The document discusses generative AI and tools for building with it on AWS. It provides an introduction to generative AI, describes common use cases like text generation and image generation, and reviews tools available on AWS for generative AI like Amazon Bedrock, Amazon EC2 Trn1n and Amazon EC2 Inf2, Amazon CodeWhisperer, and Amazon SageMaker Jumpstart. It also discusses security, customization, and cost benefits of using AWS for generative AI projects.

1. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
<@your social handle>

2. Generative AI on AWS
Suresh Poopandi
Sr. Solutions Architect
Amazon Web Services
Alak Eswaradass
Sr. Solutions Architect
Amazon Web Services

3. Agenda
Introduction to Generative AI: The what and the why
Tools for building with generative AI on AWS
How can you get started today?
3

4. 4
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Machine learning (ML) is at an inflection point
Key drivers: Compute capacity increase | Data growth | Model sophistication

5. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 5
BLOOM
Question: What is generative artificial intelligence (AI)?
• Creates new content and ideas, including conversations,
stories, images, videos, and music
• Powered by large models that are pretrained on vast corpora
of data and commonly referred to as foundation models (FMs)
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.

6. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Rise of foundation models
increase in size of model
as measured by number
of parameters
2019–2022
2019 2021 2022
1B
2B
3B
4B
5B
6B
YEAR
PARAMETERS
330M
175B
540B

7. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Where does generative AI fit?
7
Artificial intelligence (AI)
Any technique that allows computers to mimic human intelligence using logic,
if-then statements, and machine learning
Machine learning (ML)
A subset of AI that uses machines to search for patterns in data to build logic
models automatically
Deep learning (DL)
A subset of ML composed of deeply multi-layered neural networks that
perform tasks like speech and image recognition
Generative AI
Powered by large models that are pretrained on vast corpora of
data and commonly referred to as foundation models (FMs)

8. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Common use cases
Text
generation
Q&A Paraphrase
rephrase
Code
generation
Image
generation
Image
classification
Audio
generation
Video
generation
8
Text
extraction
Search
Text
summarization

9. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Generative AI application examples
Technology hardware
Chip design, robotics
Media and entertainment
Video game generation, upscaling content, face
synthesis, film preservation and coloring
Healthcare
Protein folding, drug development, personalized
medicine, improved medical imaging
Communications
Chatbot, question answering, search
Automotive
Autonomous vehicles, design parts for fuel
efficiency
Energy and utilities
Design renewable energy sources optimized for
geo, predictive maintenance
Financial services
Risk management, fraud detection
Consumer goods
Optimize pricing and inventory, correctly flag
product brand and category
9

10. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Text examples
T R A N S C R I P T S U M M A R I Z A T I O N , T E X T S U M M A R I Z A T I O N , T E X T R E P H R A S I N G
10
Cohere’s transcript summarization API AI21 Labs’ Wordtune Read AI21 Labs’ Wordtune

11. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Image examples
I M A G E G E N E R A T I O N , T R A N S F O R M A T I O N , U P S C A L I N G
11
Image transformation
4x
Upscaling
Generated by Stable Diffusion 2.0

12. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Multimodality example
P A T T E R N G E N E R A T I O N
12
LG AI Research’s Tilda, the AI artist powered by EXAONE

13. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Why foundation models?
13
Text generation
Summarization
Information
extraction
Q&A
Chatbot
Pretrain Adapt
Tasks
Unlabeled
data
FM
Text generation
Summarization
Information
extraction
Q&A
Chatbot
Train Deploy
Tasks
ML
models
…
…
…
…
Labeled
data
…
…
…
…

14. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
How foundation models work
14
Unlabeled
data
Text generation
Summarization
Information extraction
Q&A
Chatbot
FM Broad range of
general tasks
Pretrain Adapt

15. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Types of foundation models
15
Text-to-text
Generate text from simple natural-language
prompts for various applications
Text-to-embeddings
Generate numerical representation of text
for applications like search and finding
similarities between documents
“Summarize the articles on impact
of walking on heart health”
“a photo of an astronaut
riding a horse on mars”
Multimodal
Generate and edit images from
natural-language prompts
“hand soap”
Numerical representation of
“Hand soap refills
Hand soap dispenser
Hand soap antibacterial”
“Ten thousand steps per day
is optimum for maintaining
a healthy heart”
Input Output
FM

16. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Security, data control, cost,
and customization concerns
Substantial undifferentiated
work needed to operationalize
Expensive, time
consuming, and requires
deep expertise
Challenges with foundation models
16
Proprietary
models
Building your own
model from scratch
Publicly
available models
16

17. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 17
Prompt Engineering
Anatomy of a Prompt*
* Image courtesy: Cohere.ai

18. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Retrieval Augmented Generation (RAG)
18
RAG combines the power of pre-trained LLMs
with information retrieval - enabling more
accurate and context-aware responses
§ Two step process:
1. Retrieve relevant information from a
knowledge base using a retriever
2. Generate a response based on
retrieved information and input query
using a generator
§ Dynamic Knowledge Integration
o RAG allows models to access and
integrate external knowledge on-the-
fly, enhancing their ability to provide
precise answers.

19. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 19
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?

20. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 20
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?
Choose from a wide
selection of FMs built by
AI21 Labs, Anthropic,
Stability AI, and Amazon to
find the right model for your
use case.

21. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 21
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?
Customize FMs for your
business with just a few
labeled examples. None of
your data is used to train the
underlying models. Since all
data is encrypted and does
not leave your virtual private
cloud (VPC), you can trust
that your data will remain
private and confidential.

22. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 22
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?
Get the best price
performance for generative
AI with infrastructure
powered by AWS-designed
ML chips and NVIDIA GPUs.
Cost-effectively scale
infrastructure to train and
run FMs containing hundreds
of billions of parameters.

23. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 23
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?
Quickly integrate and deploy
FMs into your applications
and workloads running on
AWS using familiar controls
and integrations with the
depth and breadth of AWS
capabilities and services like
Amazon SageMaker and
Amazon S3.

24. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 24
Flexibility
The most cost-
effective
infrastructure
Secure
customization
The easiest way
to build with FMs
Generative AI-
powered solutions
Why AWS for
generative AI?
With generative AI built in,
services such as Amazon
CodeWhisperer, an AI coding
companion, can help you
improve productivity. In
addition, you can deploy
common generative AI use
cases like call summarization
and question answering using
AWS sample solutions that
combine AWS AI services with
leading FMs.

25. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
AWS Offers the broadest Generative AI Capabilities
25
AWS Generative AI
Amazon Bedrock
(Fully managed 1P & 3P FMs as API Service)
SageMaker Foundation Model Hub
(Self Managed Open Source & Custom FMs )
Amazon EC2 Trn1n
and Amazon EC2 Inf 2
Amazon CodeWhisperer

26. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
The AWS AI/ML stack
ML FRAMEWORKS
& INFRASTRUCTURE
Managed experience for building, training, hosting and monitoring ML suitable
for Data Scientist and ML practitioners
AMAZON
SAGEMAKER
Fully managed, serverless, API level access to AI suitable for LoB buyer
organizations and app developers
AI
SERVICES
26
Broadest and most complete set of services for all personas through all layers of the stack
Self managed experience for ML experts who want to build their models
Amazon CodeWhisperer
SageMaker JumpStart
AWS
Trainium
AWS
Inferentia
Amazon
Bedrock

27. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 27
Amazon Bedrock
The easiest way to build and
scale generative AI
applications with FMs
NEW

28. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon Bedrock key benefits
Enhance your data
protection using
comprehensive AWS
security capabilities
Accelerate
development of
generative AI
applications using
FMs through an API,
without managing
infrastructure
Choose FMs from
AI21 Labs, Anthropic,
Stability AI, and
Amazon to find the
right FM for your use
case
Privately customize
FMs using your
organization’s data
Use AWS tools and
capabilities that you
are familiar with to
deploy scalable,
reliable, and secure
generative AI
applications
28

29. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Bedrock supports a wide range of foundation models
29
Titan Text Titan
Embeddings
FMs from Amazon FMs from AI21 Labs, Anthropic, and Stability AI
Jurassic-2 Claude Stable
Diffusion

30. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Titan Text
focused on
NLP tasks
Titan Embeddings
for enterprise tasks
such as search and
personalization
Amazon Titan
I N N O V A T E R E S P O N S I B L Y W I T H H I G H -
P E R F O R M I N G F M s F R O M A M A Z O N
30
Benefits
• Built with 20+ years of Amazon
ML experience
• Automate language tasks such as
summarization and text generation with
Amazon Titan Text FM
• Enhance search accuracy and improve
personalized recommendations with
Amazon Titan Embeddings FM
• Support responsible use of AI by reducing
inappropriate or harmful content
30

31. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Foundation models from top AI startups
Jurassic-2 Claude
31
Stable Diffusion
31
Multilingual LLMs for text
generation in Spanish, French,
German, Portuguese, Italian, and
Dutch
LLM for conversations, question
answering, and workflow
automation based on research
into training honest and
responsible AI systems
Generation of unique, realistic,
high-quality images, art, logos,
and designs

32. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Privately customize foundation models using your
organization’s data
32
Small number of labeled examples
Maximizing accuracy for specific tasks
Fine-tune
PURPOSE
DATA NEED

33. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 33
Amazon EC2 Trn1n and
Amazon EC2 Inf2
Purpose-built ML accelerators for
the best price performance for
training and inference in the cloud
NEW

34. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Purpose-built accelerators for generative AI
AWS Inferentia
Lowest cost per inference
in the cloud for running
deep learning (DL) models
Up to 70% lower
cost per inference
than comparable
Amazon EC2 instances
AWS Trainium
The most cost-efficient, high-
performance training of
LLMs and diffusion models
Up to 50% savings
on training costs
over comparable
Amazon EC2 instances
AWS Inferentia2
High performance at the
lowest cost per inference for
LLMs and diffusion models
Up to 40% better
price performance
than comparable
Amazon EC2 instances
34

35. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved. 35
Amazon CodeWhisperer
Build applications faster and
more securely with an AI coding
companion
NEW

36. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Amazon CodeWhisperer: Now generally available and
free to use for individual developers!
36
During preview Amazon ran a productivity challenge, and participants who used
Amazon CodeWhisperer were 27% more likely to complete tasks successfully and did
so an average of 57% faster than those who did not use CodeWhisperer.
Generate code
suggestions in real time
Scan code for hard-to-
find vulnerabilities
Flag code that resembles
open-source training
data or filter by default

37. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
CodeWhisperer
B U I L D A P P L I C A T I O N S Q U I C K L Y W I T H Y O U R A I C O D I N G C O M P A N I O N
37
Code generation
AND
Go, Rust, PHP, Ruby, Kotlin, C, C++,
Shell scripting, SQL, and Scala
AND
CLion, GoLand, WebStorm, Rider,
PhpStorm, RubyMine, and DataGrip

38. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
© 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Experiment with popular
foundation models
through Amazon
SageMaker Jumpstart
38

39. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Publicly available
Foundation models available on SageMaker JumpStart
for self-managed access
39
Models
Cohere
generate-med
Tasks
Text generation
Information
extraction
Question
answering
Summarization
Models
Text2Image
Upscaling
Tasks
Generate
photo-realistic
images from
text input
Improve quality
of generated
images
Features
Fine-tuning on
SD 2.1 model
Models
Flan T-5 models
(8 variants)
DistilGPT2, GPT2
Bloom models
(3 variants)
Tasks
Machine
translation
Question
answering
Summarization
Annotation
Data generation
Models
AlexaTM
20B
Tasks
Machine
translation
Question
answering
Summarization
Annotation
Data generation
Models
Jurassic-1
Grande 17B
Tasks
Text generation
Long-form
generation
Summarization
Paraphrasing
Chat
Information
extraction
Question
answering
Classification
Models
Lyra-Fr
10B
Tasks
Text generation
Keyword
extraction
Information
extraction
Question
answering
Summarization
Sentiment
analysis
Classification
Proprietary models

40. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
3 ways to use foundation models with SageMaker
JumpStart
40
SageMaker Studio
One-step deploy
SageMaker
Notebooks
AWS Management
Console Preview

41. © 2023, Amazon Web Services, Inc. or its affiliates. All rights reserved.
Foundation models with SageMaker JumpStart:
How it works
41
Amazon SageMaker
JumpStart
Access and try out public and
proprietary FMs, and customize
and integrate them into your
generative AI applications
Deploy
Deploy the model
and run inference
for your generative
AI use case
Browse
Browse public
and proprietary FMs
Experiment
Experiment with FMs
before choosing a
model for deployment
Customize
Customize selected
FM with your own
dataset without
training from scratch

42. Start your generative AI journey today
Check out the
generative AI
webpage
Watch Werner
Vogels, AWS VP
and CTO, explain
generative AI
Read
announcement
blog post
42
Check out Image-to-
Speech generative AI
application blog post

43. Thank you!
Suresh Poopandi
ssurpo@amazon.com
LinkedIn https://www.linkedin.com/in/suresh-poopandi/
Alak Eswaradass
alaknan@amazon.com
LinkedIn https://www.linkedin.com/in/alaknantha/