The first Prompt Engineering Mastermind! 50+ people attended.
To attend the next mastermind, register here: https://www.promptengineeringmastermind.com
People presented prompts for us to collaboratively improve:
- Prompt for instructions on how to building a website using WordPress. ChatGPT fixated on prior information. Starting a new conversation cleared that out.
- Prompt for a "therapist-like" persona, that leads a conversation and responds contextually. We experimented and ultimately settled on not just roleplaying, but roleplaying with a specific famous therapist, and got good results.
- Prompt for creating carousel content (e.g. swipeable slide content typically found on Instagram and LinkedIn). The biggest improvement were telling ChatGPT to:
- Generate the text first, then "step by step" (which forces meta-cognition) break it into separate phrases for each of the 10 slides.
- Ensure each slide builds upon the previous one and creates a cohesive narrative. The final result is more connected and improved, meeting the user's objectives.
Additional topics we covered:
- Helpful extensions for ChatGPT
- The importance of using one conversation for a prompt building prompt, and testing each iteration of that prompt in a *new* conversation
- In-depth explanation of meta prompting, which involves building a persona model for ChatGPT to help it interact more effectively
- Poll of the attendees overwhelmingly voted for masterminds being weekly, rather than biweekly.
Stay in touch on:
Youtube: youtube.com/@PromptEngineeringPodcast
Telegram: https://t.me/PromptEngineeringMastermind
LinkedIn: https://www.linkedin.com/groups/14231334/
Discover how technology is reshaping our lives and livelihoods.
Listen on: Apple Podcasts Spotify
Mark as Played
Transcript
Episode Transcript
Available transcripts are automatically generated. Complete accuracy is not guaranteed.
(01:09):
Zishan, let me unmute you andmake you a co-host.
Alright, thank you.
Okay.
You should be able to share yourscreen with the ChatGPT
interface up and your prompt up.
So I actually didn't preparethat.
I just had a prompt in mind thatI was hoping you could help
guide through how to actuallylike, set it up.
(01:29):
That is okay.
I can pull it up.
Can you just tell me what theprompt is and I will share my
screen?
Yeah, so the prompt was, lemmefind it.
Basically, I want to be able tobuild a website using web
WordPress, but I don't know how,and I want ChatGPT to take me
(01:51):
through all the steps necessary,including things like,
purchasing, new domain, hosting,whatever is relevant.
And just like steps on exactlywhat needs to be done.
Awesome.
All right.
I am sharing my screen and whileI am doing that, it looks like
Ron, you are the next one inline who asked to present.
(02:14):
So go ahead and get your promptand ChatGPT up and in five
minutes we will switch over toyou.
Zhan, does this approximate whatyou had in mind?
Let's see.
Yeah.
Yeah.
Great.
Let's see what it says and thenwe will probably want to rerun
it from scratch.
But for the people who are foreveryone except Ravi and Zhan
(02:38):
put into the chat what you wouldsuggest would help make this
better.
And I see one is already highlevel.
I would start with a personaproject component.
Awesome.
Yes, that is definitely where Iwas thinking, but I wanted to
let it run on its own first sowe could see how interesting it
might be.
All right.
From the chat, we're alsogetting the suggestion of will
(03:02):
expand each component.
Yes.
Mo tried a slightly differentprompt getting very similar
results and are some asked, doesmaking a sorry, designing
persona actually make the big ofa difference?
That's a great question.
I am gonna just restart thiswith just, the only change I'm
going to make is give it a roleplay prompt.
So as a, oops, experiencedsoftware can't spell today,
(03:30):
developer with 10 years ofexperience.
So let's just.
How much of a difference.
So it was saying gimme a domain,purchase it, hosting provider.
Some of these hostingsuggestions are interesting, and
particularly having to set up dn s not necessarily where I
would've suggested at start,but, okay.
(03:52):
Now let's see.
One, one thing I'd like to haveit focus on is more like the
actual so I think it was stepfour or five where it's install
WordPress and pick a theme, andthen it just goes to the next
step.
And I feel like there's a lotthere that very much yeah.
Cool.
So interestingly in this case,it's actually not making much of
a difference.
This output is pretty similar towhat it had before.
(04:15):
So let's follow the Yes.
What Robert was just saying.
Please tell me in short andconcise steps.
From acting as an experiencedwebsite developer.
Let's see if that helps makethis more nuanced, these steps
(04:36):
of setting up WordPress.
I'm just gonna get rid of thatand yep.
More people to admit.
Man, Ravi, this is surreal.
We're at 47 for context people.
(04:58):
We had been expecting five, Ithink was the original count
that I had.
And then, yeah, boom.
This is definitely overwhelmingfor us, for sure.
Cool.
Okay.
We've got a couple differentnuances to try.
So act as an expert in WordPressand web development or act as my
consultant.
That would be a good one.
(05:19):
I'm gonna throw that in and runthis again in a second because
Yep.
It's still really focusing moreon the, give me a domain and,
sorry.
Help me buy domain oops, that'snot the one I wanted.
Act as an expert.
Tell me.
And I think we're also this isthe last one I'm gonna do
because the next one I'mactually just gonna make a new
(05:40):
chat because it's clearly, thisis one of the things that you
run into with ChatGPT.
It's fixated on the steps that Itold it originally.
And this is a good thing and abad thing.
It's a double-edged swordbecause early on I asked it to
tell me about the very technicalpieces of buying a domain.
That was literally one of thepieces of my top prompt was
(06:02):
include steps such as buying adomain.
So it's still stuck on that.
This is actually something thatjust came up in an interview I
did for the podcast withQuestGPT, and they ran into the
same issue.
Sometimes it's very easy for GPTto get very stuck on something
it had discussed before.
Okay.
We're gonna run.
(06:22):
This one more time as a newchat, if my browser will let me.
Is there like something you cansay to it forget anything I just
said And so yes.
In practice I tend to just makea new chat.
And to be clear, acting as adeveloper, you are essentially,
(06:43):
you have a new chat every time,unless you include the previous
messages, but Okay.
Do tell me in shorten completesteps, the steps of setting up
WordPress.
So this is also on 3.5, butyeah.
See on a completely new one,domain name stuff is option
number is step.
So Sean, you can see, there'ssome nuances here that you would
(07:04):
need to play with, but that isour five minutes.
So Ron, are you ready?
Thank you.
Find you.
You're welcome.
Ron?
Ron?
Ron, where did you go?
There you are.
All right.
Ron, go ahead.
I think you're unmuted, Ron.
(07:24):
I shouldn't be.
Yeah, I can hear you.
Awesome.
Yeah.
All right.
Awesome.
Cool.
I just had a general question.
So what I wanna achieve is tocreate personas I'm working on a
GPT app that has differentpersonas.
Some of the personas will giveyou right answers to the
questions that you have and someof personas should ask or should
(07:46):
engage in a conversation or leada conversation, meaning that the
persona will ask you I don'tknow how do you feel today, for
example, and you respond withthat and then the AI should
respond.
That's awesome.
Now talk about X or Y orsomething like that.
How would you go about doingthat so that the response in of
G B T will include the nextquestion as well.
(08:10):
Does that make sense?
Not quite, but, okay.
Let me take a step back here.
Yeah.
Do you have a prompt in mind interms of written out or just the
concept?
Just the concept.
Just the concept.
Okay.
So let me make a new chat.
And I'm gonna keep this in 3.5because I realized if I use
four, we're gonna run out ofprompts very quickly cuz they're
still limited.
So tell it to me again.
(08:32):
You're trying to create personasand, sorry, I'll start the five
minute timer.
Should we have a a persona thatis like a A therapist, for
example.
Okay.
Not a real therapist butsomebody who has a best friend
to to talk to.
Therapist will ask you how yourday went.
And it will follow up withquestions after you answer how
your day went, depending on theanswers.
(08:54):
So it's context sensitive ifthat makes sense.
Okay.
But yeah, it should kinda leadthe the conversation, if that
makes sense.
Yeah.
So that was gonna be my questionis where do you want the AI to
lead the person to?
Because just ask me about yourdays.
Probably gonna go in a littlebit of a.
Yeah.
It doesn't really matter in, inthat sense, I think it's, I
(09:16):
think it's more so this will bean example for somebody who
wants to just talk to someonejust talk to, like a regular
phone conversation doesn'treally matter.
Like who they're talking to.
Exactly.
Yeah.
That will be it.
That will be it.
Okay.
How my day is going.
And feel free, what do youthink, Ravi?
(09:36):
Feel free to move on to, I thinkwe need to set it up.
We did, we need to prime it tosay, it's a conversation between
you and me.
I'm gonna ask a question andbased on that we are gonna have
a conversations, something onthose lines.
Okay.
It's going.
Feel free to feel free to takeWhat's the word I'm looking for?
(09:57):
Take subjects from what Imentioned and turn it into a
positive point to help me feelbetter.
Yeah.
Not the best grammar, but atleast something like that.
Hang on, let me just scroll onthe chat for a second.
Actually, one thing I learned,is I can add in the beginning,
can you rewrite this prompt inthe best way possible?
(10:18):
And then take that and give it.
That way you get really it'slike creating a prompt, asking
you to pro create a prompt foryou.
Yes.
Make sense?
The prompt generation ones areamazing.
Let me run this and then I wannarun that cause I completely
agree.
Yep.
So what do you mean by pro?
Pro?
So that's interesting.
So I hadn't had that word.
I haven't heard that wordbefore.
Brian Otto looks like Yeah, justpasted a prompt creator in.
(10:44):
Alright, first Ron, what do youwanna say?
How's your day been so far?
Has it gone amazing?
It's going amazing.
I am currently sharing you witha community of, what are we up
to?
49 people excited to learn aboutChatGPT, and it says, that's
(11:06):
great to hear.
It sounds like you're doingsomething productive and
engaging with others.
How does it feel to be sharingyour knowledge with others and
being a part of this community?
Ok.
So there you go.
Kinda nudging you in a positivedirection.
That's awesome.
Can screenshot of this?
Yeah.
Oh yeah.
Oh, sorry, Ron go ahead and takea screenshot at all.
I got it.
I got it just in time.
Okay.
All right.
(11:27):
Is there anything I need tochange?
We're gonna try out the promptcreation and No, we're good.
So we're running the promptcreation prompt.
This is very meta, but I've usedthem too.
They're gonna be very powerful.
Okay.
So it responded.
I'd be happy to assist creatingthe best possible prompt.
So now I will paste in ouroriginal prompt and it says,
(11:54):
here is a revised prompt you'refeeling lonely and would like to
have a conversation with atherapist.
As your therapist, I'll ask youquestions about your day and
help you find positive points toimprove your mood.
And then it asks some questions.
I thought this prompt builderhad two sections.
Oh, these are, I don't think itquite did what we had in mind.
(12:17):
As part of the prompt builder,it says, ask me questions to
make the prompt better.
Ah, yes.
However, that's, yeah, that'swhere you got.
Yeah.
The questions that it's askingare not making the prompt
better, they simply are part ofthis conversation.
Interesting.
Okay, hang on.
(12:38):
That's the five minute timer.
Quick poll in the chat.
I kind of wanna run a little bitlonger on this, and five minutes
is feeling a little short.
Can I get just a thumbs up,thumbs down emoji, or a
reaction?
You can do any of those if yourvideo's on thumbs up if you'd
rather take another fiveminutes.
Thumbs down if you would like tomove on to the next person.
All right.
I'm only seeing thumbs up, soI'm inclined to, yeah.
(13:02):
All right.
I only see thumbs up.
So we're gonna go for let's, Ithink another five minutes.
What do you think, Ravi?
Let's sure.
Let's do this current paste thatprompt in there.
Yep.
And then at the beginning, sayhelp me rewrite this prompt so
that it's clear and concise.
Help me rewrite this prompt.
(13:24):
Yeah.
And then that was the one.
See what chat.
Just said something.
Oh.
Robert, the reason I'm workingin 3.5 instead of four is if I
use four, we're gonna run out ofqueries.
Yeah.
In probably about 10 minutes,because there's a limit of 25
(13:45):
every three hours.
Okay.
Looks like Jack says they have aprompt for improving.
So Jack, if you wanna paste thatin, we can definitely give that
a shot.
But for now I will run this.
Okay.
That did not help.
The only thing that it added istell me about positive
experiences or hobbies.
(14:05):
I am curious though, so I'mgonna open another and run this
in four.
I won't keep running it in fourcause like I said, we're gonna
end up running out of queries,but I'm pretty sure four is
gonna give a much betterresponse than this because four
usually does a whole lot better.
(14:25):
And Ron, I gotta say it's such avague prompt that it's having a
hard time really coming up withsomething useful.
Yeah.
Okay.
This is a little bit better, butnot a whole lot better.
And it's solving the objectivethough, right?
So it definitely is, and it'sdefinitely trying to actually
get you to the point of, make mefeel better.
(14:46):
What if you added a persona tothe question using persona of
someone that is well known inthe field?
That is a good idea.
Although, who would that be?
I can't actually think of aperson.
Yeah.
Suggestions.
Dr.
Oz or somebody?
Oh, Esther per yes.
Esther Perel is a phenomenalexample.
Yeah.
So please act as Esther.
(15:08):
I think.
Is it two Rs or one R whatever?
A therapist.
A, let's even give it a littlebit of a cue.
Famous therapist.
And I'm gonna once again make anew chat for this.
Nice.
So that's working a whole lotbetter.
Ron, you noticed not only is itgiving you a little bit more
suggestions of directions to go,like what's weighing on your
(15:30):
mind?
Why are you feeling lonely?
It also then adds a little bitof emotional support of
sometimes just talking aboutthese things can help to ease
some of the pressure that wefeel.
Sorry, I also do voice stuff, soI occasionally I'll start acting
instead of just reading.
Alright.
That's good.
That's, I think we've got timefor one more tweak and then I
(15:51):
wanna move on to gonna mess upyour name.
Ars aam.
Don't know.
But yeah, what what's one moretweak?
Okay, cool.
Alright, awesome.
Go ahead and get your ChatGPTwindow ready with your prompt in
it, and then we can all worktogether to improve this for
you.
One piece of feedback in thechat is when revising a prompt,
(16:12):
you should describe as much aspossible about your intentions,
your objectives, how it shouldwork, and then ask it to revise
and improve your prompt.
In other words, feed it withenough information.
Okay.
Somebody just suggested that weshould have had Freud be our
therapist, and I am definitelynot going to be doing that
because that is going to getinteresting way too quick, but I
(16:34):
like the idea.
Yeah.
Yeah.
That was a good suggestion.
All right, so that's our time.
So Arsam, you should be unmuted.
Yes.
Hello?
Can you hear me?
Hello?
We can.
Hi.
Hey everyone.
So let me stop sharing.
(16:54):
Oh, sorry.
Wait, do you have it set up?
Do you have a screen to share?
I do have it set up.
I think I need to share myscreen.
Yes.
Okay.
You should be able to share it.
Let me know if you can.
I do need some guidance.
So down at the bottom there's agreen share screen button.
If you click that, saw it.
Yeah, then you can click onwindow.
All right.
(17:14):
Got it.
There we go.
All right, price start.
Greg, I wanna thank you forhosting this event.
It's super cool and I'm learningso much.
I have, I wanna share somethingpretty cool, and I also wanna
ask a question.
Let's get into it.
I'm a social media contentspecialist and I write a lot of
content carousel content.
So naturally I wanna use ChatGPTto explore carousel content and
(17:36):
see if I could prompt it in away that could write me carousel
content.
Now, to do that, I've designedengineers, I guess you could say
a prompt in three sections.
At the start, I'm gonna tell itto analyze the carousel that I'm
going to send you.
Then I'm going to send it toCarousel, that's number two.
And after that, I'm going totell it based on your
(17:58):
analyzation of my content, Iwant you to write me a c
carousel content with this topicand this main point.
So that's basically awalkthrough of the prompt.
Now Basically, this is whathappened.
I went through the first step.
First step.
Oh, interesting.
Then I sent him the carer.
Yeah.
And after I sent him thecarousel, and this is something
(18:19):
I wrote, and this is somethingthat really resonated with me.
I really enjoyed it myself.
And it went on and it broke itall down.
And it told me to, okay, this isyour writings.
This is your format, this isyour voice of tone.
And it did a pretty decent job.
I would say after that I told itto, okay write me a carousel
post on the topic of educationmatters, as an example.
(18:41):
And with the main point being,humans have a need for exploring
and education helps fulfill thatneed.
So this was the basic idea.
It did give me an answer, but itwasn't flowing that well.
And so I went on and I said,okay, I want you to make it flow
better.
Using things like, because, andso furthermore, stuff like that,
(19:03):
and it did give me a decentanswer, but I guess my main
problem and my question here ishow can we make this feel more
genuine and more humane andsomething that when you read
you're actually feeling thatconnection, if you know what I'm
saying?
Yeah.
I do wanna pause for one second.
I saw a question in the chat ofwhat is carousel content?
(19:27):
I have an answer for that, but Iwould actually rather let the
expert who is asking us aboutit.
Okay.
Could you just give like a twosentence, what is carousel
content?
Okay.
So if you've been on Instagram,you've probably seen those posts
where you could swipe to maximumof 10 slides.
Basically, in a nutshell,carousel content is.
Slide content that you couldswipe and see the next slide.
(19:48):
You have it on Instagram, youhave it on LinkedIn.
It's something that a lot ofeducators use to express their
ideas, thoughts, and knowledgeand connect with their audience
basically.
Yeah, so Jack actually has theexact thing that I was gonna
say, which is, have you thoughtabout first developing the
content as a story and thenbreaking it down as a slide.
(20:10):
However, what I would suggest iswe could start with that, and I
think we probably should, butI'm gonna do something a little
crazier anyway, which is, Ithink what we should do is ask
it to step by step, which is,again, this is a specific
phrase, step-by-step, logic andstep-by-step thinking forces it
to do meta thinking,metacognition.
(20:33):
Okay.
Develop the content a and thenstep by step, split that content
into, I don't remember how manyslides you said, but, 10 slides
or 20 slides, whatever thenumber was.
So go ahead and I would say, doyou want me to write something
here?
Yeah.
Down at the bottom.
Yeah.
Go ahead and there you go.
(20:56):
Oh, sorry.
Someone else is coming in.
Yes, I'll let you, yep.
I got it.
Thanks.
Cool.
Okay the topic that you wantedto do was, what was it?
Education, I think educationmatters, I guess you could say.
Okay.
So given, take the originalprompt that you had and paste
that in, just so we can work onthat.
Should I go through the threesteps again?
(21:18):
Yep.
Okay.
Number one, okay, so it'sanalyzing the carousel post.
Great.
It is totally missing.
Oh, no.
Okay.
Got it.
A little bit.
So it the flow number five whereeach slide builds upon the
previous one, creating acohesive narrative that supports
your main message about theimportance of using fewer words
(21:40):
for greater impact.
I'm not sure where it got the,that is the main message, but
that was the nuance that I wasgonna say we need to include.
Okay.
It got that.
So in your next prompt where yousay, give me carousel material
about whatever it was.
Okay.
Yeah.
So paste that and don't hitsubmit yet.
Yeah.
Change the topic.
(22:02):
Do you have any suggestions forthe topic?
ChatGPT matters.
That's what I was gonna say.
That's one matters.
Two ts.
Alright, there you go.
Okay.
And so what's our main pointGonna be?
Everybody.
How about for a main point?
Oh, I, that's even better than Iwas about to say.
Why it will not take our jobs.
(22:25):
Okay.
That's nice one's.
Shaun.
I was gonna go a differentdirection, not take our jobs.
Cool.
Okay, so then in front of usewords, say follow number five,
colon flow from your analysis.
Should I delete this part?
(22:45):
No, go ahead and leave that.
We're gonna tweak that in asecond.
Okay.
But in front of it.
So right here, say follow.
Yep.
Follow point number five.
From your analysis that eachslide builds upon the previous
one, creating a cohesivenarrative.
(23:05):
Each slide.
Oh.
What happened?
Were each slide, could you go onagain?
Oh yeah.
I was just basically saying whatit already said.
Creating a sorry.
Each slide builds upon theprevious one, creating a
cohesive narrative.
Just cut and paste that, yeah,you can just copy and paste
(23:27):
that.
Okay.
Should I stand before you dothat, move this text afterwards.
Okay.
And then, and then put use wordslike, okay, so right here, put a
(23:47):
colon.
So use words like colon, cuz youwant to tell it a, this is a
list of words.
So that's at least tentativelyworth trying.
Gonna go a minute over the 10minutes, but let's see what it
does.
All right.
All right.
And while it is generating whois next please, that might have
(24:09):
actually been the last personwho had a prompt.
Oh no, we got one more.
Nice.
I think we, we've seen a lot ofimprovement actually.
Yeah.
Yeah.
Yeah.
So now it's starting to do the,like sentences phrases connect
between slides.
Yeah.
Can you scroll up?
What was the first bit?
AI Matters ChatGPT really goodmany Fear ChatGPT will take
(24:33):
their jobs.
Will take jobs, but it'sdesigned to assist, not replace,
because ChatGPT needs humaninput and our creativity makes
it thrive.
Yeah.
This is a lot better.
What's the end?
Yeah.
Embrace AI like ChatGPT to growtogether.
That's awesome.
Okay.
That's the, yeah.
Very appropriate.
Got a request in the chat fromStephanie.
(24:54):
If you could copy and paste thatprompt in just the last one into
the Zoom chat 100%.
Can we do a question and answermaybe at the last 15 minutes?
Yeah, totally.
Yeah.
That way we can also answer someof them who doesn't have prompt,
but may I have questions?
Yeah, totally good with that.
Good call.
Okay.
So who was it before?
(25:16):
I think it was Robert.
Yeah, Robert.
That I had called on before.
Do you have a prompt that youwanted to, do you wanna try and
unmute?
Okay.
You do have a prompt?
Are you able to unmute?
Sorry.
I know he has a prompt becauseyou just posted it in the chat.
(25:37):
Oh, you don't have a microphone.
That would make unmuting alittle bit of a problem.
Okay.
Nevermind that idea.
Let's do a q and a.
So I might as well use Zoom asintended.
If you have a question you'dlike feedback on, does not mean
you need to have a prompt raiseyour hand, which is the raise
hand button.
And I can unmute people who havequestions.
(25:59):
I'll break the ice.
How about that?
Yeah, go for it.
Alright, so I don't know howmany of you feel that, the te
the history bars that keep ongoing forever, right?
If you're working on certainthing I wanna go back to that,
but I can, it's really hard toscroll through everything.
Is that an easier way?
I'm not seeing any search buttonthere, but have you guys figured
out anything easier way to.
(26:20):
Sorry, let me make sure I'munderstanding you're talking
about the new chat, the bar onthe left Correct.
Where it shows your previousconversations.
Is that right?
Correct.
Okay.
Yeah.
Cool.
No, there is no way, there are acouple of extensions that will
let you do that.
I honestly would have to lookthem up cuz I don't remember,
but they're like, better ChatGPTor something like that.
And they do allow searching andeven organizing the prompt
(26:41):
conversations into folders,which is pretty cool.
Okay.
Okay.
Anything specific that youliked?
The extensions that worked?
Honestly, I don't use them.
Yeah.
Meaning neith, it's not whatI've needed.
Yeah.
Okay.
I swear I just saw somebody inthe chat who said something.
They were a fan of one of them.
Yeah.
A power ChatGPT Mo says and A IP R M.
(27:05):
If you could put links intothose, that would be helpful.
Also for everyone I have justput a link into the chat.
This is indeed the launch of mypodcast today.
That happened about three hoursago, so if you could take a
minute and subscribe to it onApple Podcasts and YouTube and
Spotify and rate it five starsand like the first video, cuz so
(27:28):
far that's all I have up cuz I'mdoing this instead of finishing
editing the first two episodes.
That would be awesome and veryappreciated.
Then a feedback form for today'sMastermind.
So again if you're gettingsomething out of today, do me a
really big favor and subscribeto the podcast.
And this will be published thereas well, eventually.
(27:49):
uh, Somebody's got a hand up.
Oh, Zishaun, you've got a handup.
Great.
I think you can still unmute.
If not, let me know.
Yeah, I can.
Thank you.
Cool.
Yeah, so actually I guess wejust started talking about it
naturally, but I was wonderingwhat you, and maybe even just I
guess everyone in the chat canthrow it out there as well, but
what extensions you found to bereally valuable and helpful for
ChatGPT.
I have not found any personally,does not mean that there aren't
(28:13):
a bunch of helpful ones.
Not gonna say that at all.
Just hasn't been helpful for me.
Also I'm building a tool becausethere's some very specific kind
of testing that I've been doingthat nothing really works for.
I can share about that if peoplewant, but is anybody else seeing
things in, okay, Jack says he islooking for one that can export
(28:34):
text.
Jack actually look at the Ithink it was the a i p r M one
cause I thought that one coulddo it.
I know one of them can do thatcuz I've seen a specific listing
about export your ChatGPT outputas text or P D F or image.
And I know that exists.
Let's see.
Art, thank you.
Who else is in here?
(28:55):
Oh, that, that was actually agood point.
Do, yeah, Jack just saidsearching through the old
conversations, at leastsearching through the titles,
you can just use the normalcommand or com control f to just
search the text.
That does actually work for thetitles, obviously not for the
content.
I thought I saw another hand goup.
Ah, Stephanie, let me unmuteyou.
(29:17):
Hello?
I'm speaking.
Yeah, there we go.
Now we can hear you.
Yes.
Okay, great.
So yeah, so I've beenexperimenting a lot with prompts
to get ChatGPT to generate theperfect prompt.
Yeah, I was just curious to seeif you've done any of those, how
you feel about it.
I've gotten mixed response, withthat.
Like sometimes it, it doesgenerate the perfect prompt.
Sometimes it just gets a littletoo confused.
(29:38):
So yeah, just curious if youcould talk about that.
Ravi, it sounded like you hadsomething you were about to say
on that one, so go ahead.
Yes.
If not, I can respond.
Sure.
I, what I realized is it's likea trial and error method.
Nothing, every single timeyou're gonna get a short shot at
it.
That's what is my experience sofar.
So you keep trying it until youget what you're looking for kind
of a thing.
I think, I wish I can give you abetter answer, but that's my
(30:00):
experience.
Yeah, the the one nuance I wouldadd to that because that's a
pretty good summary.
To answer the original question,yes, there are ways, no, they're
not reliable.
There are, I'd have to look itup, but there are a couple of
prompt building prompts and acouple of prompt analysis
(30:21):
prompts.
I've heard from someone thatthey actually tag team between
the two of them, where they'lldo, build me a prompt, put it
in, get the prompt, analyze theprompt, do that, and then use
that again to build it again.
I have not actually had any luckwith that, so I'm planning on
trying it myself.
Cause I'm curious.
The other thing though that Iwould say is how do I say this?
(30:42):
Doing it iteratively is whatusually works the best for me.
With very careful considerationof now I'm gonna create a new
conversation.
So like thinking about, okayI've got like a version two
prompt, still not amazing.
Okay.
Let's completely start a newconversation and go from there.
Regarding the prompt buildingprompts I can throw a couple of
(31:03):
them into the LinkedIn.
Let me just make a note of that.
Which I guess I should clarify,the LinkedIn group is in the
document that I pasted.
I'll paste it again just so thatpeople have it.
Trying some different ones isactually what I've seen works
the best.
I haven't, like this one alwaysworks.
Yeah.
Looks like just pasted a promptbuilder.
(31:24):
So Stephanie, you can check thatout in the chat and Tomer would
like to ask a question.
Tomer, you should be unmuted.
Yes.
Yeah.
Hi everyone.
Sweet.
Hey, so on my side, sinceStephanie brought up the
discussion on the promptcreator, I just wanted to ask,
(31:45):
cuz I'm like in this cycle whereit's asking the questions and
I'm trying to respond.
I'm just wondering does that endor is it, we've got enough to
the wrong foot and I shouldstart a new chat.
What's your opinion on this?
Sorry, making sure I'munderstanding the question you
asked, does it end?
It's only gonna keep going asmany times as you say, give me
(32:06):
another round, make it better.
It's not like it will take overand start just messaging you
like message message message..
Like ChatGPT doesn't really dothat.
It's keep going on you keepasking it, right?
So obviously you need to refineyour question and start asking
it.
It's just a trial and error.
I was just mentioning it toStephanie the other time.
Until you get it and then you,once you start and then you will
you will get a hang of it, andthen you know what to ask next.
(32:31):
I see what I can give you.
I can give you an example.
I'm a blog writer.
I write blogs, right?
So when I write the blog, Icreated a prompt for myself
saying, okay, I wanna rate canyou rate my prompt my blog based
upon the readability, the timingsearch keyboards value like
seven parameters, I decided,right?
(32:51):
It gives me a value for, on ascale of nine one to 10, every
single time.
Different number, but the sameblock.
So not surprised to hear that.
Yeah.
Yeah, I think I'm covered.
I think I need Excellent.
Okay, cool.
Tomer, just the last thing tomake sure of is you wanna run
one conversation prompt,upgrading, and the other
(33:12):
conversation testing the prompt.
You don't want to do it in thesame conversation.
Cause then it will get veryconfused.
I think that's what I've beendoing wrong.
Yeah.
Okay.
Yeah.
That's what I was thinking.
If you're not, if you're doingit in one conversation, the
prompt optimizer will just failafter one round.
That's a good tip.
Yeah, that's a good tip to Greg.
(33:32):
I like that.
Actually.
I didn't realize I want to.
Cool.
All right.
Let's see.
We've got five minutes left.
You're welcome.
So there was somebody talkingabout meta prompt.
Yes.
Who was that?
Would you mind whoever that, Isee the message.
Let me see if I can unmute you.
I, Hey.
Awesome.
(33:53):
Please share about metaprompting.
I have a definition, but I wouldlike to hear yours and Yeah.
What you found useful.
So many things.
Meta prompting is anything whereyou use chain of thought but you
adapt it out to higher levelthings So First, before I do
anything, go subscribe to thisman pro this dude is awesome for
being able to host this kind ofthing and getting us together on
(34:13):
LinkedIn.
So please anybody, cuz I can seehow many subs are over there and
I can see how many people are inthis thing.
So y'all not doing your job.
Go over there sub to hischannel.
Outside of that let's go aheadand get started on this stuff.
Meta prompting is going to beanything where you take a higher
level overview at the approach.
So instead of approachingChatGPT at a single point where
(34:34):
you're gonna go, Hey, I want youto build out this carousel
thing.
I want you to do a SEO blogpost.
It's okay to do that.
Then you have one higher levelwhere you can send tax it.
Right now you're gonna go, Iwant the blog post to look like
this.
Or maybe you wanna do like alevel one role play.
These are all beautiful thingsbecause they're really
attributes and characteristicsof how ChatGPT will relate with
(34:55):
you and since actually respond.
That's great.
But what if we take all of thesetraits, put it into one persona
model, shove it at the top, makethat a master model, and then
once we have this level ofmaster model, what we're gonna
end up doing is that once we'vegenerated that, we're gonna load
(35:16):
it with a persona profile So forexample, when you were talking
about the carousel content,right?
The Instagram sell carousel hostthis is what I built out for
that exactly.
So I loaded it with a simplemaster model.
This is a very simple one mindyou.
Eva and I have other things thatare like six post deep.
Now that we have the besteducator in the world that we're
(35:38):
interacting with, ChatGPT hasthis construct of who it should
be, how it should be, how itshould talk.
So why does this matter?
Your time is the most valuablething in the world, right?
So what if I told you couldspend less time on ChatGPT
getting more out of it.
That's what this is.
If you take the time to build amaster model with your specialty
and experiences, because I'm notyou.
(36:00):
You know what your specialthings is?
Maybe you're a coder, maybeyou're a marketer, you know how
you want to define the model torespond to you or to do things,
but you don't have to thinkabout it from such a how do I
want the carousel post?
No, ignore that.
You're defining a person.
If you were to hire a person todo that job for you, how do you
want that person to be?
(36:22):
That's what you build first.
That's your persona model.
Then you secondarily move intowhat you want it to be.
So you give it a job.
You are named builder.
You're gonna create a project.
This project's gonna be thisthing.
So now you have all of thesealiases or variables you could
(36:43):
say to reference back to thethings you want.
If you want to call back theproject, you just say, I'm doing
this project, and Tai will knowexactly what you're talking
about.
So now instead of having toredefine the whole damn thing
every post or do this hugesyntax, programmatic language,
you can get lazier and lazierwhile ChatGPT reinforces itself
and get smarter and smarterbecause you preemptively told
(37:06):
it.
That's what it should do.
In fact, there was a researchpaper that was released on this
called Chat Gcpt plusReflection, which shows the
accuracy of GPT four goes fromsomething like, what is it, 77
to I, I don't wanna misquote it,but it's something like 10 to
20% increase in accuracy.
It's insane levels and if that'swhat's is amazing.
(37:29):
Yeah.
And just realized we haveactually, we just hit the hour.
Real quick before everybody hopsoff, I'm gonna share a quick
poll How often do you want to dothese?
You wanna do them every week?
Do you wanna do them every otherweek?
So if you could answer thatreally quick and Wow.
Very strong reaction for weeklyawesome.
(37:49):
Okay.
Thank you all so much forattending.
This was a ton of fun for me.
I hope it was a fun I hope itwas a lot of fun for you.
This recording will be shared onthe YouTube channel aside from
me wanting you to support thepodcast, also, if you want the
recording and you don't wannasubscribe, the link is in that
document that I put which sure Ican put that again.
(38:11):
This one.
And if this has been helpful foryou please do me a solid and go
subscribe to the podcast inwherever you do it.
Rate it, five stars.
I'm gonna be posting a wholebunch more episodes of people
coming in and havingconversations with them about
like, How do you do promptengineering?
(38:33):
What have you learned, all thatkind of stuff.
Two other things.
If you are a person who writesprompts and would like to be on
the podcasts honestly, just sendme an email, email redacted but
also you can comment in theLinkedIn or reply to any of the
dozen places that I think Iposted this.
Oh one last thing.
The paper about reflection justgot posted in the comments, so
(38:57):
make sure that you grab that.
It is a very cool technique andI'm hopefully going to be
talking about that on anupcoming episode.
And if not, I might just do anepisode myself on it cuz it's
pretty cool.
Awesome.
Thank you all so much and talkto you soon.
Greg Schwartz (39:26):
Thanks for coming to the Prompt Engineering
podcast, the podcast dedicatedto helping you learn how to be a
better prompt engineer, how tosell your prompts if that's
something you're into, or justuse them in your day job.
See you soon.
Zishan, let me unmute you andmake you a co-host.
Alright, thank you.
Okay.
You should be able to share yourscreen with the ChatGPT
interface up and your prompt up.
So I actually didn't preparethat.
I just had a prompt in mind thatI was hoping you could help
guide through how to actuallylike, set it up.
(01:29):
That is okay.
I can pull it up.
Can you just tell me what theprompt is and I will share my
screen?
Yeah, so the prompt was, lemmefind it.
Basically, I want to be able tobuild a website using web
WordPress, but I don't know how,and I want ChatGPT to take me
(01:51):
through all the steps necessary,including things like,
purchasing, new domain, hosting,whatever is relevant.
And just like steps on exactlywhat needs to be done.
Awesome.
All right.
I am sharing my screen and whileI am doing that, it looks like
Ron, you are the next one inline who asked to present.
(02:14):
So go ahead and get your promptand ChatGPT up and in five
minutes we will switch over toyou.
Zhan, does this approximate whatyou had in mind?
Let's see.
Yeah.
Yeah.
Great.
Let's see what it says and thenwe will probably want to rerun
it from scratch.
But for the people who are foreveryone except Ravi and Zhan
(02:38):
put into the chat what you wouldsuggest would help make this
better.
And I see one is already highlevel.
I would start with a personaproject component.
Awesome.
Yes, that is definitely where Iwas thinking, but I wanted to
let it run on its own first sowe could see how interesting it
might be.
All right.
From the chat, we're alsogetting the suggestion of will
(03:02):
expand each component.
Yes.
Mo tried a slightly differentprompt getting very similar
results and are some asked, doesmaking a sorry, designing
persona actually make the big ofa difference?
That's a great question.
I am gonna just restart thiswith just, the only change I'm
going to make is give it a roleplay prompt.
So as a, oops, experiencedsoftware can't spell today,
(03:30):
developer with 10 years ofexperience.
So let's just.
How much of a difference.
So it was saying gimme a domain,purchase it, hosting provider.
Some of these hostingsuggestions are interesting, and
particularly having to set up dn s not necessarily where I
would've suggested at start,but, okay.
(03:52):
Now let's see.
One, one thing I'd like to haveit focus on is more like the
actual so I think it was stepfour or five where it's install
WordPress and pick a theme, andthen it just goes to the next
step.
And I feel like there's a lotthere that very much yeah.
Cool.
So interestingly in this case,it's actually not making much of
a difference.
This output is pretty similar towhat it had before.
(04:15):
So let's follow the Yes.
What Robert was just saying.
Please tell me in short andconcise steps.
From acting as an experiencedwebsite developer.
Let's see if that helps makethis more nuanced, these steps
(04:36):
of setting up WordPress.
I'm just gonna get rid of thatand yep.
More people to admit.
Man, Ravi, this is surreal.
We're at 47 for context people.
(04:58):
We had been expecting five, Ithink was the original count
that I had.
And then, yeah, boom.
This is definitely overwhelmingfor us, for sure.
Cool.
Okay.
We've got a couple differentnuances to try.
So act as an expert in WordPressand web development or act as my
consultant.
That would be a good one.
(05:19):
I'm gonna throw that in and runthis again in a second because
Yep.
It's still really focusing moreon the, give me a domain and,
sorry.
Help me buy domain oops, that'snot the one I wanted.
Act as an expert.
Tell me.
And I think we're also this isthe last one I'm gonna do
because the next one I'mactually just gonna make a new
(05:40):
chat because it's clearly, thisis one of the things that you
run into with ChatGPT.
It's fixated on the steps that Itold it originally.
And this is a good thing and abad thing.
It's a double-edged swordbecause early on I asked it to
tell me about the very technicalpieces of buying a domain.
That was literally one of thepieces of my top prompt was
(06:02):
include steps such as buying adomain.
So it's still stuck on that.
This is actually something thatjust came up in an interview I
did for the podcast withQuestGPT, and they ran into the
same issue.
Sometimes it's very easy for GPTto get very stuck on something
it had discussed before.
Okay.
We're gonna run.
(06:22):
This one more time as a newchat, if my browser will let me.
Is there like something you cansay to it forget anything I just
said And so yes.
In practice I tend to just makea new chat.
And to be clear, acting as adeveloper, you are essentially,
(06:43):
you have a new chat every time,unless you include the previous
messages, but Okay.
Do tell me in shorten completesteps, the steps of setting up
WordPress.
So this is also on 3.5, butyeah.
See on a completely new one,domain name stuff is option
number is step.
So Sean, you can see, there'ssome nuances here that you would
(07:04):
need to play with, but that isour five minutes.
So Ron, are you ready?
Thank you.
Find you.
You're welcome.
Ron?
Ron?
Ron, where did you go?
There you are.
All right.
Ron, go ahead.
I think you're unmuted, Ron.
(07:24):
I shouldn't be.
Yeah, I can hear you.
Awesome.
Yeah.
All right.
Awesome.
Cool.
I just had a general question.
So what I wanna achieve is tocreate personas I'm working on a
GPT app that has differentpersonas.
Some of the personas will giveyou right answers to the
questions that you have and someof personas should ask or should
(07:46):
engage in a conversation or leada conversation, meaning that the
persona will ask you I don'tknow how do you feel today, for
example, and you respond withthat and then the AI should
respond.
That's awesome.
Now talk about X or Y orsomething like that.
How would you go about doingthat so that the response in of
G B T will include the nextquestion as well.
(08:10):
Does that make sense?
Not quite, but, okay.
Let me take a step back here.
Yeah.
Do you have a prompt in mind interms of written out or just the
concept?
Just the concept.
Just the concept.
Okay.
So let me make a new chat.
And I'm gonna keep this in 3.5because I realized if I use
four, we're gonna run out ofprompts very quickly cuz they're
still limited.
So tell it to me again.
(08:32):
You're trying to create personasand, sorry, I'll start the five
minute timer.
Should we have a a persona thatis like a A therapist, for
example.
Okay.
Not a real therapist butsomebody who has a best friend
to to talk to.
Therapist will ask you how yourday went.
And it will follow up withquestions after you answer how
your day went, depending on theanswers.
(08:54):
So it's context sensitive ifthat makes sense.
Okay.
But yeah, it should kinda leadthe the conversation, if that
makes sense.
Yeah.
So that was gonna be my questionis where do you want the AI to
lead the person to?
Because just ask me about yourdays.
Probably gonna go in a littlebit of a.
Yeah.
It doesn't really matter in, inthat sense, I think it's, I
(09:16):
think it's more so this will bean example for somebody who
wants to just talk to someonejust talk to, like a regular
phone conversation doesn'treally matter.
Like who they're talking to.
Exactly.
Yeah.
That will be it.
That will be it.
Okay.
How my day is going.
And feel free, what do youthink, Ravi?
(09:36):
Feel free to move on to, I thinkwe need to set it up.
We did, we need to prime it tosay, it's a conversation between
you and me.
I'm gonna ask a question andbased on that we are gonna have
a conversations, something onthose lines.
Okay.
It's going.
Feel free to feel free to takeWhat's the word I'm looking for?
(09:57):
Take subjects from what Imentioned and turn it into a
positive point to help me feelbetter.
Yeah.
Not the best grammar, but atleast something like that.
Hang on, let me just scroll onthe chat for a second.
Actually, one thing I learned,is I can add in the beginning,
can you rewrite this prompt inthe best way possible?
(10:18):
And then take that and give it.
That way you get really it'slike creating a prompt, asking
you to pro create a prompt foryou.
Yes.
Make sense?
The prompt generation ones areamazing.
Let me run this and then I wannarun that cause I completely
agree.
Yep.
So what do you mean by pro?
Pro?
So that's interesting.
So I hadn't had that word.
I haven't heard that wordbefore.
Brian Otto looks like Yeah, justpasted a prompt creator in.
(10:44):
Alright, first Ron, what do youwanna say?
How's your day been so far?
Has it gone amazing?
It's going amazing.
I am currently sharing you witha community of, what are we up
to?
49 people excited to learn aboutChatGPT, and it says, that's
(11:06):
great to hear.
It sounds like you're doingsomething productive and
engaging with others.
How does it feel to be sharingyour knowledge with others and
being a part of this community?
Ok.
So there you go.
Kinda nudging you in a positivedirection.
That's awesome.
Can screenshot of this?
Yeah.
Oh yeah.
Oh, sorry, Ron go ahead and takea screenshot at all.
I got it.
I got it just in time.
Okay.
All right.
(11:27):
Is there anything I need tochange?
We're gonna try out the promptcreation and No, we're good.
So we're running the promptcreation prompt.
This is very meta, but I've usedthem too.
They're gonna be very powerful.
Okay.
So it responded.
I'd be happy to assist creatingthe best possible prompt.
So now I will paste in ouroriginal prompt and it says,
(11:54):
here is a revised prompt you'refeeling lonely and would like to
have a conversation with atherapist.
As your therapist, I'll ask youquestions about your day and
help you find positive points toimprove your mood.
And then it asks some questions.
I thought this prompt builderhad two sections.
Oh, these are, I don't think itquite did what we had in mind.
(12:17):
As part of the prompt builder,it says, ask me questions to
make the prompt better.
Ah, yes.
However, that's, yeah, that'swhere you got.
Yeah.
The questions that it's askingare not making the prompt
better, they simply are part ofthis conversation.
Interesting.
Okay, hang on.
(12:38):
That's the five minute timer.
Quick poll in the chat.
I kind of wanna run a little bitlonger on this, and five minutes
is feeling a little short.
Can I get just a thumbs up,thumbs down emoji, or a
reaction?
You can do any of those if yourvideo's on thumbs up if you'd
rather take another fiveminutes.
Thumbs down if you would like tomove on to the next person.
All right.
I'm only seeing thumbs up, soI'm inclined to, yeah.
(13:02):
All right.
I only see thumbs up.
So we're gonna go for let's, Ithink another five minutes.
What do you think, Ravi?
Let's sure.
Let's do this current paste thatprompt in there.
Yep.
And then at the beginning, sayhelp me rewrite this prompt so
that it's clear and concise.
Help me rewrite this prompt.
(13:24):
Yeah.
And then that was the one.
See what chat.
Just said something.
Oh.
Robert, the reason I'm workingin 3.5 instead of four is if I
use four, we're gonna run out ofqueries.
Yeah.
In probably about 10 minutes,because there's a limit of 25
(13:45):
every three hours.
Okay.
Looks like Jack says they have aprompt for improving.
So Jack, if you wanna paste thatin, we can definitely give that
a shot.
But for now I will run this.
Okay.
That did not help.
The only thing that it added istell me about positive
experiences or hobbies.
(14:05):
I am curious though, so I'mgonna open another and run this
in four.
I won't keep running it in fourcause like I said, we're gonna
end up running out of queries,but I'm pretty sure four is
gonna give a much betterresponse than this because four
usually does a whole lot better.
(14:25):
And Ron, I gotta say it's such avague prompt that it's having a
hard time really coming up withsomething useful.
Yeah.
Okay.
This is a little bit better, butnot a whole lot better.
And it's solving the objectivethough, right?
So it definitely is, and it'sdefinitely trying to actually
get you to the point of, make mefeel better.
(14:46):
What if you added a persona tothe question using persona of
someone that is well known inthe field?
That is a good idea.
Although, who would that be?
I can't actually think of aperson.
Yeah.
Suggestions.
Dr.
Oz or somebody?
Oh, Esther per yes.
Esther Perel is a phenomenalexample.
Yeah.
So please act as Esther.
(15:08):
I think.
Is it two Rs or one R whatever?
A therapist.
A, let's even give it a littlebit of a cue.
Famous therapist.
And I'm gonna once again make anew chat for this.
Nice.
So that's working a whole lotbetter.
Ron, you noticed not only is itgiving you a little bit more
suggestions of directions to go,like what's weighing on your
(15:30):
mind?
Why are you feeling lonely?
It also then adds a little bitof emotional support of
sometimes just talking aboutthese things can help to ease
some of the pressure that wefeel.
Sorry, I also do voice stuff, soI occasionally I'll start acting
instead of just reading.
Alright.
That's good.
That's, I think we've got timefor one more tweak and then I
(15:51):
wanna move on to gonna mess upyour name.
Ars aam.
Don't know.
But yeah, what what's one moretweak?
Okay, cool.
Alright, awesome.
Go ahead and get your ChatGPTwindow ready with your prompt in
it, and then we can all worktogether to improve this for
you.
One piece of feedback in thechat is when revising a prompt,
(16:12):
you should describe as much aspossible about your intentions,
your objectives, how it shouldwork, and then ask it to revise
and improve your prompt.
In other words, feed it withenough information.
Okay.
Somebody just suggested that weshould have had Freud be our
therapist, and I am definitelynot going to be doing that
because that is going to getinteresting way too quick, but I
(16:34):
like the idea.
Yeah.
Yeah.
That was a good suggestion.
All right, so that's our time.
So Arsam, you should be unmuted.
Yes.
Hello?
Can you hear me?
Hello?
We can.
Hi.
Hey everyone.
So let me stop sharing.
(16:54):
Oh, sorry.
Wait, do you have it set up?
Do you have a screen to share?
I do have it set up.
I think I need to share myscreen.
Yes.
Okay.
You should be able to share it.
Let me know if you can.
I do need some guidance.
So down at the bottom there's agreen share screen button.
If you click that, saw it.
Yeah, then you can click onwindow.
All right.
(17:14):
Got it.
There we go.
All right, price start.
Greg, I wanna thank you forhosting this event.
It's super cool and I'm learningso much.
I have, I wanna share somethingpretty cool, and I also wanna
ask a question.
Let's get into it.
I'm a social media contentspecialist and I write a lot of
content carousel content.
So naturally I wanna use ChatGPTto explore carousel content and
(17:36):
see if I could prompt it in away that could write me carousel
content.
Now, to do that, I've designedengineers, I guess you could say
a prompt in three sections.
At the start, I'm gonna tell itto analyze the carousel that I'm
going to send you.
Then I'm going to send it toCarousel, that's number two.
And after that, I'm going totell it based on your
(17:58):
analyzation of my content, Iwant you to write me a c
carousel content with this topicand this main point.
So that's basically awalkthrough of the prompt.
Now Basically, this is whathappened.
I went through the first step.
First step.
Oh, interesting.
Then I sent him the carer.
Yeah.
And after I sent him thecarousel, and this is something
(18:19):
I wrote, and this is somethingthat really resonated with me.
I really enjoyed it myself.
And it went on and it broke itall down.
And it told me to, okay, this isyour writings.
This is your format, this isyour voice of tone.
And it did a pretty decent job.
I would say after that I told itto, okay write me a carousel
post on the topic of educationmatters, as an example.
(18:41):
And with the main point being,humans have a need for exploring
and education helps fulfill thatneed.
So this was the basic idea.
It did give me an answer, but itwasn't flowing that well.
And so I went on and I said,okay, I want you to make it flow
better.
Using things like, because, andso furthermore, stuff like that,
(19:03):
and it did give me a decentanswer, but I guess my main
problem and my question here ishow can we make this feel more
genuine and more humane andsomething that when you read
you're actually feeling thatconnection, if you know what I'm
saying?
Yeah.
I do wanna pause for one second.
I saw a question in the chat ofwhat is carousel content?
(19:27):
I have an answer for that, but Iwould actually rather let the
expert who is asking us aboutit.
Okay.
Could you just give like a twosentence, what is carousel
content?
Okay.
So if you've been on Instagram,you've probably seen those posts
where you could swipe to maximumof 10 slides.
Basically, in a nutshell,carousel content is.
Slide content that you couldswipe and see the next slide.
(19:48):
You have it on Instagram, youhave it on LinkedIn.
It's something that a lot ofeducators use to express their
ideas, thoughts, and knowledgeand connect with their audience
basically.
Yeah, so Jack actually has theexact thing that I was gonna
say, which is, have you thoughtabout first developing the
content as a story and thenbreaking it down as a slide.
(20:10):
However, what I would suggest iswe could start with that, and I
think we probably should, butI'm gonna do something a little
crazier anyway, which is, Ithink what we should do is ask
it to step by step, which is,again, this is a specific
phrase, step-by-step, logic andstep-by-step thinking forces it
to do meta thinking,metacognition.
(20:33):
Okay.
Develop the content a and thenstep by step, split that content
into, I don't remember how manyslides you said, but, 10 slides
or 20 slides, whatever thenumber was.
So go ahead and I would say, doyou want me to write something
here?
Yeah.
Down at the bottom.
Yeah.
Go ahead and there you go.
(20:56):
Oh, sorry.
Someone else is coming in.
Yes, I'll let you, yep.
I got it.
Thanks.
Cool.
Okay the topic that you wantedto do was, what was it?
Education, I think educationmatters, I guess you could say.
Okay.
So given, take the originalprompt that you had and paste
that in, just so we can work onthat.
Should I go through the threesteps again?
(21:18):
Yep.
Okay.
Number one, okay, so it'sanalyzing the carousel post.
Great.
It is totally missing.
Oh, no.
Okay.
Got it.
A little bit.
So it the flow number five whereeach slide builds upon the
previous one, creating acohesive narrative that supports
your main message about theimportance of using fewer words
(21:40):
for greater impact.
I'm not sure where it got the,that is the main message, but
that was the nuance that I wasgonna say we need to include.
Okay.
It got that.
So in your next prompt where yousay, give me carousel material
about whatever it was.
Okay.
Yeah.
So paste that and don't hitsubmit yet.
Yeah.
Change the topic.
(22:02):
Do you have any suggestions forthe topic?
ChatGPT matters.
That's what I was gonna say.
That's one matters.
Two ts.
Alright, there you go.
Okay.
And so what's our main pointGonna be?
Everybody.
How about for a main point?
Oh, I, that's even better than Iwas about to say.
Why it will not take our jobs.
(22:25):
Okay.
That's nice one's.
Shaun.
I was gonna go a differentdirection, not take our jobs.
Cool.
Okay, so then in front of usewords, say follow number five,
colon flow from your analysis.
Should I delete this part?
(22:45):
No, go ahead and leave that.
We're gonna tweak that in asecond.
Okay.
But in front of it.
So right here, say follow.
Yep.
Follow point number five.
From your analysis that eachslide builds upon the previous
one, creating a cohesivenarrative.
(23:05):
Each slide.
Oh.
What happened?
Were each slide, could you go onagain?
Oh yeah.
I was just basically saying whatit already said.
Creating a sorry.
Each slide builds upon theprevious one, creating a
cohesive narrative.
Just cut and paste that, yeah,you can just copy and paste
(23:27):
that.
Okay.
Should I stand before you dothat, move this text afterwards.
Okay.
And then, and then put use wordslike, okay, so right here, put a
(23:47):
colon.
So use words like colon, cuz youwant to tell it a, this is a
list of words.
So that's at least tentativelyworth trying.
Gonna go a minute over the 10minutes, but let's see what it
does.
All right.
All right.
And while it is generating whois next please, that might have
(24:09):
actually been the last personwho had a prompt.
Oh no, we got one more.
Nice.
I think we, we've seen a lot ofimprovement actually.
Yeah.
Yeah.
Yeah.
So now it's starting to do the,like sentences phrases connect
between slides.
Yeah.
Can you scroll up?
What was the first bit?
AI Matters ChatGPT really goodmany Fear ChatGPT will take
(24:33):
their jobs.
Will take jobs, but it'sdesigned to assist, not replace,
because ChatGPT needs humaninput and our creativity makes
it thrive.
Yeah.
This is a lot better.
What's the end?
Yeah.
Embrace AI like ChatGPT to growtogether.
That's awesome.
Okay.
That's the, yeah.
Very appropriate.
Got a request in the chat fromStephanie.
(24:54):
If you could copy and paste thatprompt in just the last one into
the Zoom chat 100%.
Can we do a question and answermaybe at the last 15 minutes?
Yeah, totally.
Yeah.
That way we can also answer someof them who doesn't have prompt,
but may I have questions?
Yeah, totally good with that.
Good call.
Okay.
So who was it before?
(25:16):
I think it was Robert.
Yeah, Robert.
That I had called on before.
Do you have a prompt that youwanted to, do you wanna try and
unmute?
Okay.
You do have a prompt?
Are you able to unmute?
Sorry.
I know he has a prompt becauseyou just posted it in the chat.
(25:37):
Oh, you don't have a microphone.
That would make unmuting alittle bit of a problem.
Okay.
Nevermind that idea.
Let's do a q and a.
So I might as well use Zoom asintended.
If you have a question you'dlike feedback on, does not mean
you need to have a prompt raiseyour hand, which is the raise
hand button.
And I can unmute people who havequestions.
(25:59):
I'll break the ice.
How about that?
Yeah, go for it.
Alright, so I don't know howmany of you feel that, the te
the history bars that keep ongoing forever, right?
If you're working on certainthing I wanna go back to that,
but I can, it's really hard toscroll through everything.
Is that an easier way?
I'm not seeing any search buttonthere, but have you guys figured
out anything easier way to.
(26:20):
Sorry, let me make sure I'munderstanding you're talking
about the new chat, the bar onthe left Correct.
Where it shows your previousconversations.
Is that right?
Correct.
Okay.
Yeah.
Cool.
No, there is no way, there are acouple of extensions that will
let you do that.
I honestly would have to lookthem up cuz I don't remember,
but they're like, better ChatGPTor something like that.
And they do allow searching andeven organizing the prompt
(26:41):
conversations into folders,which is pretty cool.
Okay.
Okay.
Anything specific that youliked?
The extensions that worked?
Honestly, I don't use them.
Yeah.
Meaning neith, it's not whatI've needed.
Yeah.
Okay.
I swear I just saw somebody inthe chat who said something.
They were a fan of one of them.
Yeah.
A power ChatGPT Mo says and A IP R M.
(27:05):
If you could put links intothose, that would be helpful.
Also for everyone I have justput a link into the chat.
This is indeed the launch of mypodcast today.
That happened about three hoursago, so if you could take a
minute and subscribe to it onApple Podcasts and YouTube and
Spotify and rate it five starsand like the first video, cuz so
(27:28):
far that's all I have up cuz I'mdoing this instead of finishing
editing the first two episodes.
That would be awesome and veryappreciated.
Then a feedback form for today'sMastermind.
So again if you're gettingsomething out of today, do me a
really big favor and subscribeto the podcast.
And this will be published thereas well, eventually.
(27:49):
uh, Somebody's got a hand up.
Oh, Zishaun, you've got a handup.
Great.
I think you can still unmute.
If not, let me know.
Yeah, I can.
Thank you.
Cool.
Yeah, so actually I guess wejust started talking about it
naturally, but I was wonderingwhat you, and maybe even just I
guess everyone in the chat canthrow it out there as well, but
what extensions you found to bereally valuable and helpful for
ChatGPT.
I have not found any personally,does not mean that there aren't
(28:13):
a bunch of helpful ones.
Not gonna say that at all.
Just hasn't been helpful for me.
Also I'm building a tool becausethere's some very specific kind
of testing that I've been doingthat nothing really works for.
I can share about that if peoplewant, but is anybody else seeing
things in, okay, Jack says he islooking for one that can export
(28:34):
text.
Jack actually look at the Ithink it was the a i p r M one
cause I thought that one coulddo it.
I know one of them can do thatcuz I've seen a specific listing
about export your ChatGPT outputas text or P D F or image.
And I know that exists.
Let's see.
Art, thank you.
Who else is in here?
(28:55):
Oh, that, that was actually agood point.
Do, yeah, Jack just saidsearching through the old
conversations, at leastsearching through the titles,
you can just use the normalcommand or com control f to just
search the text.
That does actually work for thetitles, obviously not for the
content.
I thought I saw another hand goup.
Ah, Stephanie, let me unmuteyou.
(29:17):
Hello?
I'm speaking.
Yeah, there we go.
Now we can hear you.
Yes.
Okay, great.
So yeah, so I've beenexperimenting a lot with prompts
to get ChatGPT to generate theperfect prompt.
Yeah, I was just curious to seeif you've done any of those, how
you feel about it.
I've gotten mixed response, withthat.
Like sometimes it, it doesgenerate the perfect prompt.
Sometimes it just gets a littletoo confused.
(29:38):
So yeah, just curious if youcould talk about that.
Ravi, it sounded like you hadsomething you were about to say
on that one, so go ahead.
Yes.
If not, I can respond.
Sure.
I, what I realized is it's likea trial and error method.
Nothing, every single timeyou're gonna get a short shot at
it.
That's what is my experience sofar.
So you keep trying it until youget what you're looking for kind
of a thing.
I think, I wish I can give you abetter answer, but that's my
(30:00):
experience.
Yeah, the the one nuance I wouldadd to that because that's a
pretty good summary.
To answer the original question,yes, there are ways, no, they're
not reliable.
There are, I'd have to look itup, but there are a couple of
prompt building prompts and acouple of prompt analysis
(30:21):
prompts.
I've heard from someone thatthey actually tag team between
the two of them, where they'lldo, build me a prompt, put it
in, get the prompt, analyze theprompt, do that, and then use
that again to build it again.
I have not actually had any luckwith that, so I'm planning on
trying it myself.
Cause I'm curious.
The other thing though that Iwould say is how do I say this?
(30:42):
Doing it iteratively is whatusually works the best for me.
With very careful considerationof now I'm gonna create a new
conversation.
So like thinking about, okayI've got like a version two
prompt, still not amazing.
Okay.
Let's completely start a newconversation and go from there.
Regarding the prompt buildingprompts I can throw a couple of
(31:03):
them into the LinkedIn.
Let me just make a note of that.
Which I guess I should clarify,the LinkedIn group is in the
document that I pasted.
I'll paste it again just so thatpeople have it.
Trying some different ones isactually what I've seen works
the best.
I haven't, like this one alwaysworks.
Yeah.
Looks like just pasted a promptbuilder.
(31:24):
So Stephanie, you can check thatout in the chat and Tomer would
like to ask a question.
Tomer, you should be unmuted.
Yes.
Yeah.
Hi everyone.
Sweet.
Hey, so on my side, sinceStephanie brought up the
discussion on the promptcreator, I just wanted to ask,
(31:45):
cuz I'm like in this cycle whereit's asking the questions and
I'm trying to respond.
I'm just wondering does that endor is it, we've got enough to
the wrong foot and I shouldstart a new chat.
What's your opinion on this?
Sorry, making sure I'munderstanding the question you
asked, does it end?
It's only gonna keep going asmany times as you say, give me
(32:06):
another round, make it better.
It's not like it will take overand start just messaging you
like message message message..
Like ChatGPT doesn't really dothat.
It's keep going on you keepasking it, right?
So obviously you need to refineyour question and start asking
it.
It's just a trial and error.
I was just mentioning it toStephanie the other time.
Until you get it and then you,once you start and then you will
you will get a hang of it, andthen you know what to ask next.
(32:31):
I see what I can give you.
I can give you an example.
I'm a blog writer.
I write blogs, right?
So when I write the blog, Icreated a prompt for myself
saying, okay, I wanna rate canyou rate my prompt my blog based
upon the readability, the timingsearch keyboards value like
seven parameters, I decided,right?
(32:51):
It gives me a value for, on ascale of nine one to 10, every
single time.
Different number, but the sameblock.
So not surprised to hear that.
Yeah.
Yeah, I think I'm covered.
I think I need Excellent.
Okay, cool.
Tomer, just the last thing tomake sure of is you wanna run
one conversation prompt,upgrading, and the other
(33:12):
conversation testing the prompt.
You don't want to do it in thesame conversation.
Cause then it will get veryconfused.
I think that's what I've beendoing wrong.
Yeah.
Okay.
Yeah.
That's what I was thinking.
If you're not, if you're doingit in one conversation, the
prompt optimizer will just failafter one round.
That's a good tip.
Yeah, that's a good tip to Greg.
(33:32):
I like that.
Actually.
I didn't realize I want to.
Cool.
All right.
Let's see.
We've got five minutes left.
You're welcome.
So there was somebody talkingabout meta prompt.
Yes.
Who was that?
Would you mind whoever that, Isee the message.
Let me see if I can unmute you.
I, Hey.
Awesome.
(33:53):
Please share about metaprompting.
I have a definition, but I wouldlike to hear yours and Yeah.
What you found useful.
So many things.
Meta prompting is anything whereyou use chain of thought but you
adapt it out to higher levelthings So First, before I do
anything, go subscribe to thisman pro this dude is awesome for
being able to host this kind ofthing and getting us together on
(34:13):
LinkedIn.
So please anybody, cuz I can seehow many subs are over there and
I can see how many people are inthis thing.
So y'all not doing your job.
Go over there sub to hischannel.
Outside of that let's go aheadand get started on this stuff.
Meta prompting is going to beanything where you take a higher
level overview at the approach.
So instead of approachingChatGPT at a single point where
(34:34):
you're gonna go, Hey, I want youto build out this carousel
thing.
I want you to do a SEO blogpost.
It's okay to do that.
Then you have one higher levelwhere you can send tax it.
Right now you're gonna go, Iwant the blog post to look like
this.
Or maybe you wanna do like alevel one role play.
These are all beautiful thingsbecause they're really
attributes and characteristicsof how ChatGPT will relate with
(34:55):
you and since actually respond.
That's great.
But what if we take all of thesetraits, put it into one persona
model, shove it at the top, makethat a master model, and then
once we have this level ofmaster model, what we're gonna
end up doing is that once we'vegenerated that, we're gonna load
(35:16):
it with a persona profile So forexample, when you were talking
about the carousel content,right?
The Instagram sell carousel hostthis is what I built out for
that exactly.
So I loaded it with a simplemaster model.
This is a very simple one mindyou.
Eva and I have other things thatare like six post deep.
Now that we have the besteducator in the world that we're
(35:38):
interacting with, ChatGPT hasthis construct of who it should
be, how it should be, how itshould talk.
So why does this matter?
Your time is the most valuablething in the world, right?
So what if I told you couldspend less time on ChatGPT
getting more out of it.
That's what this is.
If you take the time to build amaster model with your specialty
and experiences, because I'm notyou.
(36:00):
You know what your specialthings is?
Maybe you're a coder, maybeyou're a marketer, you know how
you want to define the model torespond to you or to do things,
but you don't have to thinkabout it from such a how do I
want the carousel post?
No, ignore that.
You're defining a person.
If you were to hire a person todo that job for you, how do you
want that person to be?
(36:22):
That's what you build first.
That's your persona model.
Then you secondarily move intowhat you want it to be.
So you give it a job.
You are named builder.
You're gonna create a project.
This project's gonna be thisthing.
So now you have all of thesealiases or variables you could
(36:43):
say to reference back to thethings you want.
If you want to call back theproject, you just say, I'm doing
this project, and Tai will knowexactly what you're talking
about.
So now instead of having toredefine the whole damn thing
every post or do this hugesyntax, programmatic language,
you can get lazier and lazierwhile ChatGPT reinforces itself
and get smarter and smarterbecause you preemptively told
(37:06):
it.
That's what it should do.
In fact, there was a researchpaper that was released on this
called Chat Gcpt plusReflection, which shows the
accuracy of GPT four goes fromsomething like, what is it, 77
to I, I don't wanna misquote it,but it's something like 10 to
20% increase in accuracy.
It's insane levels and if that'swhat's is amazing.
(37:29):
Yeah.
And just realized we haveactually, we just hit the hour.
Real quick before everybody hopsoff, I'm gonna share a quick
poll How often do you want to dothese?
You wanna do them every week?
Do you wanna do them every otherweek?
So if you could answer thatreally quick and Wow.
Very strong reaction for weeklyawesome.
(37:49):
Okay.
Thank you all so much forattending.
This was a ton of fun for me.
I hope it was a fun I hope itwas a lot of fun for you.
This recording will be shared onthe YouTube channel aside from
me wanting you to support thepodcast, also, if you want the
recording and you don't wannasubscribe, the link is in that
document that I put which sure Ican put that again.
(38:11):
This one.
And if this has been helpful foryou please do me a solid and go
subscribe to the podcast inwherever you do it.
Rate it, five stars.
I'm gonna be posting a wholebunch more episodes of people
coming in and havingconversations with them about
like, How do you do promptengineering?
(38:33):
What have you learned, all thatkind of stuff.
Two other things.
If you are a person who writesprompts and would like to be on
the podcasts honestly, just sendme an email, email redacted but
also you can comment in theLinkedIn or reply to any of the
dozen places that I think Iposted this.
Oh one last thing.
The paper about reflection justgot posted in the comments, so
(38:57):
make sure that you grab that.
It is a very cool technique andI'm hopefully going to be
talking about that on anupcoming episode.
And if not, I might just do anepisode myself on it cuz it's
pretty cool.
Awesome.
Thank you all so much and talkto you soon.
Greg Schwartz (39:26):
Thanks for coming to the Prompt Engineering
podcast, the podcast dedicatedto helping you learn how to be a
better prompt engineer, how tosell your prompts if that's
something you're into, or justuse them in your day job.
See you soon.
Popular Podcasts
Stuff You Should Know
If you've ever wanted to know about champagne, satanism, the Stonewall Uprising, chaos theory, LSD, El Nino, true crime and Rosa Parks, then look no further. Josh and Chuck have you covered.
Dateline NBC
Current and classic episodes, featuring compelling true-crime mysteries, powerful documentaries and in-depth investigations. Follow now to get the latest episodes of Dateline NBC completely free, or subscribe to Dateline Premium for ad-free listening and exclusive bonus content: DatelinePremium.com
On Purpose with Jay Shetty
I’m Jay Shetty host of On Purpose the worlds #1 Mental Health podcast and I’m so grateful you found us. I started this podcast 5 years ago to invite you into conversations and workshops that are designed to help make you happier, healthier and more healed. I believe that when you (yes you) feel seen, heard and understood you’re able to deal with relationship struggles, work challenges and life’s ups and downs with more ease and grace. I interview experts, celebrities, thought leaders and athletes so that we can grow our mindset, build better habits and uncover a side of them we’ve never seen before. New episodes every Monday and Friday. Your support means the world to me and I don’t take it for granted — click the follow button and leave a review to help us spread the love with On Purpose. I can’t wait for you to listen to your first or 500th episode!