00:00:00
[Music]
00:00:09
chances are you've already experimented
00:00:11
with generative Ai and you've probably
00:00:13
gotten some results that have been
00:00:15
helpful and maybe some that fell short
00:00:18
throughout this course AI experts at
00:00:20
Google will teach you the difference
00:00:21
between a good prompt and a great prompt
00:00:24
so you can work faster and smarter with
00:00:26
Gen at your side and we'll share
00:00:29
practical examples of where you can use
00:00:31
gen at work hi I'm Amina I work on
00:00:35
generative AI at Google in this course
00:00:37
my colleagues and I are going to teach
00:00:38
you how you can get the most out of gen
00:00:41
you'll learn when to use gen and how by
00:00:43
designing better prompts to get the best
00:00:46
results you'll apply what you've learned
00:00:48
with Hands-On activities and quizzes to
00:00:50
level up your prompting skills after
00:00:52
you've completed this course you'll have
00:00:54
lots of practice applying gen in ways
00:00:57
that matter to you and your job as
00:01:00
recognition of your work you'll earn a
00:01:02
certificate from Google to share with
00:01:04
your network and potential employers we
00:01:06
have a lot of exciting stuff in store so
00:01:08
let's get to
00:01:09
[Music]
00:01:20
it hi I'm Timothy a director of
00:01:23
developer relations at Google for the
00:01:25
last 14 years I've been helping
00:01:27
developers and Google work better
00:01:29
together I've recently been working a
00:01:31
lot more with Gen to do things like
00:01:33
technical writing and generating code
00:01:35
I've also been helping more developers
00:01:37
integrate gen into their apps prompting
00:01:40
is a new skill that a lot of us are
00:01:42
learning and trying to get better at
00:01:44
myself included now my first experience
00:01:46
using geni that was transformative was
00:01:49
for a pretty simple task I needed to
00:01:52
quickly collect everyone's availability
00:01:54
for an important team meeting I asked
00:01:56
over chat and everyone responded in a
00:01:58
different format as people are likely to
00:02:00
do and it was a lot to track but with
00:02:02
the help of gen AI I was able to
00:02:04
organize everyone's availability into a
00:02:06
table and then transposed it so the
00:02:08
table was sorted by date not by chat
00:02:11
message a task that would have taken
00:02:13
forever manually only took me a few
00:02:15
minutes with Genai and that was my
00:02:17
breakthrough moment using Genai in my
00:02:19
everyday tasks to turn things that used
00:02:22
to be a headache into something simple
00:02:24
and easy and that's what this course is
00:02:26
about using gen to help you get your job
00:02:30
done so what is prompting anyway put
00:02:34
simply prompting is the process of
00:02:36
providing Specific Instructions to a gen
00:02:39
tool to receive new information or to
00:02:42
achieve a desired outcome on a task
00:02:45
those instructions are called prompts
00:02:47
when we write a prompt for a gen tool
00:02:50
we're giving it a series of inputs and
00:02:51
telling it what we would like it to
00:02:53
generate some gen tools can generate
00:02:56
text or images While others generate
00:02:58
video audio or even code a prompting is
00:03:02
both an art and a science to get the
00:03:04
best results we need to be precise in
00:03:07
defining what we need now this is
00:03:09
similar to the way you would help your
00:03:11
teammate get started on a new project
00:03:13
providing context and setting parameters
00:03:16
will get you the best output from gen
00:03:19
the first thing you'll learn is the
00:03:20
prompting framework it's a formula for
00:03:22
writing great prompts you'll use this
00:03:24
framework throughout the course and
00:03:27
after that it's all about putting
00:03:29
prompts to use on specific tasks that
00:03:31
can save you time in your job you'll use
00:03:34
gen to brainstorm ideas develop plans
00:03:37
and draft emails for different audiences
00:03:40
we'll teach you how to summarize meeting
00:03:41
notes assign action items and more we'll
00:03:45
also teach you how to analyze data and
00:03:47
spreadsheets with geni you'll write
00:03:49
prompts that can help you find insights
00:03:51
buried in data you'll then use gen to
00:03:54
turn those insights into visuals and
00:03:56
eventually turn it all into a slide Deck
00:03:59
with talking points for presentation
00:04:01
next you'll learn Advanced prompting
00:04:03
techniques to help you untangle complex
00:04:06
tasks for example you'll learn how to
00:04:09
create prompts that can help make
00:04:10
long-term complicated projects easier to
00:04:13
plan and execute you'll also learn how
00:04:16
to design a prompt to create your own
00:04:19
personalized AI agent to do things like
00:04:22
practice before an interview or prepare
00:04:25
for difficult work
00:04:27
conversations and finally you'll learn
00:04:29
how to use geni responsibly including
00:04:32
guidelines for using it in your job and
00:04:34
on your team this is crucial gen tools
00:04:39
help you with the work that you do but
00:04:41
they don't do it for you anyone using
00:04:44
gen should always be a valuating and
00:04:46
factchecking outputs there are a lot of
00:04:48
gen tools out there and in this course
00:04:51
we're going to demonstrate how to prompt
00:04:53
using Gemini and other Google AI tools
00:04:55
like Gemini for Google workspace and
00:04:58
Google AI Studio but all of the
00:05:00
techniques and best practices you'll
00:05:01
learn in this course can be applied to
00:05:04
other geni tools like chat GPT co-pilot
00:05:07
or CLA last thing we designed this
00:05:10
course to give you skills that you can
00:05:11
use at work right away so all of these
00:05:14
lessons and techniques you're going to
00:05:15
learn are rooted in real world scenarios
00:05:18
you should experiment and play around to
00:05:20
figure out what works best for you and
00:05:23
as you go through this course feel free
00:05:24
to pause the video and test what you
00:05:26
just learned with something you're
00:05:28
working on right now
00:05:30
now let's get started with our first
00:05:33
[Music]
00:05:43
prompts in this lesson you're going to
00:05:46
learn how to create effective prompts a
00:05:48
good prompt follows a simple framework
00:05:51
task context
00:05:54
references evaluate and iterate if you
00:05:58
ever forget a step just remember
00:06:01
thoughtfully create really excellent
00:06:05
inputs first is Task you need to
00:06:08
describe the task you want the
00:06:10
generative AI tool to help you with now
00:06:13
this should include a Persona and a
00:06:15
format preference so that the task is
00:06:18
specific Persona refers to what
00:06:21
expertise you want the Gen tool to draw
00:06:24
from you can ask the tool to take on a
00:06:26
Persona like a professional speech
00:06:28
writer or or a marketing executive with
00:06:31
15 years of experience or you can ask it
00:06:34
to create output for a specific audience
00:06:37
a customer or even your manager you can
00:06:40
be as detailed as you'd like when adding
00:06:42
a Persona to your task format refers to
00:06:46
how you want the output to appear
00:06:48
whether that's a bulleted list short
00:06:50
sentences or a table so there you have
00:06:52
it task next you'll include context or
00:06:56
the necessary details to help the Gen
00:06:58
tool understand what you need from it
00:07:01
this is the difference between writing
00:07:03
give me some ideas for a birthday
00:07:05
present under
00:07:06
$30 and give me five ideas for a
00:07:09
birthday present my budget is $30 the
00:07:12
gift is for a 29-year-old who loves
00:07:14
winter sports and has recently switched
00:07:16
from snowboarding to
00:07:18
skiing sometimes you'll add references
00:07:22
for the Gen tool to use while creating
00:07:25
its output you just asked a gen tool to
00:07:28
give you ideas for birthday present
00:07:30
right well if you add examples of
00:07:32
birthday presents you've given in the
00:07:34
past as references the Gen tool can give
00:07:37
you a more useful output there aren't
00:07:40
always going to be clear references of
00:07:41
what you need especially if you're
00:07:43
working on something more abstract or
00:07:46
searching for ideas and inspiration once
00:07:48
you have your output it's time to
00:07:52
evaluate ask yourself if the input you
00:07:54
provided gave you the output you needed
00:07:57
this leads us to the final part of the
00:07:58
framework iterate if you evaluate your
00:08:01
output and determine that you're not
00:08:03
getting what you need you can try again
00:08:05
by adding more information or tweaking
00:08:08
your prompt and this is a key part of
00:08:10
prompting effectively and we'll explore
00:08:11
it in depth later on in the course one
00:08:15
more note on the framework there are
00:08:17
plenty of ways to construct an effective
00:08:18
prompt the order of how you construct a
00:08:21
prompt is less important than the
00:08:23
substance of the prompt itself as long
00:08:25
as you're thoughtfully creating really
00:08:27
excellent inputs you're outputs should
00:08:30
be
00:08:35
[Music]
00:08:41
great let's put the framework into
00:08:45
action first we'll log into Gemini and
00:08:48
then use the tool to help us brainstorm
00:08:50
ideas for a new high performance sneaker
00:08:53
line first let's add the
00:08:56
task generate five ideas for a new high
00:09:01
performance sneaker line okay we've
00:09:04
asked Gemini to complete a task but
00:09:06
we're not really applying the prompting
00:09:08
framework yet remember thoughtfully
00:09:11
create really excellent inputs this
00:09:15
prompt is all task and nothing else
00:09:17
which might give us an output that's too
00:09:19
broad and not very useful still Gemini
00:09:22
generated five ideas with unique names
00:09:25
and descriptions this isn't a bad start
00:09:27
but we can do better let's add some more
00:09:30
details like our desired format and a
00:09:32
more specific task for the tool to
00:09:35
complete list the
00:09:38
concepts and materials for each sneaker
00:09:43
in an
00:09:45
outline that's much better now we have a
00:09:48
set of unique ideas for a sneaker line
00:09:51
that includes the materials for each
00:09:53
shoe and it came in our preferred format
00:09:57
I think we can do even better don't you
00:09:58
let's add some
00:10:00
context the sneakers should be made for
00:10:05
athletes doing cross trining
00:10:09
activities with the new information
00:10:11
Gemini created Five new sneaker ideas
00:10:14
that are more suited to our specific
00:10:16
goals remember getting tailored outputs
00:10:19
means we need to provide a Genai tool
00:10:22
with more details and context in order
00:10:25
to generate more useful results success
00:10:28
is all about the details so let's give
00:10:32
references a
00:10:33
try references give gen tools examples
00:10:37
to work from and that can mean asking a
00:10:39
gen tool to learn from the tone style or
00:10:43
length of a given reference providing
00:10:45
multiple references is also known as few
00:10:48
shot prompting shots are just references
00:10:51
or examples and the term is used a lot
00:10:55
there's also singleshot prompting which
00:10:57
means we're giving it one reference and
00:11:00
zero shot prompting which means we don't
00:11:03
give the AI tool any references now most
00:11:06
of the time between two and five
00:11:08
references is the sweet spot for a geni
00:11:11
tool too few references and we don't
00:11:13
give enough context too many we could
00:11:15
skew the results and limit creativity to
00:11:18
practice few shot prompting with our new
00:11:20
sneaker line let's include descriptions
00:11:22
of shoes that already exist one of them
00:11:24
is from a budget line of shoes and the
00:11:27
other one has a new adaptive soul
00:11:29
we can input those descriptions like
00:11:33
this keep the five ideas generated but
00:11:37
refine them using these two examples as
00:11:42
references here as we paste in the
00:11:48
references ah there's a lot of choices
00:11:51
here and they all seem like good options
00:11:54
for the task and this is cool a shoe
00:11:57
that regulates temperature
00:11:59
evaluating the output and iterating
00:12:01
might be the last parts of our prompting
00:12:03
framework but they're also where we get
00:12:05
to experiment and get creative each new
00:12:08
output is an opportunity to further
00:12:10
refine your prompt until you get the
00:12:12
response you want in fact we've been
00:12:14
evaluating and iterating this whole time
00:12:18
we evaluated the sneaker ideas from our
00:12:19
first prompt and we iterated by adding
00:12:22
context we evaluated the output again
00:12:25
and we iterated by adding references and
00:12:28
remember we can always add details or
00:12:30
tweak phrasing in order to change our
00:12:33
outputs we like to say ABI or always be
00:12:37
iterating give the prompting framework a
00:12:39
try yourself remember it's always better
00:12:41
to start simple and then slowly add
00:12:44
complexity iterating as you go if your
00:12:47
outputs start to lose quality you might
00:12:49
need to go back and make your prompts
00:12:51
simpler and that's okay learning what
00:12:53
works and what doesn't is all part of
00:12:56
the journey if you ever get stuck just
00:12:58
remember to thoughtfully create really
00:13:01
excellent inputs and you'll get back on
00:13:06
[Music]
00:13:16
track there are going to be times when
00:13:18
your prompt simply isn't giving you what
00:13:20
you want but instead of scrapping all
00:13:22
your work and starting again from zero
00:13:25
think about how you can always be
00:13:27
iterating or Abi to try and mold the
00:13:30
outputs into something more useful by
00:13:33
the end of this video you'll learn four
00:13:35
helpful iteration methods the first
00:13:37
method is to revisit the prompting
00:13:39
framework and make sure you're providing
00:13:41
enough specificity in your task context
00:13:45
and references for example if you wrote
00:13:48
give me five blog post ideas a
00:13:50
generative AI tool might respond better
00:13:53
if you adjusted your prompt to include
00:13:55
the Persona and format for example you
00:13:58
are an expert on Sports Nutrition
00:14:01
provide five blog post headlines that
00:14:03
summarize the biggest Trends happening
00:14:05
in the industry for an audience of
00:14:07
physical therapists working with
00:14:10
professional basketball players the
00:14:12
second method is to separate your prompt
00:14:14
into shorter sentences start by taking a
00:14:17
long input and breaking it down into
00:14:19
smaller tasks this is the long input
00:14:22
summarize the key data points and
00:14:24
information in this report then create
00:14:27
visual graphs from the data and shorten
00:14:28
the key information into bullets you can
00:14:31
break this up into shorter sentences and
00:14:33
input them as separate prompts you'll
00:14:36
input each prompt receive an output and
00:14:38
then follow up with a new prompt until
00:14:41
all of your tasks have been submitted
00:14:44
first summarize the key data points and
00:14:46
information in this report then follow
00:14:49
that up with create visual graphs with
00:14:52
the data you summarized and finally
00:14:56
shorten the key information you
00:14:57
summarized into bullets sometimes
00:15:00
shorter sentences can yield more precise
00:15:02
results because the Gen tool can parse
00:15:05
one small task at a time instead of
00:15:08
identifying the relationships between
00:15:09
all of them at once you can also try
00:15:12
using different phrasing or switching to
00:15:15
an analogous task which is a task that
00:15:17
is very similar to the one you're trying
00:15:19
to complete but different enough to
00:15:22
trigger a new response for example if
00:15:25
you're asking a gen tool to help write a
00:15:28
marketing plan plan for a product or
00:15:30
service you could instead ask it to
00:15:32
write a story about how this product
00:15:34
fits into the lives of our Target
00:15:36
customer demographic by moving from
00:15:39
write a marketing plan to write a story
00:15:43
you're asking the Gen tool to approach
00:15:45
the task differently which might lead
00:15:48
you closer to a useful output finally
00:15:50
introducing constraints might also help
00:15:53
focus a gen tools outputs maybe you want
00:15:56
to make a playlist for an upcoming road
00:15:58
trip and you're trying to figure out
00:15:59
what artists you want to include you've
00:16:02
added some context about your favorite
00:16:03
genre but the results are kind of boring
00:16:06
you've heard all these songs a million
00:16:08
times before to get better output and
00:16:10
something more unexpected you could
00:16:13
start adding constraints like specifying
00:16:16
you only want artists from a certain
00:16:18
region or artists that have released
00:16:20
music in The Last 5 Years adding
00:16:23
constraints to your prompt will help the
00:16:25
Gen tool narrow down its outputs and
00:16:27
give you something more helpful or
00:16:30
unique the better you can evaluate and
00:16:32
iterate the better your output will
00:16:35
[Music]
00:16:46
be images and visuals can be as
00:16:49
important as words when you want to
00:16:51
communicate ideas in this lesson I'm
00:16:54
going to teach you how to use generative
00:16:56
AI tools to create visuals
00:16:59
so far we've asked gen tools to produce
00:17:02
responses in what's called a text based
00:17:06
modality modalities are the different
00:17:09
formats in which gen tools receive or
00:17:12
produce information whether that's text
00:17:15
images video audio or code different gen
00:17:20
tools are better at working in certain
00:17:22
modalities be sure to check the Gen tool
00:17:25
you're using to find out which
00:17:27
modalities it's capable of using or
00:17:29
producing let's start with image
00:17:31
generation some gen AI tools can create
00:17:34
images a sunrise a bouquet of flowers or
00:17:38
even a crab right into a dolphin but
00:17:40
those same tools can also make images
00:17:42
for a business or a professional
00:17:46
presentation maybe you're a musician
00:17:48
playing a gig in New Orleans and you
00:17:50
want to promote your concert so you use
00:17:53
a gen tool to help you create a poster
00:17:55
to advertise the show let's prompt
00:17:58
Gemini to to create both text and images
00:18:00
so we can discuss the subtle differences
00:18:02
between prompting for each modality
00:18:05
we'll start with the text first remember
00:18:08
to keep the thoughtfully create really
00:18:11
excellent inputs framework in mind text
00:18:15
based prompts work best when we specify
00:18:17
our task and add some clear context so
00:18:20
we could prompt generate headlines for a
00:18:24
poster promoting a rock concert in New
00:18:28
Orleans
00:18:30
and to add a little more context about
00:18:32
the task we could write the concert is
00:18:35
one night only and the headlines should
00:18:39
encourage the audience not to miss
00:18:43
out by specifying our task and adding
00:18:46
context we're guiding the Gen tool to
00:18:49
the text based output we
00:18:52
want and just like that Gemini came up
00:18:56
with a few catchy headlines for the
00:18:57
poster
00:18:59
and this is a good one right here Nola
00:19:02
this is it Unforgettable Rock one night
00:19:05
only it's catchy and it gets to the
00:19:08
point now in order to prompt the Gen
00:19:11
tool for an image we'll need to tweak
00:19:13
our language we'll still use the
00:19:15
prompting framework but we'll need to
00:19:17
provide more vivid descriptions that
00:19:19
help the Gen tool determine the type of
00:19:22
image it needs to create this means
00:19:24
specifying the size color and position
00:19:27
of things in the image and the overall
00:19:31
aesthetic we want so first we'll specify
00:19:33
our task and
00:19:36
format generate an image of an electric
00:19:40
guitar for a poster it should be a
00:19:44
photographic
00:19:46
style and how about some vivid
00:19:49
descriptions the guitar should be
00:19:52
glittery or sparkly and create a sense
00:19:57
of excitement
00:20:00
the guitar should be in the
00:20:02
foreground and give a sense that it's
00:20:05
floating in the
00:20:08
sky great Gemini created four different
00:20:12
images that you can use on your
00:20:14
poster so how can we make these images
00:20:17
even better let's break down how to
00:20:19
iterate and refine a prompt for images
00:20:22
we're still going to use the prompting
00:20:23
framework but with a few little tweaks
00:20:26
for the concert poster maybe you liked
00:20:28
the appearance of the guitar but you
00:20:30
want to make it even more exciting by
00:20:32
adding a storm with lightning striking
00:20:35
the guitar we could refine it by
00:20:38
writing now make the sky
00:20:41
Stormy with lightning hitting the
00:20:46
guitar there we go you could keep this
00:20:49
image or keep evaluating and iterating
00:20:52
again and again adding relevant details
00:20:55
from each new output until you get one
00:20:57
that works
00:21:02
[Music]
00:21:11
okay we just used text to create an
00:21:13
image but we can also use an image as
00:21:16
part of our prompt to create a different
00:21:19
type of output let me introduce you to
00:21:21
multimodal prompting the essence of
00:21:24
multimodal prompting is using different
00:21:26
types of media to prompt a gener ative
00:21:28
AI tool like inputting image and text or
00:21:33
audio and text this can be especially
00:21:36
useful in the workplace you can take a
00:21:38
picture of a chart and ask a gen tool to
00:21:41
explain the data in plain language you
00:21:44
could upload different logo options for
00:21:46
your company's Rebrand as a set of
00:21:48
references and then prompt the Gen tool
00:21:51
to give you more choices based on each
00:21:53
direction or you could capture audio of
00:21:56
another language and ask for a transcrip
00:21:58
destion in the language you understand
00:22:01
here's an example where we'll prompt
00:22:02
with both image and text to receive a
00:22:05
text based output from Gemini let's
00:22:07
imagine you're an entrepreneur who needs
00:22:09
help creating social media captions for
00:22:12
a new design of nail art you're selling
00:22:15
you can take a picture of your nail art
00:22:17
and ask for help writing a caption
00:22:20
here's a photo of the nail art and we'll
00:22:22
input this into Gemini and prompt write
00:22:25
a social media post featuring this this
00:22:29
image the post should be fun short and
00:22:34
focus on the fact it's a collection of
00:22:37
new designs I'm selling note that in
00:22:41
addition to including a reference photo
00:22:43
of the nail art we still used the other
00:22:46
elements of our prompting framework we
00:22:48
specified our task added some context
00:22:52
and included the format besides the
00:22:55
image itself we didn't provide other
00:22:58
reference ref es but if we have a
00:23:00
specific tone or voice we want the Gen
00:23:03
tool to match we could always input a
00:23:06
few captions from previous posts to
00:23:12
reference this is great Gemini analyzed
00:23:16
the image and created a fun caption you
00:23:18
can use to market the nail art notice
00:23:20
how it uses emojis to break up the text
00:23:23
and how it engages followers by asking a
00:23:26
question about their favorite design the
00:23:28
cool thing about multimodal prompting is
00:23:30
that it reflects the way you experience
00:23:32
the world you don't just discuss the
00:23:35
words or images in a work presentation
00:23:37
you build connections between them to
00:23:40
get a fuller understanding of the topic
00:23:42
in question a mix of text images and
00:23:46
other modalities can open up new ways of
00:23:49
solving problems or saving time you
00:23:51
could use a gen tool to turn a picture
00:23:53
of a city map into a list of notable
00:23:56
landmarks find key insights within an
00:23:59
audio file or quickly extract a list of
00:24:03
room names from an office floor plan
00:24:06
here's another example you go to a
00:24:08
conference and receive a schedule of
00:24:10
events and you want your team to focus
00:24:12
on a few of the events in particular I
00:24:15
want to send a reminder to my colleagues
00:24:19
about certain events from a conference
00:24:23
schedule extract the times of the
00:24:25
keynote speaker and two panel
00:24:29
discussions from this schedule into a
00:24:33
table again we specified our task
00:24:36
provided helpful context and included
00:24:38
the format before inputting the picture
00:24:40
of the schedule let's check it
00:24:45
out great the table makes it really easy
00:24:49
to see where your team needs to go and
00:24:51
when you can even take it a step further
00:24:53
and prompt Gemini to draft an email
00:24:56
about these events we'll get into
00:24:57
prompting for email drafts later in the
00:24:59
course just remember to keep the
00:25:01
prompting framework in mind no matter
00:25:03
what modality you are prompting in to
00:25:06
achieve the best results how might you
00:25:08
leverage different modalities in your
00:25:10
prompts to help you at
00:25:12
[Music]
00:25:20
work generative AI tools are powerful
00:25:24
but like any tool it's important you use
00:25:26
them responsibly especially at work
00:25:29
first consider the problem you're using
00:25:30
gen to help you solve does it align with
00:25:33
your goals and your obligations to your
00:25:35
clients and co-workers what about your
00:25:37
organization's policies and local laws
00:25:40
about using gen to perform this type of
00:25:43
task if it doesn't align then you should
00:25:45
rethink your process and whether or not
00:25:47
a gen tool is right for the job second
00:25:51
consult your company's rules or policies
00:25:53
before entering confidential or
00:25:55
sensitive data into gen tools you can
00:25:57
also check if your company has an
00:25:59
Enterprise version of a gen tool that is
00:26:02
okay for other types of use and remember
00:26:05
if you're using geni tools for personal
00:26:07
use avoid entering personal or
00:26:10
confidential information about yourself
00:26:12
into publicly available tools and always
00:26:15
check how the data you enter might be
00:26:18
used finally being a responsible gen
00:26:21
user means evaluating outputs for
00:26:23
potential bias and errors and disclosing
00:26:26
any use of gen when sharing content with
00:26:29
others while it's okay to enlist the
00:26:32
help of geni you'll still need to
00:26:34
evaluate the outputs for accuracy the
00:26:36
way you would for any output the same
00:26:39
goes for hallucinations which is when a
00:26:41
gen tool provides outputs that are
00:26:44
inconsistent incorrect or even
00:26:47
nonsensical hallucinations most often
00:26:50
happen when someone gives a gen tool
00:26:52
vague or unclear instructions or when a
00:26:55
tool guesses at an answer to something
00:26:57
it didn't quite understand
00:27:00
hallucinations can be hard to recognize
00:27:03
that's why it's so crucial to fact check
00:27:05
and cross reference outputs to confirm
00:27:08
if a fact or statement in an output is
00:27:10
true remember gen tools aren't thinking
00:27:13
critically the way humans can it's
00:27:16
important to keep what we call a human
00:27:18
in the loop approach meaning a human
00:27:21
should verify gen outputs before using
00:27:24
them I recently generated an image for a
00:27:27
presentation
00:27:28
I wanted to have a bunch of cats on a
00:27:31
rocket going to the moon now instead the
00:27:34
output was a little bit off the cats
00:27:36
were on top of the rocket rather than
00:27:39
inside it and that's not exactly safe
00:27:41
for cats is it while I did write in my
00:27:44
prompt that cats needed to be on a
00:27:46
rocket I didn't mean that literally but
00:27:49
the tool didn't know that so I iterated
00:27:52
and specified that the cats should
00:27:54
appear safe and sound inside the rocket
00:27:57
instead of on top of it some gen tools
00:27:59
such as Gemini have a built-in fact
00:28:01
Checker that allows you to cross
00:28:03
reference the outputs using Google
00:28:05
search comparing outputs side by side
00:28:08
makes it easier to determine how
00:28:11
accurate your initial output is and to
00:28:14
find any discrepancies so how can you
00:28:16
avoid these issues before they become a
00:28:19
problem try to recognize biases and
00:28:21
outputs and the negative consequences
00:28:23
they can have they may appear as
00:28:25
stereotypes or unfair represent
00:28:28
presentations of a group of people
00:28:30
avoiding biased to negative outputs
00:28:32
starts with inputting specific detailed
00:28:34
prompts and iterating as needed another
00:28:37
key part of this is using language that
00:28:40
includes people of all backgrounds
00:28:42
genders and ethnicities and avoid
00:28:45
stereotypes and generalizations in your
00:28:47
inputs for example if you were using a
00:28:50
gen tool to help you write the
00:28:52
description for a job posting you should
00:28:54
avoid the gendered terms like serviceman
00:28:57
or Workman instead use service person or
00:29:02
worker so the tool doesn't write a
00:29:04
description that only speaks to someone
00:29:06
who identifies as male remember gen
00:29:09
tools are only tools they don't think
00:29:11
critically and can't understand Nuance
00:29:13
the way humans can it's your job to
00:29:16
bring that human perspective every time
00:29:18
you use a gen tool
00:29:21
[Music]
