[00:00]
I don't usually overhype myself but in
[00:02]
this video we're going over the only
[00:04]
formula you will ever need to master
[00:06]
prompting on ChatGPT and Google Bard
[00:08]
so let's get started hey friends welcome
[00:11]
back to channel if you're new here my
[00:12]
name is Jeff I work full-time in Tech
[00:14]
and if you're anything like me a couple
[00:15]
months ago you know prompting is an
[00:18]
important skill to learn but you're not
[00:19]
exactly sure why some prompts generate
[00:22]
outputs that are super generic While
[00:24]
others give you precisely what you're
[00:26]
looking for since then I've spent
[00:27]
hundreds of hours taking prompt
[00:29]
engineering courses and applying what
[00:31]
I've learned in my daily life and so in
[00:33]
this video I'm sharing the six building
[00:34]
blocks that make up a good prompt so
[00:36]
that you can use this formula to
[00:38]
consistently generate high quality
[00:41]
outputs first it's critical to not only
[00:43]
know what the six components are task
[00:45]
context exemplars Persona format and
[00:47]
tone but also know that there's an order
[00:50]
of importance to these six components to
[00:53]
show you what I mean let's use this
[00:54]
simple example I'm a 70kg male give me a
[00:57]
three-month training program the first
[00:59]
part is context followed by the task the
[01:01]
reason why the task is higher up in the
[01:03]
form of the hierarchy is if we just
[01:05]
input the task without the context
[01:06]
there's still some sort of meaningful
[01:08]
output but if we just give ChatGPT the
[01:11]
context nothing really happens put
[01:13]
another way it's mandatory to have a
[01:15]
task in your prompt it's important to
[01:18]
include relevant context and exemplars
[01:20]
and it's nice to have Persona format and
[01:23]
tone when you think of writing your
[01:25]
prompt go down this mental checklist so
[01:28]
this formula will act as a constant
[01:29]
reminder for you to include just enough
[01:31]
relevant information when writing
[01:33]
prompts and as you'll see in this next part
[01:35]
you do not need all six components in
[01:38]
every prompt to have a good output now
[01:40]
let's break down each building block
[01:42]
with specific examples starting with the
[01:44]
task the rule of thumb is to always
[01:46]
start the task sentence with an action
[01:48]
verb generate give write analyze Etc and
[01:51]
clearly articulate what your end goal is
[01:54]
it could be one simple task like
[01:56]
generating a three-month training
[01:57]
program or a complex three-step ask like
[02:00]
analyzing hundreds of user feedback
[02:02]
sharing the top three takeaways and
[02:05]
categorizing the feedback based on the
[02:07]
team responsible for following up the
[02:09]
second component context is the
[02:10]
trickiest to get right because
[02:11]
technically there's an infinite amount
[02:13]
of information you can give so I found
[02:16]
asking myself these three questions to
[02:18]
be super helpful in coming up with just
[02:20]
enough information to get a good result
[02:22]
from ChatGPT first what's the user's
[02:24]
background second what does success look
[02:27]
like and third what environment are they
[02:29]
in Back to the workout example we now
[02:31]
have I'm a 70kg male looking to put on
[02:34]
five kilograms of muscle mass over the
[02:35]
next three months I only have time to go
[02:37]
to the gym twice a week and for one hour
[02:39]
each session give me a three-month
[02:41]
training program to follow could I have
[02:43]
added more background information of
[02:45]
course only prioritize the muscle groups
[02:47]
that make me look good on Instagram
[02:49]
but the key to staying productive with
[02:51]
ChatGPT and Bard is giving just enough
[02:53]
information to constrain the endless
[02:55]
possibilities by the way although this
[02:57]
video is not sponsored it is supported
[02:59]
by those of you who subscribe to my paid
[03:01]
productivity newsletter on Google
[03:02]
workspace tips Link in the description
[03:04]
to learn more moving over to the
[03:06]
exemplars component it's just a fancy
[03:08]
way of saying examples basically all the
[03:10]
research on large language models LLMs
[03:13]
have shown that including examples
[03:15]
within the prompt drastically improves
[03:18]
the quality of the output starting with
[03:20]
a simple example this is a poorly written
[03:22]
bullet point from a resume we can now
[03:25]
ask ChatGPT to rewrite this bullet
[03:27]
point using this structure I accomplished
[03:29]
X by the measure y that resulted in Z
[03:31]
which is actually best practice by the
[03:33]
way so actually do this in your resume
[03:34]
for example I lowered Hospital mortality
[03:37]
rate by 10% by educating nurses in new
[03:39]
protocols which translates to 200 lives
[03:41]
saved per year here's a slightly more
[03:43]
complicated example for interview prep
[03:45]
based on my own resume write me an
[03:47]
answer to the interview question what's
[03:49]
your biggest weakness use the star
[03:51]
answer framework situation task action
[03:53]
and results here instead of using a
[03:56]
full-blown interview answer as an
[03:57]
example which would be overkill the star
[04:00]
framework acts as an example structure
[04:02]
for ChatGPT to follow last example
[04:04]
let's say you need to write a job
[04:05]
description you give some context around
[04:07]
the opening and ask ChatGPT to
[04:09]
reference an existing job description if
[04:12]
I use this one I found on LinkedIn the
[04:13]
output will follow the same formatting
[04:15]
and use the same professional HR-y
[04:17]
language saving me a bunch of time main
[04:20]
takeaway here exemplars are not
[04:22]
necessary for every prompt but including
[04:25]
a relevant example or framework will
[04:27]
greatly improve the quality of your
[04:29]
output moving along the Persona
[04:31]
component is basically who you want
[04:33]
ChatGPT and Bard to be and the pro tip
[04:36]
here is to think of someone you wish you
[04:38]
had instant access to with the task
[04:40]
you're facing if you enjoyed yourself
[04:42]
working out that person might be a
[04:43]
physical therapist with experience
[04:44]
helping athletes recover if you're a job
[04:47]
Seeker that person might be a recruiter
[04:48]
or hiring manager if you're working on a
[04:51]
creative brief that person might be a
[04:52]
senior product marketing manager who's
[04:54]
great at storytelling Pro tip you can
[04:56]
also name specific individuals but I
[04:58]
found the results to be good only when
[05:00]
they're famous enough like Warren
[05:01]
Buffett Steve Jobs Jeff Su
[05:04]
by the way I just have to share this we
[05:06]
have a team off-site with a superheroes
[05:08]
theme so I asked ChatGPT to draft an
[05:10]
email from Batman and it even included
[05:12]
things like please let Alfred know and
[05:14]
signed off as Your Dark Knight
[05:16]
um so fictional characters work as well
[05:18]
and I'm actually going to use this the
[05:19]
fifth component format the pro tip here
[05:22]
is to literally close your eyes and
[05:24]
visualize how exactly you want the end
[05:26]
result to look like a million likes on
[05:28]
my thirst bomb Instagram photo damn it
[05:30]
didn't work back to the user feedback
[05:32]
example I don't want to read each
[05:34]
sentence so I asked ChatGPT to take
[05:36]
all the feedback and output a table with
[05:39]
three headers the original feedback the
[05:41]
team responsible for following up and
[05:43]
priority and now I can copy this
[05:46]
directly and paste it into a Google
[05:48]
sheet sort by priority and filter by team
[05:51]
other common formats include emails
[05:53]
bullet points and code blocks but the
[05:55]
one I found to be the most useful as a
[05:57]
full-time working professional is
[05:58]
paragraphs and markdown for example I
[06:01]
just received a lengthy industry report
[06:03]
for my director first give me the three
[06:04]
key takeaways then summarize based on
[06:07]
topic use H2 as section headers here is
[06:10]
the report Pro tip whenever I use ChatGPT
[06:12]
to proofread any document I specify
[06:15]
that all changes need to be bolded so I
[06:17]
can easily see exactly what has been
[06:20]
changed let's quickly go through the
[06:21]
last component tone before we put all
[06:23]
this together in one example the good
[06:25]
news is tone is easy to understand use a
[06:28]
casual or formal tone of voice give me a
[06:31]
witty output show enthusiasm sound
[06:34]
pessimistic the bad news is we're
[06:37]
usually not very good at recalling the
[06:39]
thousands of potential adjectives and
[06:41]
adverbs at a moment's notice so here's a
[06:43]
pro tip tell ChatGPT the feeling
[06:46]
you're going for for example I'm writing
[06:48]
an email to a team I haven't worked with
[06:49]
before and I want to be taken seriously
[06:51]
without coming off as too stuck up and
[06:53]
cringy can you please give me a list of
[06:56]
five tone keywords I can include in a
[06:58]
prompt for ChatGPT and look now in the
[07:00]
actual prompt I can say use clear and
[07:02]
concise language and write in a friendly
[07:04]
get confident tone putting all this
[07:07]
together let's look at this
[07:08]
comprehensive prompt you are a senior
[07:10]
product marketing manager at Apple
[07:12]
Persona and you have just unveiled the
[07:14]
latest Apple product in collaboration
[07:16]
with Tesla the Apple car and received 12
[07:18]
000 pre-orders which is 200% higher than
[07:21]
Target context write an email to your
[07:23]
boss Tim Cookie sharing this positive
[07:25]
news task and format the email should
[07:28]
include at tl;dr too long didn't read
[07:30]
section project background why this
[07:32]
product came into existence business
[07:34]
results section quantifiable business
[07:36]
metrics and end with a section thanking
[07:39]
the product and Engineering teams
[07:41]
example structure use clear and concise
[07:44]
language and write in a confident yet
[07:46]
friendly tone tone note that if I had an
[07:49]
existing email to reference I could
[07:51]
delete the instructions around the
[07:53]
structure and simply tell ChatGPT the email
[07:56]
should follow the exact same format as a
[07:58]
one I'll share below and paste the email
[08:00]
from before by the way you can compare
[08:02]
the output from this prompt to that of a
[08:04]
simpler prompt I just launched a new
[08:06]
product at the Apple car I received 12
[08:08]
000 pre-orders please write an email to my
[08:10]
boss with this update there's a pretty
[08:12]
big difference in terms of how generic
[08:14]
and usable the end result is now that
[08:16]
you know the basics of prompting my next
[08:18]
video is going to take you from beginner
[08:19]
to Pro so make sure you're subscribed
[08:21]
for that check out my top five ChatGPT
[08:23]
productivity tips for work see you on
[08:25]
the next video in the meantime
[08:27]
have a great one