www.washingtonpost.com Open in urlscan Pro
23.204.2.78  Public Scan

Form analysis
0 forms found in the DOM

Text Content

Accessibility statementSkip to main content

Democracy Dies in Darkness
SubscribeSign in
This content is paid for by an advertiser and published by WP Creative Group.
The Washington Post newsroom was not involved in the creation of this content.
Content from AWS
Illustration of an evergreen tree
Illustration of an evergreen tree
Illustration of an evergreen tree


Welcome
Illustration of an evergreen tree
Illustration of an evergreen tree
G
E
N
E
R
A
T
I
V
E
A
I
E
X
P
L
O
R
E
R
’
S
G
U
I
D
E

E
S
S
E
N
T
I
A
L
L
E
A
R
N
I
N
G
S
F
O
R
B
U
S
I
N
E
S
S
A
N
D
T
E
C
H
L
E
A
D
E
R
S
N
A
V
I
G
A
T
I
N
G
T
H
E
A
I
W
I
L
D
E
R
N
E
S
S

Share on FacebookShare on TwitterShare on LinkedIn Send e-mail Copy urlUrl
copied
explore

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration

Click the book to start your explorationTap the book to start your exploration
Scroll to explore
Subtitle Settings
Font Default Mono Sans Mono Serif Sans Serif Comic Fancy Small Caps
Font Size Default X-Small Small Medium Large X-Large XX-Large
Font Edge Default Outline Dark Outline Light Outline Dark Bold Outline Light
Bold Shadow Dark Shadow Light Shadow Dark Bold Shadow Light Bold
Font Color Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
Background Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
AWS_2024_Guide_Welcome video
1:27




THIS VIDEO IS BEING VIEWED IN ANOTHER WINDOW


0:00 / 1:28
Settings



Scroll to explore

Welcome to the Generative AI Explorer’s Guide. Check out the introductory video
above, then keep scrolling to learn more about where to start your AI journey.





From streamlining logistics to boosting customer service, generative AI promises
to
take business operations to the next level. Indeed, the technology is expected
to
increase U.S. productivity growth by up to 1.5 percent annually over the next
decade1
as more companies use it to drive innovation.





But for leaders who must navigate this future, the path to success with large
language
models (LLMs) can be full of twists, turns and forks in the trail. That’s where
this
Generative AI Explorer’s Guide comes in. A compendium of short videos, snackable
insights and quick case studies, it’s filled with actionable information that
will help
guide you toward generative AI success.





The AI Explorer’s Guide is divided into three sections that mark major waypoints
along
the path:



 * MOVE:Getting started

 * BUILD:Piloting solution

 * SCALE:Going into production

At the end, you’ll find a summary of key learnings and a look at what’s ahead
for
generative AI.





1 Goldman Sachs



L
E
S
S
O
N
1
:

M
O
V
E



Turn the page to continue



Move


Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring
Scroll to explore
Subtitle Settings
Font Default Mono Sans Mono Serif Sans Serif Comic Fancy Small Caps
Font Size Default X-Small Small Medium Large X-Large XX-Large
Font Edge Default Outline Dark Outline Light Outline Dark Bold Outline Light
Bold Shadow Dark Shadow Light Shadow Dark Bold Shadow Light Bold
Font Color Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
Background Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
AWS_2024_Guide_Move video
1:59




THIS VIDEO IS BEING VIEWED IN ANOTHER WINDOW


0:00 / 2:00
Settings



Scroll to explore
> Y
> o
> u
> c
> a
> n
> ’
> t
> a
> f
> f
> o
> r
> d
> n
> o
> t
> t
> o
> d
> o
> a
> n
> y
> t
> h
> i
> n
> g
> ,
> y
> o
> u
> h
> a
> v
> e
> t
> o
> g
> e
> t
> o
> n
> t
> h
> i
> s
> j
> o
> u
> r
> n
> e
> y
> .

Francessca Vasquez,
VP of Professional Services and Generative AI Innovation Center, AWS

The time to go from thinking about generative AI to implementing it is now.
Early movers will enjoy sustained advantages over competitors. Conversely,
organizations that sit on
the sidelines will fall behind. “You can’t afford not to do anything,” says
Francessca Vasquez, VP of
Professional Services and Generative AI Innovation Center, AWS. “You have to get
on this journey.”



Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more

 * What generative AI can do
   
   Generative AI is widely known for text-based applications like chatbots and
   writing assistants. But the latest foundation models are multimodal. That
   means that your inputs and outputs can include:
   
    * Images: Text-to-image models are becoming more creative, artistic and
      versatile.
    * Audio: Text-to-speech applications can lend a voice to customer
      assistants, create audio tracks and support accessibility.
    * Video: Text-to-video can create promotional spots, educational and
      training materials and even short films.
    * Computer code: Text-to-code output can modernize legacy code and make
      coding recommendations to code faster and improve reliability and
      security.

 * Where to begin
   
   Leaders who are implementing generative AI should start by getting buy-in
   from across the organization – from line-of-business heads to stakeholders in
   specialized functions. Important areas to include are:
   
    * Legal and compliance
    * Engineering
    * Security
    * End-user teams
    * Data and analytics teams
   
   This will help ensure the effort is properly resourced and broadly supported.
   “You need sponsorship, whether it’s from the board or senior leadership.
   That’s where we’ve seen acceleration for companies that are extracting value
   from generative AI and moving to production,” says Vasquez.

 * Get the data
   
   Data is what drives generative AI, and a company’s own information is its
   most valuable asset in this era. Organizations must proactively assess and
   prepare their data to ensure it is AI-ready, taking the necessary steps to
   cleanse, curate and structure their data for optimal use by generative AI
   applications.
   
   Valuable data sources companies can tap for generative AI applications
   include:
   
    * Internal data: Customer records, transaction logs, products for
      personalized content, customer service response and recommendations
    * Public datasets: Open data from governments and researchers for augmenting
      and customizing LLMs
    * User-generated content: Social media, reviews, online communities for
      sentiment analysis and recommendation systems
    * Sensor data: Internet of Things (IoT) devices, machines, infrastructure
      sensors for predictive maintenance and process optimization
   
   Quick tip: You can use an LLM to analyze and evaluate the quality of your
   data and recommend remedial actions if there’s a problem, such as
   inconsistencies.

 * Where to use generative AI
   
   A company’s business priorities should determine its use cases for generative
   AI. “Are you trying to drive more revenue, improve the customer experience or
   boost retention?” asks Shaown Nandi, Director of Technology, AWS.
   
   Areas where generative AI can be most impactful include:
   
    * Process optimization: Fraud/anomaly detection, demand forecasting,
      supply-chain improvements
    * Employee productivity: Code writing, employee assistance, report
      generation, task automation
    * Customer experience: Virtual assistants and customer support, personalized
      experiences in fields such as healthcare and finance
    * Creativity and content creation: Automated creation of sales collateral,
      marketing content, personalized emails

 * How to select a pilot project
   
   With leaders on board and priorities identified, it’s time to put generative
   AI into action. Experts suggest beginning with a pilot project that:
   
    * Is low risk
    * Can quickly drive business value
    * Will deliver a positive return on investment
    * Will garner broad organizational support
   
   Resources permitting, there’s no need to stick to just one pilot. Experiment
   with various possibilities. “I haven’t come across a customer that is looking
   at a single pilot,” says Vasquez. “They have a good number of experiments
   that are controlled but are happening up and down their organization.”

G
E
N
A
I
I
N
A
C
T
I
O
N


M
I
T
S
U
B
I
S
H
I
M
A
N
U
F
A
C
T
U
R
E
S
S
P
E
E
D
A
N
D
E
F
F
I
C
I
E
N
C
Y

Mitsubishi Electric is one of the world’s leading electronics manufacturers. As
it
navigates the complexities of the manufacturing landscape – marked by
shorter release cycles and increasing software demands – the company is
dedicated to empowering its engineers to concentrate on high-value tasks. By
using generative AI to automate repetitive processes, Mitsubishi Electric aims
to significantly enhance the quality and speed of software development.





In collaboration with AWS, Mitsubishi Electric initiated a generative AI
discovery workshop to identify and validate impactful use cases that address
real business challenges. The workshop identified an opportunity to streamline
requests from the manufacturing department to revise code that is embedded
in a device’s hardware, known as firmware code.





This solution, powered by Amazon Bedrock, will save software developers time
by automating the search of internal documentation. The company estimates
that this approach could reduce its internal team’s workload by 20 to 40
percent, as it expands across its 18 production sites and continues to gather
user feedback.



Interactive insight
T
H
E
R
I
G
H
T
T
O
O
L
S
F
O
R
B
E
T
T
E
R
B
A
N
K
I
N
G

Organizations should get started with generative AI by identifying business
goals that
the technology can support. Let’s explore how this applies to a specific
industry like
financial services.



Click each to explore
 * Illustration of a clipboard, pen and small coin purse
   
   Goal:
   
   Automate loan processing
   
   
    * Document verification and analysis: Generative AI tools verify
      documentation and evaluate completeness of loan application.
    * Approve/reject decisioning: Generative AI tools extract and summarize data
      from structured, unstructured and public sources.
    * Enhanced due diligence: Generative AI tools help make initial credit
      assessments, creating personalized requests for further information when
      required.

 * Illustration of a human wearing a headset and microphone
   
   Goal:
   
   Improve call center experience
   
   
    * Call routing: Generative AI tools interpret customer queries and provide
      resolution options.
    * Agent referral and investigation: Generative AI tools suggest next steps
      to the agent using customer history. Agent can easily ask questions of
      knowledge management assets to decrease time to resolution.

 * Iillustration of sliders that can be adjusted, representing the ability to
   create personalized services
   
   Goal:
   
   Offer personalized recommendations
   
   
    * Segmented customer data and content generation: Generative AI tools
      generate content relevant to the customer segment.
    * Process automation: Generative AI tools help reduce low value manual
      processes and leverage a “human-in-the-loop”.
    * Personalized messaging: Generative AI tools optimize channel selection
      based on customer behaviors.

L
E
S
S
O
N
2
:

B
U
I
L
D



Turn the page to continue



Build


Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring
Scroll to explore
Subtitle Settings
Font Default Mono Sans Mono Serif Sans Serif Comic Fancy Small Caps
Font Size Default X-Small Small Medium Large X-Large XX-Large
Font Edge Default Outline Dark Outline Light Outline Dark Bold Outline Light
Bold Shadow Dark Shadow Light Shadow Dark Bold Shadow Light Bold
Font Color Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
Background Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
AWS_2024_Guide_Build video
1:40




THIS VIDEO IS BEING VIEWED IN ANOTHER WINDOW


0:00 / 1:41
Settings



Scroll to explore
> Y
> o
> u
> d
> o
> n
> ’
> t
> d
> o
> g
> e
> n
> A
> I
> u
> n
> l
> e
> s
> s
> y
> o
> u
> ’
> v
> e
> g
> o
> t
> a
> r
> o
> b
> u
> s
> t
> s
> e
> c
> u
> r
> i
> t
> y
> a
> n
> d
> g
> o
> v
> e
> r
> n
> a
> n
> c
> e
> s
> t
> r
> u
> c
> t
> u
> r
> e
> .

Francessca Vasquez,
VP of Professional Services and Generative AI Innovation Center, AWS

It’s time to build a generative AI application and the right decisions will help
you stay on the path. Choosing a generative AI tool like Amazon Bedrock, which
provides access to the latest LLMs, can help to future-proof your solution as
generative AI evolves. “Models are changing rapidly,” says Nandi. “You don’t
want to get stuck with an old model.”

What follows are some key considerations for building generative AI
applications.

Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more

 * How to choose the right LLM
   
   This decision should be driven by use case. Not all LLMs are created equal,
   as some excel more than others at specific tasks. “There are going to be
   tradeoffs,” says Nandi. Consider that some models are optimized for:
   
    * Customization and reasoning: Ideal for driving customer-facing
      applications like chatbots.
    * High-speed summarization and categorization: Well-suited for
      information-intensive industries.
    * Multimodal models: May be the best choice for companies that engage with
      customers across multiple platforms, including voice, video and chat.
    * Industry-specific models: These are pre-trained on domain knowledge for
      use in areas such as healthcare, legal and finance.
   
   Quick tip: Make your data model agnostic. That is, don’t code your solution
   to a specific foundation model. You’ll want to bake in some future-proofing
   as models evolve. A managed service like Amazon Bedrock offers a range of
   models and is continually updated.

 * Take control of your data
   
   High-quality data is essential to generative AI success. To drive desired
   outcomes, your data must be timely, complete and accurate. It also must be
   well-governed, unique and validated. Here are some steps that can help:
   
    * Get data into the cloud: From a cloud service like Amazon S3 Glacier, data
      can be quickly accessed and used without having to provision local
      infrastructure.
    * Curate diverse datasets: Ensure datasets have a broad range of
      demographics and experiences to counter harmful biases.
    * Continually refresh your data: Provide ongoing access to new data so LLMs’
      responses stay relevant and reduce the chance of erroneous output, also
      known as hallucinations.
    * Use the right tools: Amazon Bedrock and Amazon SageMaker can be used to
      cleanse, process, transform and analyze data to ensure it is AI ready.

 * Customize your LLM
   
   Customizing publicly available LLMs with more narrowly focused internal data
   helps ensure that the output is consistent with company goals and brand
   identity. For example, a personalized customer service app could draw data
   from prior contacts and customer knowledge bases. “Leveraging your own data
   leads to the best outcome,” says Nandi.

 * Building trustworthy AI
   
   As you build generative AI, earning the trust of your employees and customers
   is paramount. To responsibly innovate, you must work to mitigate risks and
   take steps to control hallucinations, introduce guardrails and prioritize
   education.
   
   For generative AI to be trustworthy, it must be:
   
    * Secure: Data that drives AI models can contain sensitive personal and
      financial information, so it must be safeguarded to the same standards as
      other enterprise data. At the same time, generative AI applications must
      be protected against emerging threats like prompt injections.
    * Responsible: To build responsibly, key dimensions must be considered.
      These include fairness, explainability, controllability, safety, privacy,
      security, governance, transparency, veracity and robustness.
    * Accurate: LLMs are improving in factual accuracy, but hallucinations
      remain a major challenge for organizations deploying generative AI. Tools
      like Amazon Bedrock Guardrails help detect and mitigate these
      hallucinations.

G
E
N
A
I
I
N
A
C
T
I
O
N


G
O
D
A
D
D
Y
L
I
G
H
T
S
U
P
C
U
S
T
O
M
E
R
C
A
R
E

GoDaddy hosts millions of websites, from blogs to sophisticated e-commerce
portals. Looking to better understand the factors driving the most common
customer care interactions and resolve them more quickly, GoDaddy’s Care &
Services team worked with AWS’s GenAI Labs team to build Lighthouse.





Lighthouse is a generative AI-powered system that uses LLMs managed
through Amazon Bedrock to generate insights informed by customer
support interactions.





GoDaddy product experts draw upon a pre-built prompt library to input
natural language queries into Lighthouse to mine information from transcripts
of customer interactions. This data is aggregated and visualized in dashboards
and other tools. This enables GoDaddy analysts to quickly spot recurring
problems and develop repeatable solutions.





GoDaddy customer care experts can also craft one-time prompts that reveal
insights for highly specific customer scenarios.





Armed with insights from Lighthouse, GoDaddy can:





Identify the most common customer care issues
Develop website and product experience improvements
Create more efficient customer call routing systems
Improve and personalize customer experiences


Interactive insight
T
A
I
L
O
R
E
D
T
O
P
E
R
F
E
C
T
I
O
N

There are several methods to build trusted and differentiated generative AI
experiences using
your organization’s data. See how each approach works, and the amount of data
required.



Click each to explore

 * PROMPT ENGINEERING
   
   Data needed: Low to medium
   | Customization level: Low
   | Accuracy: Low to medium
   
   
   
   Data needed: Low to medium
   
   Prompt engineering can work with limited data by leveraging the existing
   knowledge of the pre-trained model. However, having some task-specific or
   domain-specific data can help create more effective prompts.
   
   Customization level: Low
   
   While prompts can be highly customized and tailored to the specific tasks,
   the output generated relies on the knowledge of the foundation model itself.
   
   Accuracy: Low to medium
   
   Well-crafted prompts can significantly improve the accuracy and relevance of
   the model’s outputs for the targeted task or domain, but the underlying
   model’s capabilities also play a role.

 * RAG
   
   Data needed: Medium to high
   | Customization level: Medium
   | Accuracy: High
   
   
   
   Data needed: Medium to high
   
   RAG requires a vast amount of data that is curated as well as access to
   external data sources to generate results. More data helps models build
   comprehensive knowledge and generate coherent, contextually appropriate
   responses.
   
   Customization level: Medium
   
   RAG is relatively easy to set up and is one of the easier ways to get started
   with generative AI. The level of complexity increases as the application
   scales. The data needs to be high-quality and diverse for accurate and
   relevant outputs.
   
   Accuracy: High
   
   RAG can significantly improve response accuracy, especially for providing
   relevant information. Use it to generate up-to-date responses and content
   that can cite trusted sources.

 * FINE TUNING
   
   Data needed: Medium to high
   | Customization level: Medium
   | Accuracy: High
   
   
   
   Data needed: Medium to high
   
   Fine-tuning requires a moderate amount of task-specific or domain-specific
   data to effectively specialize the pre-trained model.
   
   Customization level: Medium
   
   Fine-tuning allows for customization by specializing the foundation model to
   a specific task or domain. The level of customization is limited by the
   foundation model’s existing knowledge or training data.
   
   Accuracy: High
   
   Fine-tuning on relevant data can significantly improve the accuracy and
   performance of the model for the targeted task or domain, as the model is
   explicitly trained on that data.

L
E
S
S
O
N
3
:

S
C
A
L
E



Turn the page to continue



Scale


Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring
Scroll to explore
Subtitle Settings
Font Default Mono Sans Mono Serif Sans Serif Comic Fancy Small Caps
Font Size Default X-Small Small Medium Large X-Large XX-Large
Font Edge Default Outline Dark Outline Light Outline Dark Bold Outline Light
Bold Shadow Dark Shadow Light Shadow Dark Bold Shadow Light Bold
Font Color Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
Background Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
AWS_2024_Guide_Scale video
1:47




THIS VIDEO IS BEING VIEWED IN ANOTHER WINDOW


0:00 / 1:48
Settings



Scroll to explore
> G
> e
> n
> e
> r
> a
> t
> i
> v
> e
> A
> I
> i
> s
> a
> c
> t
> u
> a
> l
> l
> y
> m
> a
> k
> i
> n
> g
> w
> o
> r
> k
> e
> r
> s
> m
> o
> r
> e
> p
> r
> o
> d
> u
> c
> t
> i
> v
> e
> ,
> m
> o
> r
> e
> e
> f
> f
> e
> c
> t
> i
> v
> e
> a
> n
> d
> g
> i
> v
> i
> n
> g
> t
> h
> e
> m
> b
> a
> c
> k
> t
> i
> m
> e
> t
> o
> d
> o
> t
> h
> e
> w
> o
> r
> k
> t
> h
> a
> t
> m
> a
> t
> t
> e
> r
> s
> .

Shaown Nandi,
Director of Technology, AWS

Once your project has been built and successfully piloted, it’s time to roll it
out at scale. That could
mean supporting thousands or even millions of users, so it’s important to get
things right. Using a
managed service like Amazon Bedrock to ensure data integrity and establish
guardrails allows
engineers to focus on business outcomes as the solution grows. “It’s a massive
accelerant for scaling,”
says Nandi.



Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more
Click the to learn moreTap the to learn more

 * When to move from pilot to production
   
   Knowing when it’s time to scale is key. Moving too soon could mean rolling
   out a solution that’s not ready for prime time. Waiting too long could mean
   losing ground to competitors. Organizations with experience working in the
   cloud can apply the same fundamentals for scaling generative AI.
   
   Here are some key considerations on when to scale:
   
    * Your data foundation is comprehensive and integrated.
    * Guardrails must be put in place to help ensure that data is private and
      secure.
    * The solution is delivering proven return on investment.
    * All stakeholders are aligned on the application’s purpose, goals and
      value.
    * Appropriate governance and compliance rules are established.
   
   Organizations can also establish metrics that must be hit before moving into
   production. For example, achieving latency of about 1 second with 90 percent
   accuracy.

 * Understanding costs
   
   Leaders must understand all of the costs associated with their generative AI
   initiatives, and how scaling those efforts will impact the bottom line. Some
   costs, like infrastructure, are mostly fixed. Others that are tied to model
   usage, potentially including fine tuning and RAG optimizations and the length
   and quantity of prompts, will grow as usage grows.
   
   “That transition from training to inference-based capacity plays a big role
   in the cost element,” says Vasquez.

 * Evaluating ROI
   
   Organizations should take a broad view when calculating return on investment
   on a generative AI project running at scale. The calculation should go beyond
   traditional metrics like internal rate of return. Factors that can be
   included are:
   
    * Reduced time to value of new initiatives
    * Productivity gains across the organization
    * Ease of modernization, for example automatically rewriting code to go from
      mainframe to cloud native
    * Increased customer satisfaction

 * Democratizing AI
   
   Scaling should include making generative AI-powered applications widely
   available throughout an organization. The goal is to empower as many workers
   as possible – giving everyone the power to innovate. Generative AI tools like
   Amazon Q, a generative AI assistant, can help both developers and business
   users accelerate software development and leverage companies’ internal data.
   
   Democratizing AI in this way can raise workers’ comfort level with the
   technology. “A lot of teams have had a bit of a fear factor around gen AI,”
   says Nandi. “‘Is this going to take my job, is this going to replace me?’ In
   most use cases we see, it’s actually making workers more productive, more
   effective and giving them back time to do the work that matters.”
   
   Democratizing AI may require some organizations to shift their mindset around
   access to technology. “It means that you want to be able to have your
   developers, your business unit users, all your personas actually doing things
   with gen AI.” says Vasquez.
   
   Quick tip: To encourage widespread, internal use of generative AI,
   organizations should apply the following:
   
    * Educate: Upskill everyone
    * Empower: Encourage experimentation
    * Evangelize: Highlight the wins

R
E
V
I
E
W



Turn the page to continue



Review


Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring

Click the book to continue ExploringTap the book to continue Exploring
Scroll to explore
Subtitle Settings
Font Default Mono Sans Mono Serif Sans Serif Comic Fancy Small Caps
Font Size Default X-Small Small Medium Large X-Large XX-Large
Font Edge Default Outline Dark Outline Light Outline Dark Bold Outline Light
Bold Shadow Dark Shadow Light Shadow Dark Bold Shadow Light Bold
Font Color Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
Background Default Black Silver Gray White Maroon Red Purple Fuchsia Green Lime
Olive Yellow Navy Blue Teal Aqua Orange Default 100% 75% 50% 25% 0%
AWS_2024_Guide_review video
0:57




THIS VIDEO IS BEING VIEWED IN ANOTHER WINDOW


0:00 / 0:58
Settings



Scroll to explore


THE PATH FORWARD

Congratulations, you’ve completed this phase of your generative AI journey. But
the adventure is just beginning. Here are some things to keep in mind as you
move forward:

 * Generative AI is quickly evolving and gaining new capabilities. There will be
   continuous improvement of models, and new approaches to implementation. One
   of the most important trends is agents. These are interconnected AI services
   that work in concert to perform complex tasks, such as end-to-end workflow
   automation for HR or accounting.
 * Multimodal models will become more prevalent as use cases grow.
 * Standards and regulations will continue to evolve, so organizations should
   keep a close eye on AI policy.

Illustration of an evergreen tree
Illustration of an evergreen tree

S
U
M
M
A
R
Y
O
F
K
E
Y
L
E
A
R
N
I
N
G
S

Click to explore

 * Move
   
   This decision should be driven by use case. Not all LLMs are created equal,
   as some excel more than others at specific tasks. “There are going to be
   tradeoffs,” says Nandi. Consider that some models are optimized for:
   
    * Start by identifying a business priority and work backward from there.
    * Select a pilot project that is low-risk, positively impactful and will
      have broad appeal.
    * Capture as much internal data as possible to feed current and future
      generative AI projects.
    * Obtain buy-in from the C-suite, as well as heads of key departments.

 * Build
   
    * Let the use case guide your choice of large language model.
    * Data that drives LLMs should be continually refreshed and efficiently
      managed with tools like Amazon Bedrock or Amazon SageMaker.
    * Don’t code everything to a specific LLM. Make your data model agnostic for
      maximum flexibility.
    * Use methods like RAG, fine tuning and prompt engineering to customize
      output.

 * Scale
   
    * Use predetermined metrics, such as latency and accuracy, to inform the
      decision on when to move from pilot to production.
    * Don’t consider scaling until your generative AI application is built
      securely and responsibly to earn trust.
    * Understand the true costs of scaling your model. Variable costs may rise
      as more users are added.
    * When calculating ROI, take a broad view that encompasses factors like
      customer satisfaction and productivity gains.

AWS has the tools and expertise to help you stay on the path to generative AI
success.
Learn more

Share this article:

Share on FacebookShare on TwitterShare on LinkedIn Send e-mail Copy urlUrl
copied
washingtonpost.com © 1996-2024 The Washington Post



 * 
 * Welcome
 * Move
 * Build
 * Scale
 * Review