr. Swami Sivasubramanian, AWS VP Data & ML.

New AWS services announced are preceded by (New) & bold font

Keynote by Dr. Swami Sivasubramanian, AWS VP Data & ML

Summary

Dr. Swami covered new launches in AWS’s AI/ML stack: Amazon Bedrock & Amazon SageMaker and explored how AWS data services are at the core of AI/ML. The focus of the keynote was on Generative AI. Today, there is a powerful symbiotic relationship between Data, Gen AI & Humans which is accelerating our ability to create new innovations & we are just getting started. AWS has everything you need to accelerate your Gen AI journey.

Highlights

The Analytical Engine…can follow analysis…but its province is to assist us in making available what we are already acquainted with - Ada Lovelace
More ML workloads run of AWS than any other cloud provider - HG Insights
Generative AI stack:
Amazon Bedrock
- The easiest way to build & scale gen AI apps on LLMs & other FMs
- Broad choice of models:
  - (New) Models now available in Amazon Bedrock
    - Anthropic’s Claude 2.1
    - Meta’s Llama 2 70B
- Amazon Titan Foundation Models
  - Vector Embeddings are numerical representations of text (humans understand text & machines numbers)
    - Vectors are critical for Gen AI apps: VE is helping FMs to produce more relevant response to your search query
    - Amazon Titan Text Embeddings (translates text into numerical representations)
    - (New) Amazon Titan Multimodal Embeddings
      - Search, Recommendation & Personalization
      - Accepts text, image, or combination of text-image to generate embeddings
      - Adapts to unique & proprietary data
      - Built-in mitigation to reduce biased search results
  - (New) Amazon Titan Text Lite - Summarization, copywriting & ideal for fine-tuning
  - (New) Amazon Titan Text Express - Open-ended text generation, conversational chat, RAG support
  - (New) Amazon Titan Image Generator
    - Generates Realistic & Studio-quality images
    - contains invisible watermark inside image for responsible AI
- Fine-tune additional models in Amazon Bedrock
  - using your own data & the fine-tuned model remains private & available to you only
  - Example: Create Marketing campaign for New product launch - shoes, using Llama 2:
  - What happens when your price changes or inventory changes, it’s not practical to fine tune models each time? Knowledge Bases for Amazon Bedrock launched yesterday, to address this problem:
  - Vector Databases for Amazon Bedrock
- AWS Gen AI Innovation Center - accelerating success with tools like Amazon Bedrock - training from AWS
Amazon SageMaker
- (New) Amazon SageMaker HyperPod - Reduces time to train models by up to 40%
- (New) Amazon SageMaker Innvoations - for Inference, Training & MLOps
A strong data foundation is critical to Gen AI
- A strong data foundation is
  - Comprehensive
  - Integrated
  - Governed
- A Comprehensive set of services for your strong data foundation:
- Storing Vectors & Business data speed up Models - Enabled Vector Search across AWS services:
  - Amazon Aurora PostgreSQL
  - Amazon RDS for PostgreSQL
  - Amazon OpenSearch Service
  - Amazon OpenSearch Serverless
  - (New) Amazon DocumentDB
  - (New) Amazon DynamoDB
  - (New) Amazon MemoryDB for Redis
- (New) Amazon Neptune Analytics
- (New) Amazon OpenSearch Service zero-ETL integration with Amazon S3
- Amazon DataZone - Catalog , Discover, Share & Govern Data across Organization
- AWS Clean Rooms - Create Clean Rooms in minutes & collaborate with your partners, without sharing raw data
- (New) AWS Clean Rooms ML - Apply ML models with your partners without sharing underlying data
- (New) AI-driven scaling & optimizations in Amazon Redshift Serverless
- Amazon Q - Gen AI powered assistant for your business
- (New) Amazon Q generative SQL in Amazon Redshift
- (New) Amazon Q data integration in AWS Glue - Integrate data using natural language
- Amazon CodeWhisperer
- (New) Model Evaluation on Amazon Bedrock - Evaluate, compare & select the best FM for your use-case
- Survey by World Economic Forum - Nearly 75% of companies will adopt AI by 2027
- AWS Investing in Educational Programs:
  - AWS Generative AI Scholarship - Udacity
  - AWS AI & ML Scholarship
  - 100+ AI & ML digital courses
  - (New) PartyRock - An Amazon BedRock playground