AI for Business: #3 Generative AI Use-cases

Summary

The video script explores the rapid rise and diverse applications of generative AI, from writing articles and generating code to creating realistic images, videos, and music. It discusses the technology's economic impact, highlighting sectors like sales, marketing, and software engineering. The script delves into use cases like Microsoft's Office 365 co-pilot, marketing content creation, and data analysis, emphasizing the potential of generative AI in transforming business operations and everyday tasks through natural language interfaces and autonomous AI agents.

Takeaways

• 🚀 Chad GPT became the fastest-growing app, reaching 100 million users and showcasing the potential of generative AI in various fields.
• 📈 Generative AI is predicted to contribute between $2.6 to $4.4 trillion annually, impacting 63 different use cases across various sectors.
• 💡 Large language models are capable of generating not just text, but also code, images, videos, 3D models, voice, and music, expanding creative and productive possibilities.
• 🔧 Tools like GitHub Copilot and Amazon CodeWhisperer are assisting software engineers in writing and debugging code more efficiently.
• 🎨 AI tools are generating highly realistic images and videos from text inputs, with companies like Runway making strides in video generation.
• 🎼 Music creation is also being revolutionized by AI, as demonstrated by tools like Stable Audio, which can produce original music compositions.
• 📊 Generative AI is being used for knowledge retrieval, allowing employees to interact with enterprise data through natural language queries.
• 📈 The economic impact of generative AI is substantial, with sectors like sales, marketing, software engineering, customer operations, and product R&D showing significant potential.
• 🛠️ Microsoft's Office 365 co-pilot is an example of how AI can automate tasks in presentation creation, data analysis, and document writing.
• 📝 Generative AI is streamlining marketing tasks, from crafting copy to generating social media content and personalized outreach, enhancing personalization and content richness.
• 🏢 Enterprise generative AI systems are integrating with proprietary company data to create smart chatbots and internal tools that can provide the latest information and insights.

Q & A

What milestone did Chad GPT achieve in January 2023?

In January 2023, Chad GPT became the fastest-growing app of all time, reaching 100 million users just two months after its launch.

What are some capabilities of Chad GPT mentioned in the script?

Chad GPT is capable of writing entire articles, generating software code, and aiding students in learning about virtually any topic.

What economic potential does generative AI hold according to the McKenzie report?

Generative AI holds the potential to contribute between $2.6 to $4.4 trillion annually across 63 different use cases.

How are organizations leveraging large language models?

Organizations use large language models to transform their business by automating customer operations, crafting marketing content, and assisting software engineers in writing code 56% faster.

What types of data can large language models generate beyond text?

Large language models can generate highly realistic images, videos, 3D models, voice, music, and more.

Which tools assist software engineers with code generation?

Tools like GitHub Copilot, Amazon CodeWhisperer, and Meta's Code Llama assist software engineers in writing software code, writing unit tests, and finding bugs.

What are some tools mentioned for generating realistic images from text inputs?

Stable Diffusion, OpenAI's DALL-E, and Midjourney are mentioned as tools that can generate highly realistic and beautiful images from text inputs.

How is generative AI being used in the music industry?

Tools like Stable Audio from Stability AI are used to produce music, demonstrating AI's capability to generate nice music from text inputs.

What is one significant use case of generative AI in the marketing domain?

Generative AI is used to automate the creation of marketing content, such as crafting marketing copy, generating social media content, writing personalized outreach emails, and producing visual elements.

What are the two major ways of integrating generative AI with enterprise data?

The two major ways are fine-tuning, which involves refining a pre-trained model on a specific dataset, and retrieval-augmented generation (RAG), which allows models to access up-to-date and verifiable knowledge sources.

Outlines

00:00

🚀 Generative AI's Rapid Growth and Versatility

The script introduces Chad GPT, an app that achieved unprecedented growth by reaching 100 million users, showcasing the potential of generative AI in various fields. It highlights the technology's ability to produce text, code, images, videos, 3D models, voice, and music. The economic impact is underscored by a report estimating a contribution of $2.6 to $4.4 trillion annually across multiple sectors. The episode aims to explore over 30 use cases of generative AI in different domains, emphasizing its practical applications for non-AI experts and encouraging viewers to stay updated with the rapidly evolving field.

05:02

🛠️ Applications of Generative AI in Business and Creativity

This paragraph delves into the practical applications of generative AI in business, focusing on how it can automate tasks and enhance content creation. It mentions tools like GitHub Copilot and Amazon CodeWhisperer that assist in coding, and visual tools for generating images and videos from text prompts. The economic impact is illustrated through a chart from a McKenzie report, emphasizing sectors like sales, marketing, software engineering, and customer operations. The paragraph also discusses knowledge retrieval, where AI can access and retrieve enterprise data through natural language queries, exemplified by Microsoft's Office 365 co-pilot and its capabilities in presentation, spreadsheet, and document creation.

10:04

🎨 Creative AI Tools and Enterprise Generative AI Systems

The script discusses the use of generative AI in creative fields such as music and visual arts, with tools like Stable Diffusion and OpenAI's DALL-E. It also covers the application of AI in software development, user interface design, and synthetic data generation for machine learning. The paragraph transitions to enterprise-level use of generative AI, where it can be integrated with proprietary data to create smart chatbots and internal tools that can understand and respond to company-specific inquiries, either through fine-tuning or retrieval-augmented generation.

15:05

🗂️ Enterprise Data Integration and Retrieval-Augmented Generation (RAG)

This section explains how generative AI can be integrated with enterprise data to create intelligent systems that can retrieve information and generate responses based on proprietary datasets. It introduces the concept of Retrieval-Augmented Generation (RAG), which allows models to access and use the latest data to provide accurate and relevant answers. Examples include a chatbot for internal health insurance documents and an e-commerce application that provides product recommendations based on the shopping cart contents, illustrating the potential for natural language interfaces in various industries.

20:05

🌐 Natural Language Interfaces and Their Impact on Various Industries

The script highlights the rise of natural language interfaces facilitated by generative AI, which allows users to interact with software more intuitively. It provides examples from various industries, such as education with Duolingo's language learning features, e-commerce with Instacart's 'Ask Instacart', legal and contractual applications with Duckin's contract summarization, and health care with Amazon's AWS Health Scribe and Google's MedPaLM. The paragraph emphasizes the transformative potential of these interfaces in making software interactions more natural and efficient.

25:07

🤖 Autonomous AI Agents and Their Potential Applications

The final paragraph discusses the concept of autonomous AI agents, which can break down complex tasks into smaller subtasks and execute them using large language models. Examples include Meta's GPT, which can generate a comprehensive set of outputs for building a game or a website, and other agents that can order food, find houses, or interact with CRM systems like Salesforce. The paragraph concludes by emphasizing the transformative impact of generative AI across all domains and invites viewers to learn more about AI and how to apply it in their businesses.

Keywords

💡Generative AI

Generative AI refers to artificial intelligence that can generate new content, such as text, images, video, and music, from simple prompts. In the video, generative AI is highlighted as a transformative technology with applications in writing articles, generating software code, creating realistic images, and more. It marks an explosive growth in AI capabilities and economic impact.

💡Large Language Models (LLMs)

Large Language Models are AI systems trained on vast amounts of text data to understand and generate human-like language. These models, like ChatGPT, are capable of writing text, summarizing information, and assisting in various tasks. The video discusses how LLMs are used in tools like GitHub Copilot and how they enable natural language interfaces.

💡Retrieval-Augmented Generation (RAG)

RAG is a method where large language models access up-to-date, external knowledge sources to improve the quality and accuracy of generated responses. It reduces the risk of AI producing incorrect information by grounding answers in verifiable data. The video explains RAG's application in scenarios like internal enterprise data retrieval and customer interactions.

💡Economic Impact

The economic impact of generative AI refers to its potential to contribute significant value across various industries. The video cites a McKinsey report estimating that generative AI could add between $2.6 to $4.4 trillion annually across multiple sectors, including sales, marketing, software engineering, and customer operations.

💡Automation

Automation involves using technology to perform tasks without human intervention. The video highlights how generative AI automates tasks like customer operations, marketing content creation, and software development, thereby increasing efficiency and productivity.

💡Knowledge Retrieval

Knowledge retrieval in the context of generative AI involves accessing and utilizing an organization's data through natural language queries. The video discusses how employees can ask questions and retrieve information from documents, PDFs, and other data sources, transforming how businesses manage and use their data.

💡Enterprise Applications

Enterprise applications refer to the use of generative AI in business settings to enhance operations, customer service, and internal processes. The video showcases examples like Microsoft's Office 365 Copilot, which automates tasks in Word, Excel, and other applications, and internal chatbots that provide information based on company data.

💡Natural Language Interface

A natural language interface allows users to interact with software using everyday language instead of specialized commands. The video emphasizes how generative AI enables such interfaces, making it easier for users to access information and perform tasks by simply asking questions in natural language.

💡Autonomous AI Agents

Autonomous AI agents are AI systems that can perform complex tasks independently by breaking them down into smaller sub-tasks. The video mentions examples like Auto GPT and Baby AGI, which demonstrate the potential of these agents to execute complicated objectives autonomously.

💡Fine-Tuning

Fine-tuning involves refining a pre-trained large language model on a specific dataset to customize it for particular tasks or styles. The video explains that fine-tuning is resource-intensive but useful for achieving a deep understanding of domain-specific language or generating content in a specific style.

Highlights

Chad GPT became the fastest-growing app of all time, reaching 100 million users within two months of its launch in January 2023.

Generative AI can produce highly realistic images, videos, 3D models, voice, music, and more, significantly expanding its application beyond text and code.

Generative AI holds the potential to contribute between $2.6 to $4.4 trillion annually across 63 different use cases, according to a McKinsey report.

Thousands of organizations leverage large language models to transform their business operations, such as automating customer operations and assisting software engineers in writing code 56% faster.

Tools like GitHub Copilot, Amazon CodeWhisperer, and Meta's Code Llama assist software engineers in writing code, creating unit tests, and finding bugs.

Visual AI tools like Stable Diffusion, OpenAI's DALL-E, and MidJourney generate highly realistic images from text inputs.

Runway's advancements in video generation allow users to create video content from text prompts.

Stability AI's Stable Audio tool showcases AI's capability to produce music, transforming creative processes in the music industry.

Generative AI enables advanced data analysis using natural language, simplifying complex data tasks for non-experts.

In marketing, generative AI automates tasks such as crafting marketing copy, generating social media content, and writing personalized outreach emails.

Microsoft's Office 365 Copilot suite demonstrates generative AI's power in creating presentations and analyzing business data through natural language prompts.

Knowledge retrieval using generative AI allows organizations to access their data through natural language queries, enhancing productivity and information retrieval.

Enterprise generative AI systems, such as fine-tuning and retrieval-augmented generation (RAG), provide deep integration with proprietary company data for customized AI solutions.

Autonomous AI agents like Auto GPT and Baby AGI showcase the potential of AI in autonomously executing complex tasks by breaking them down into subtasks.

Meta GPT's multi-agent framework can take a single prompt and generate comprehensive outputs needed to build a solution, showcasing the future of project management and development.

Transcripts

in January 2023 only two months after

its launch Chad GPT made history by

becoming the fastest growing app of all

time reaching 100 million users capable

of writing entire articles generating

software code and eaing students in

learning about virtually any topic Chad

GPT marked the onset of an explosive

growth in generative AI the technology

powering it this surge wasn't just

confined to creating text and code it

extended to gener rating highly

realistic images videos 3D voice music

and more according to a report by

McKenzie generative AI holds the

potential to contribute somewhere

between 2.6 to $4.4 trillion annually

across 63 different use cases today

thousands of organizations leverage

large language models to transform their

business using them for tasks like

automating customer operations crafting

marketing content or even assisting

their software Engineers write code 56%

faster the possibilities seem endless

welcome to the third episode of AI for

business where we decode Ai and how to

use it in your day-to-day work for the

non AI experts in the previous episodes

we got introduced into AI machine

learning and explored more than 50

practical use cases in this episode our

focus is going to be on the rising star

generative AI we're going to explore

more than 30 30 different use cases

across different domains from Smart

knowledge retrieval and chatting with

Enterprise data to co-pilots and

autonomous agents to a lot of other

applications in health education Finance

retail marketing and more you'll get

introduced to the major patterns and

types of use cases in this growing field

this will help you understand how to

apply generative AI into your day-to-day

work if you want to understand what

generative AI is and how it

distinguishes from other types of AI I

highly recommend you revisit the first

episode of this course as it delves

deeper into this

differentiation a big disclaimer before

we start generative AI is moving so fast

the use cases you're going to hear about

today are somehow relevant as of the

time we record this video which is the

end of September 2023 but I really

encourage you to keep an eye on this

growing field because it changes

literally every single day with that

said let's Dive Right

In

[Music]

first of all I want to start with the

fact that large language models are

capable of generating stuff Beyond text

people might be familiar with the

different applications of chat GPT in

writing articles and summarizing text

and so forth but large language models

are very powerful in producing different

types of data today tools like GitHub

co-pilot Amazon code whisper and meta's

code llama are able to assist software

Engineers drastically in writing

software code writing unit tests finding

bugs in software and more in the visual

realm we have a lot of tools today that

are able to generate highly realistic

and beautiful images from text inputs

these include tools like stable

diffusion open eyes Dolly mid journey

and more companies like Runway are

making significant strides in the video

generation space where you can type

specific prompts and get video output

music cativity has not been left behind

either with tools like stable audio from

stability AI we can see how AI could be

used to produce really nice music let's

listen to some

[Music]

samples now that we have seen the

different types of data that gen could

produce let's explore what the real

economic value lies and uncovered the

significant use cases where companies

are reaping the benefits of this amazing

technology today let's start with this

chart from the same mckenzi report that

highlights specific sectors with the

most substantial economic impact

glancing at the chart you'll observe the

vertical scale representing impact in

billions while the horizontal scale

illustrates impact as a percentage of

functional spend the graph highlights

promising domains such as sales

marketing software engineering customer

operations and product R&D areas I

personally concur hold immense potential

in addition to those I'd like to add

another pattern that has been

tremendously helping companies recently

knowledge retrieval generative AI with

its ability to understand natural

language can help different

organizations access their data through

natural language employees can ask

questions and chat with their Enterprise

data whether they have documents PDFs

presentations knowledge bases product

features you name it now you can ask

those questions simply in natural

language and generative a and large

language models will be able to retrieve

that information more on this later

we're going to expand on it give a lot

of examples but I just wanted to add

this as a major pattern to the previous

patterns that we just discussed a great

example to start with with these out of

the box tools that anyone could use is

the suite of tools that Microsoft has

created for the Office 365 co-pilot

let's take an example here you're asking

it for example to create a presentation

based on some proposal document you're

attaching the document and boom it

created the presentation from scratch

wow isn't this amazing now you can keep

adjusting the presentation like for

example hey add a cost benefit analysis

and you would add this for you and then

you can go on and add some visuals for

example again through simple natural

language prompts boom now imagine how

this would transform presentation

creation right now let's take another

example with spreadsheets here for

example we have a spreadsheet with

transactional sales data you know we

have different dimensions like countries

customers products Etc uh you want to do

some analysis so we got and ask hey

Analyze This quarters business results

and summarize key trends and then it

would do that right and then you can go

on and you know hey show me a breakdown

of some sales growth and it would show

you this you can keep on adding these

questions you know adding visualizations

extracting insights adding charts and

you know these things historically used

to take a lot of time a lot of effort

you know writing different equations and

stuff here you would go to what if

analysis like what happens if this and

this happens and it would provide the

scenario and then you would build a

model and so forth now let's transition

to another example in word when you're

writing documents for example writing

first drafts writing proposals is one of

the most tedious tasks that takes a lot

of time let's see what you can do with

Office 365 in that case you can ask it

simply to write write a proposal based

on some meeting notes and a product road

map document and if we go and do that

these things by the way is so helpful in

generating first drafts that you can

iterate on don't take it you know as the

final version or something and then you

can start like iterating like asking it

or hey make it look like this style like

you can convert the style to make it

mimic something and pull images from

some other presentation and it go goes

on and does that and so forth you can

add a summary at the beginning and keep

on doing things how would that transform

writing in general and you know

Enterprise documents in specific for

those interested in data analysis but

may not be experts in coding or

sophisticated software Chad GT's

Advanced Data analysis is a

GameChanger it enables you to analyze

complicated data sets using natural

language you can upload whatever files

you want you know like images

spreadsheets text documents or whatever

data files and one queries using natural

language ask questions and generate

insights it can clean and manipulate

data with just simple instructions in

plain language and also provides data

visualization capabilities enabling you

to see your data in charts and graphs

without needing to create them manually

in marketing many companies are using

generative AI today to bring massive

amounts of automation to a wide array of

tasks including crafting marketing copy

generating engaging social media content

writing personalized Outreach emails and

even producing a diverse range of visual

elements it's not just about

streamlining the process but actually

enriching the content and bringing a

whole new level of

personalization let's explore a real

word example type face a company using

generative AI to revolutionize content

creation for marketing a classical use

case is creating nice images for

products so for example you can select a

product write a simple prompt describing

what you want to see in the image of the

product and boom you end up with a bunch

of nice photos with nice backgrounds

that are on brand and Enterprise safe

and the next step you might want to take

these images and create some marketing

campaigns with them let's see how that's

going to happen you can go on select a

template for an Instagram ad and start

filling some attributes like the goal of

the post which product are you

showcasing your target audience the tone

you're want to use and so forth and it's

going to generate the campaign for you

you have a nice image with a nice

description that you can take right away

and put on social media now imagine the

amount of automation you can bring to

social media with a technology like this

on various social platforms like

Instagram LinkedIn Facebook and more

applications in other domains include

the following when you look at the area

of like software in general you have

like code generation for accelerating

application development you have

application prototype and design to

quickly generate user interface designs

you have things like data set generation

you know generating synthetic data to

train machine learning models in the

area of audio you can think of things

like text to voice generation for

creating educational voice over sound

creation which could be used for making

Custom Sounds without copyright

violations audio editing for editing

podcasts in post without having to

re-record in the 3D world you can look

at applications like 3D object

generation which could be used for video

games digital

representation and you know creating

interior design mockups and virtual

staging for architecture design in video

there are a lot of applications like

video creation which could be used in

entertainment you know like generating

short form videos for social media for

example or training or learning and

there is also voice translation and

adjustments which could be used for

video dubbing life translation and voice

cloning for most of these use cases

there are existing tools that you can

subscribe to write a simple prompt and

start generating the output that you

want whether it's a video an image code

or whatever now I want to shift gears to

talk with you about the second level of

using generative VI which I call it the

Enterprise generative AI systems that's

where we can start seeing how generative

AI could be used in an Enterprise

setting for Enterprise use cases in a

little more sophisticated way but brings

generally speaking more value let's see

it Enterprise generative AI is mostly

about using foundational models with

your Enterprise data this data is not

usually published on the web for example

let's say you would like to have a smart

chat bot that you'd like your customers

to have intuitive conversations with you

know to ask questions about your

products or Services facilitate returns

make bookings and so forth that bot will

require to have access to your latest

data about those customers those those

products you know the purchasing history

and so forth right or if you'd like to

have an internal chatbot where your

teams could collaborate and ask

questions about company specific

information like you know again specific

aspects related to products or Services

historical proposals you have sent to a

client company proprietary data and so

forth again that bot would require to

have access to large amounts of your

company's data now in these cases you

would like these models to have a deeper

level of integration with your company's

data and that data could exist in many

unstructured forms like documents emails

spreadsheets historical slack

conversations presentations and so forth

there are two major ways for having

these models generate content based on

your proprietary data those are

fine-tuning and retrieval augmented

generation I'm not going to go through

the technical details of each right now

you can look this up on the Internet a

lot of Greater resources but I'd like to

provide a gentle introduction and some

tips on when to consider each method

fine-tuning involves refining a

pre-trained large language model on a

smaller specific data set to customize

it for Unique needs this method while

impactful is resource intensive

necessitating significant compute power

infrastructure and deep AI expertise rag

or retrieval augmented generation on the

other side enables large language models

to access the latest proprietary

knowledge without the hurdles of model

fine-tuning it enhances the quality of

responses by grounding large language

models on external upto-date and

verifiable knowledge sources thereby

reducing hallucinations or

misinformation fine tuning is usually

helpful when you'd like the model to

have a certain style or tone with the

output it generates or a specific format

like code generation for example or when

you'd like the model to have a deep

understanding for the specific words uh

used in a specific domain right ragon

the other side is pretty helpful when

you'd like the model to be able to reply

based on the latest data that you have

like latest customer data latest you

know products data services data stuff

like that so whenever you need this kind

of recency in the kind of replies the

model generat rag is the way to go there

is no such thing that you know one is

better than the other it's just about

what are you trying to achieve in the

following uh you know part of the video

I'm going to show some examples for

retrieval augmented generation

that would help showcase how it could be

used in an Enterprise setting so as

discussed before the whole idea about

frag is to enable large language models

to generate the replies based on

specific data sets that you'd like to

use to ground the answers of these

models in this case this is a demo built

by Microsoft using the chat GPT model on

Asia open Ai and what they're doing is

they want to create a chat experience

where the employees of a fictitious

company called kosu can chat with

internal health insurance documents this

is as you can see this is the health

insurance policy uh consisting of about

like 103 Pages the traditional way of

looking uh for information inside that

document would take a lot of time for

anyone to understand the specifics about

their policy and whatnot so here they

are enabling a chat experience as you

can see you can go on and start chatting

like asking for example what is included

in my North Wind Health Plus plan that

is not in standard

and boom it generates the answer as you

can see Northwind Health Plus offers

more comprehensive coverage than

Northwind standard and so forth the

thing here is that you can find the

citations these are the sources where

this answer was generated based upon you

can click on any of these citations and

see the source part or piece in the

document that actually was used to

generate this reply this brings a lot of

credibility you know one thing common

about large language models is

hallucination producing you know replies

that might not be factual so seeing

those citations and being able to track

where exactly they are in the document

actually somehow solve this problem and

then the workflow goes on uh the person

is asking a followup question does my

plan cover ey exams the model

understands that they are referring to

the plan that they've asked about before

so there is no need to you know explain

things again and then the answer as you

can see yes Northwind Health Plus offers

coverage for vision etc etc and the same

thing you have

citations imagine if you can apply the

same methodology over whatever set of

unstructured data that you have whether

again this is find in presentations

documents PDFs emails Etc rag is widely

used by many companies and organizations

today to enable easy access to

proprietary data through natural

language interface very similar to what

you see in chat GPT now let's see a

similar example but this time it's more

for consumers in the e-commerce space

let's have a look so in that case it's

an e-commerce store for again the

fictitious company Koso and someone is

asking you know um give me some

information about your hacking jackets

so it goes and retriev some information

about the hacking jackets but it takes

into consideration what's available in

the shopping cart for that customer

right now we have three products so the

reply is grounded and is related to

those three products as you can see here

and it provides some suggestions and

recommendations based on what's

available in that shopping cart that's a

quick example for how retrieval

augmented generation could work in case

of e-commerce as well by now I'm sure

you are starting to notice the pattern

here the emergence of natural language

interfaces this ability for users to

just talk with software instead of going

through graphical user interface or

different models this ability is enabled

by generative Ai and this is opening a

whole set of possibilities for consumers

businesses employees and individuals to

interact with software in an incredibly

natural way using language I want to

show you a set of examples for early

adopters who are starting to embrace

this idea of natural language interfaces

in legal Travel Health online shopping

and more let's have a look in the

education space dual lingo a famous

language learning app has used gp4 to

enable two major features one is explain

my answer which explains in details the

mistakes that you have done while using

the app for example so we can ask it

like hey what was the mistake and you

know ask it to elaborate again please

and it's going to provide details

helping you to understand why that was a

mistake the other feature is roleplay

where you can imagine different

scenarios like here for example you know

engaging with a waiter in a restaurant

you know asking for your order and stuff

like that through which you can learn

specific languages like Spanish and

French isn't that beautiful another

beautiful example that I like is what in

cart did with its new feature ask

instacart that is enabled by gp4 users

can go and start chatting with the

application for example you want

something related to lunch you can type

it and then set of questions would come

up and you can pick up a question like

what's a healthy lunch for my kids and

then the app would understand that go

and retrieve some products related to

this like healthy lunches for kids

provide some tips brings the the related

products and then you can follow up with

other questions like what are some

healthy snacks for my kids and go on and

ask questions and you know pick items

isn't this totally transforming the

shopping experience in the legal and

contractual space I like what duckin

showed before bringing generative AI to

the contract space like summarizing the

most important aspects in a contract for

example and extracting the specific

items that you're interested at another

thing is chatting with the contracts

right like asking questions like Hey

will this contract be automatically

renewed or what is the payment due date

or what is going to happen in the event

of Act of God etc etc you can ask these

questions J VI is going to go retrieve

the answer and come back instead of you

searching manually for these insights in

the health space Amazon created a

service called AWS Health scribe which

is able to transcribe voice

conversations between the health

professionals and patients and extract

meaningful insights like in that case

for example as you see here it's able to

a highlight who's The Speaker

B extract key uh items like the chief

complaint history uh the plan and so

forth which can totally automate the

idea of writing clinical reports

summarizing these reports and extracting

meaningful insights from Health

conversations Google has shown some

interesting work as well using Med Palm

2 which is a large language model

optimized and fine-tuned on Health Data

where for example you can pass

multimodal data in that case it's like

an x-ray and you know question what does

this film show and then you know the

model would process these two inputs and

come up with an answer of course these

insights and answers when it comes to

health need to be taken with a grain of

salt and you know with high care to

accuracy and stuff like that but again

it's a very nice step in the health uh

space in the travel space imagine if

Travelers can chat with an app like

Expedia speak up their mind what their

preferences are and the app would help

them plan their trips in this example

xedia is showing an example of someone

planning their honeymoon so they go on

launch the app and you know start

chatting hey I'm going to Hawaii for my

honeymoon etc etc and then start

exchanging conversation with the app

right so next step that person is going

to start you know asking some questions

is April a good month for surfing you

know getting some tips and tricks about

the duration and then starting to ask uh

for some recommendations for a couple of

romantic resorts in Maui and the app

would return this based on the data you

know this is definitely using retrial

augmented generation here as we

discussed before and then moving forward

until you book your trips wouldn't that

be fantastic if you instead of going

through a lot of comparisons you can

just share your preferences with an app

see the insights and book directly

through natural

language another interesting Paradigm

leveraging large language models is

autonomous AI agents although not why

they adop in production yet there are

increasing indications of their massive

potential and promise in the field

agents like Auto GPT and baby AGI

exemplify how complicated tasks could be

broken down into smaller chunks and

tackled step by step with the help of

large language models here is how agents

generally work once they receive the big

objective from the user prompt like

build a website or order pza or build me

that game for example or whatever they

start dividing that objective into

subtasks and each subtask will be

further divided into another subtask now

these subtasks get allocated to agents

who are employing large language models

for reasoning and to execute these

actions on their own now these agents to

achieve these subtasks they can use

tools these tools range from for example

search engines to look up something on

the Internet and come back or leveraging

or integrating with apis from other

systems you know or other things now

once the subtask is achieved it rolls

back and once the big objective is

achieved the loop ends that's in a

nutshell how these agents work you can

start seeing that this somehow showcase

the ability of these agents to

autonomously execute complicated tasks

in the future let's take an example meta

GPT imagine you have a team of software

experts who can build whatever app you

have in mind with a simple brief

description well that's more less what

met GPT doeses it's a multi-agent

framework which can take a single line

prompt a description of like hey I need

to build that game or that website or

whatever and turns it into a

comprehensive set of outputs that are

required to build that solution for

example requirement documents user

stories data structures apis code

project you know description competitive

analysis you name it so how exactly does

it do that well

again it's a multi-agent framework and

it has multiple roles inside it for

example project manager who can put

together a project plan and review the

progress and so forth a product manager

who can come up with the requirements

needed for that

product you know software Engineers who

can write code and execute that code QA

testers who can write unit tests and

execute those tests and so forth each of

these agents do it like you know they do

whatever it takes to complete that

project let's take an example of seeing

met gbt building a game called

2048 from the paper that the authors put

for that framework let's have a look so

it all starts with the human input as

you can see here the user starts with

hey make the 2048 sliding Tile game this

is the seed from which the whole project

grows and then comes the product manager

agent's role acting as the Project's

Visionary met gpt's product manager

agent takes your idea and crafts a

product requirements document detailing

what the game should do and how it will

hook

players and then passing this to the

architect agent with a plan in hand the

Artic agent steps in to design the

game's technical structure deciding on

the tools like pame in that case and how

the components fit together and then

comes the engineer agent's role next the

engineer agent rolls up its sleeves and

starts coding based on the architect's

design building the game mechanics and

user interface piece by piece finally

the QA engineer agent meticulously tests

the game ensuring everything works as

intended and the game is ready for hours

of fun another example is ordering food

on door Dash through agents this is an

app called multi-on user can types what

they're looking for like ordering a

burger from the melt in Palo Alto and it

goes on and start searching the web for

that info it finds the Mel door Dash it

goes to the page and it starts you know

clicking on the link to that page and

you know clicking the melberger item

adding it to the cart proceeding to

checkout you know and it does all these

actions on the website and then finally

it executes the order and voila another

interesting use Case by Adept AI labs

they built a model called action

transformer act1 for short and that

model can basically help you find a

suitable house house through natural

language instructions right let's see it

in action you can go and say like find

me a house in Houston that works for a

family of four my budget is 600k and

boom it goes to the website starts

searching uh you know entering the

criteria for example you know based on

the requirements the max budget would be

600k you know uh beds would be four five

plus and there you go the final

application I want to show using the

same model from adep AI is the this one

this is dealing with Salesforce you know

if you have been using CRM or like

Salesforce or others you know that there

are usually multiple steps involved when

you're you know registering leads or you

know taking memos on or notes on some

clients and stuff like that now using

natural language you can do cool stuff

like this one for example add Max n at

Adept as a new lead so it goes on opens

the needed modules start adding the

information update it save the

information and voila you know log a c

with James field saying that he's

thinking about buying 100 widgets same

thing the agent would go on open the

native modules or open the profile

update James field information add the

note and that's it to summarize

generative AI is truly transforming the

world it's absolutely going to change

every single domain moving forward you

could be part of this your company could

be part of it and that's exactly why I

created that course to help everyone

understand what the heck is going on in

the AI space and most importantly be

able to act on it if you enjoyed this

episode I appreciate if you can share it

with others who could enjoy it too and

now after we have seen so many use cases

for classical Ai and generative AI it's

time to take our first steps towards

applying this knowledge in the next

episode we're going to see how can we

start selecting the best AI projects for

your company and evaluate these projects

based on multiple criteria with that

said thank you so much for your time

today and see you on the next next

[Music]

one