June 10, 2023 • 42 mins
A mastermind on ChatGPT Prompt Engineering. We covered:
- Can we get browsing to do analysis?
- Idea for analyzing stocks
- Writing a who-dunit mystery story
Story outline generation prompt:
https://chat.openai.com/share/61e97176-72ce-4ac7-8480-6e85da342975
Chapter generation: https://chat.openai.com/share/4ddc5a8a-6e5d-469c-af4c-64d3889edf3c (chapters 1-3 of 7)
Next mastermind:
To attend the next mastermind (no cost): PromptEngineeringMastermind.com
Stay in touch on:
Youtube: youtube.com/@PromptEngineeringPodcast
Telegram: https://t.me/PromptEngineeringMastermind
LinkedIn: https://www.linkedin.com/groups/14231334
Links:
If you want to up your game selling on PromptBase, here is the analysis of all text-output prompts: https://gregschwartz.gumroad.com Use coupon code "podcast" for 10% off!
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.
Greg (00:15):
All right.
Looks like this should be goodto go.
Welcome...
Neil, in particular, I saw yourquestion.
A few other people had some, butwe'll start with yours.
Yeah, thank you.
I Actually I, rather than mereading it, why don't you just
explain in your own words, causeyou can do a better job.
Neil (00:31):
Yeah, of course.
So I've been trying to use thebasically the web search
functionality in ChatGPT to dosome kind of deeper research
into different topics.
And I'll, some of it has to dowith what sort of things are
trending and kind of predictivestuff.
And when I first saw that, thatwas an option I tried a couple
of different, techniques tobuild the prompts.
(00:54):
The prompts had, persona andknowledge and traits and like
steps to the task and all that.
And the first attempt that Itook at it, I really focus in on
the steps to the task.
And I would say start with thesesources that are more broad,
then dig into these sources,then dig into these sources.
Thinking like, how would I goabout researching something?
(01:16):
And then I put examples of URLsto look at.
And then, names of specificcompanies and things like that.
And it, it didn't seem to dovery good at all with the URL
examples.
So I pulled those out and thenit seemed to do better without
that information, but I almostlike by giving it maybe 11, 12
(01:38):
examples, yeah.
It didn't seem to do so well.
I narrowed it down a little bitmore.
It did better.
It was very odd.
And then eventually I just, Tookthe specific examples away, made
it more broad.
Only had three examples thatseemed to do the best, but it
seemed like it just didn'treally go super deep each time.
(01:59):
It would maybe hone in on two orthree websites and gather
information from that andpresent the findings.
And I was really looking for itto go a lot deeper and I
couldn't quite figure out, eeven try and continue exactly
where you left off or somethinglike that.
It would just do the task overagain and find the same thing.
So I wasn't sure what otherpeople had done and had success
(02:21):
with using that capability
Greg (02:23):
cuz it clearly
Neil (02:24):
was a lot better than being in barred and everything.
But just, I don't know, I washaving a hard time really
getting into it and getting itto really
Greg (02:34):
go to the depths
Neil (02:34):
That I wanted
Greg (02:35):
it to go to.
Got it.
Okay.
Yeah, I actually just did apodcast episode about this topic
interesting.
Although that was a little morebrainstorming and a little less
just flat out research.
But what's an example?
Just lost the video.
What's an example of the kind ofstuff you were researching?
Neil (02:57):
So it, it had to do with trends in 401K plans.
Okay.
Like risk trends.
Okay.
Yeah.
I was asking it to, take a lookat overall kind of trends.
And then I was directing it tospecific, like law firms and
things like that would typicallywrite about those trends or
(03:20):
industry organizations or thingslike that.
But, may I wish I could ch I notreally for compliance for, I
can't really share my screen oranything, but Okay.
Maybe something like similaralong those lines.
I'm not sure if someone else haslike a particular area that
they, I don't know, maybe evenAI trends or something like
that.
Something that we could try totrack down and see, like we
(03:41):
could all look at and go, oh,that's really good.
I don't know if there's anotherexample that we could use.
Similar along those
Greg (03:46):
lines.
Yeah, I'm not sure.
To me hearing, hearing like riskfor retirement funds sounds like
something that you'd be able to.
Maybe have to go into a PDF topull out, but at least it's
gonna be there as opposed tomore of an analytical, like what
directions are AI tools trendingin would be a much more thought
(04:08):
processy.
Yeah.
Yeah.
But I guess we could start withsomething like I don't know,
what are the return amounts forthe top 10, retirement plans?
And then start trying to digfrom there and see, can you give
me more data?
Can you explore?
Cuz one thing I have noticed isit seems like the browsing, at
(04:31):
least from my experience, ismuch more go get one webpage and
analyze it as opposed to go giveme 30 web pages and analyze
them.
Okay.
So I wonder if that might bejust part of it but let's try.
So I guess.
Starting off role wise act as aexperienced financial planner
(04:54):
and please go retrieve the rateof return for the top 10
investment plans retirementplans within the US Cause of
course that's worth clarifying.
And then I think we're gonnaneed to say yeah, ask me
(05:16):
questions to generate the bestoutput.
And actually I can alreadyimagine it asking what the top
10 retirement plans would whenthe US are.
So maybe just based on rate ofreturn.
It's a little recursive, but itmight be okay with that.
And then No, actually I justwanna run it from there and see
(05:40):
what it comes up with cuz.
Okay, good.
I'll look up the most currentdata specific types of
retirement plans.
Oh okay.
I feel like some of that'sobvious and implied, but Sure.
Yeah.
I guess let's go with type as a401k, historical rate of return.
(06:02):
Obviously not future.
That seems a little silly toeven ask, but I guess it.
Maybe makes sense.
Would you like That would bevery
Neil (06:11):
interesting to see what it comes up with cuz that's yeah.
Easy information to, there'spublicly available for some
plans, but it's not very easy toconnect all the dots to
determine
Greg (06:21):
something like that.
I think general plans,regardless of company.
And then any specific industriesor sectors you're interested in?
No, please.
Really just wanted to keep itbroad.
Okay.
It's browsing good.
What is it?
Browsing to historical rate ofreturn for 401k plans in the us.
(06:46):
I don't imagine that's gonnagive me a very good result, but
let's see what it does.
It's also just slow.
So this happens to me a lot.
It tried clicking on it andapparently failed.
It said click failed.
Why did the click fail?
I don't know.
Can I even open this link?
I can, it's a search on Bing forlots of stuff.
(07:06):
Okay.
Okay.
So now it's going off of apocket sense article and then it
went back to the search.
Neil (07:17):
This is a lot better than what I've most of the, so it
was, what I was looking in wasif you have, if you're like a,
an officer of a company, youare, you might be what is called
a fiduciary of a re you're likeresponsible for the retirement
plan.
So that was the lens I wastrying to get a sense for is
(07:39):
what are the how good would itdo?
Looking out there and getting asense for what the current
trending risk areas are, forthose individuals.
And then also trying to get itto think like of the stuff
that's happening out there,what's likely to become a more
meaningful risk in future yearsthat like what's trending and
(08:02):
just developing.
And anytime I would run it likethat, I definitely would not get
that many successfulclickthroughs on
Greg (08:10):
anything.
It would maybe just
Neil (08:12):
find one or two articles and then most of the time it
would say click failed retrying.
Even in that example, you lookin it's I don't know, seven,
eight websites or something.
Yeah.
Greg (08:21):
Yeah.
Honestly, this is the mostsuccessful clicks I think I've
gotten and it still didn't get avery helpful result.
Yeah.
It's particularly spending abunch of time on telling what a
401k is, which is strange, but,okay.
Even assuming nine and a halfpercent is accurate, who knows?
That's it says it's coming froma link.
(08:42):
It's then not well, okay.
Yeah.
It then even says, I ran out oftime before I could find
information about the average onsep.
Oh, okay.
But that's not what I was askingfor.
All right.
All right, then let's justre-ask the same-ish question.
Because what I want you to do isretrieve the rate of return for
(09:04):
the top, or actually let's evenjust say list the top 10
retirement plans based on rateof return, and then see if that
does a better job refining it.
It is a little strange to methat it was like, yes, I will go
answer that, and then it didn'tstay on track.
Neil (09:25):
Yeah.
I didn't realize there was atime component that it, it was
on that, maybe allotted only acertain amount of time to
Greg (09:32):
perform the search.
I am honestly not surprised.
You want it to be responsive.
Why is it loading airline creditcards, cashback, credit cards.
What in the world?
Okay, now it's failing to readanything, so I wonder what would
be a better way to, okay let'ssee if we could prompt it for a
(09:54):
second here.
If I just did top 10 retirementaccounts and I guess I should be
saying 401k.
All right, so here is an exampleof one who knows how accurate it
is, but does this then list the,okay, so it lists the average
(10:14):
annual return.
Great.
So let's just see if I give itstill on credit cards.
Okay we'll see what it comes upwith.
Based on the analysis at here,pick the top 10.
Interesting.
Still giving me some of the sametopic content as before.
(10:39):
Okay.
You can stop typing at any time.
ChatGPT.
Wow, that's a really weirddirection to take.
All right.
How about, I'm tell
Neil (10:53):
you to go here.
Curious in if anyone else is,from a strategy standpoint,
found, it, let's say it doeslook at eight websites.
A prompt that it's okay, nowlook at eight now.
Now find new information, butfactor the pri.
Like how many times can you haveit, do a search and collect
information and organize thatinformation and then apply it in
(11:17):
some way through a
Greg (11:18):
prompt somehow.
Yeah, I've definitely done dataanalysis where I just go get the
data and then I throw it in.
And there's a few tools thatI've seen that somewhat try and
work on that.
Wow.
It really doesn't like that USNews website.
I wonder out of curiosity ifit's not an H T G P thing.
(11:40):
All right.
So it'll work without that.
Stop typing.
Rhonda? (11:45):
It might be a financial thing.
I've noticed sometimes if I'vetried, I'm trying to build a
stock app and yeah, there aresome
Greg (11:55):
blocks.
Oh, that's interesting.
I am a little surprised, but Iguess there could be a lot of
things that, that open AI issaying, Nope, let's not go
there.
Okay.
Yeah, it's not htt p s that'sweird.
Rhonda? (12:16):
I think it looks like a good search because you're
asking historical, so it shouldbe able to pull that information
up.
It d it doesn't seem like itwould be proprietary in any way.
If you were, you're asking it todo a stock picking for future,
it won't do that.
But if you were to do it basedon history of 401ks, what's the
average, the best return rate?
(12:36):
I would think that would be ano-brainer.
Greg (12:39):
Yeah.
I have yet to have any goodexperiences with Bard, but let's
just go try that.
See if it does anything useful.
Okay.
That at least got.
That, that got something, butI'm not sure.
Hang on.
Okay.
It did correctly pull a bunch ofthese, it looks like.
(13:02):
Nice.
Okay.
Not so much.
So VTax is 19.1, but it doesn'tlist that.
It immediately then goes on toBaron Partners, which is 18.7,
which is the next one.
But clearly it has just failedat the task and actually skips
the next one.
Two, what?
(13:23):
That's weird.
Okay, hang on.
Why is it coming up with artisanmidcap, which is 18.4 and is not
mentioned in this article?
Okay.
All right.
So yeah this is probably justnot using that data at all.
It's pulling it from somewhereelse.
I wonder, does it say no?
(13:47):
Yeah, so it's not even, it's noteven loading it.
Use the article at, to pick,like I said, I have not had good
luck with Bard, so I'm not sure.
Yeah.
Okay.
That's about what I wasexpecting from Bard.
(14:08):
Okay.
I wonder if we take it the otherdirection.
So if we say retrieve the bestperforming funds from companies,
fidelity, Vanguard.
(14:28):
Baron.
I assume that's Baron Partners,but I don't actually know.
Is it gonna actually come upwith anything useful or is it
just gonna Okay, it's browsingthe web.
That's a good sign.
Oops.
Okay, that's a good one to run.
Still coming to a Forbes andthen Best Student credit cards.
(14:52):
It really likes looking atcredit cards.
Apparently.
I don't understand this.
This is weird.
That's really interesting.
I almost wanna try and dosomething else like financial,
but not related to this, just tosee if it decides to be like,
yeah, I'm gonna talk about homeloans, credit cards again or
something.
But we will wait on that.
Oh my God.
(15:13):
Bunch of bit credit card stuffand then it failed.
Huh?
Let's actually throw in a stepby step here, cuz I'm wondering
work step by step to retrieveooh, this might actually be a
(15:33):
good point to try react, but letme try this first one, retrieve
top companies two, retrieve fivefunds from each.
Three retrieve rate of returnfor each fund.
Four.
Compare to select top funds byrate of return.
(15:59):
And let's actually just give itto it as a bulleted list instead
of an in line.
All right.
See it.
One more try.
Wow.
No, it does not like the,something about the way I
provided that list is making itreally unhappy.
Let me just refresh the browser.
(16:21):
Wow.
I think it actually just lostthat entire pro.
It did.
It lost an entire prompt.
Okay, let's lemme see if I canrecreate that really quick.
That was this one.
And then work step by step.
Retrieve tall companies retrievefive of their funds, retrieve
(16:41):
the rate of return for eachfund, and then provide a table
sorted by rate of return withcompany, with columns company
fund, rate of return.
And I think that
Neil (17:01):
that'd be good cuz that, that's something, that's a good
example where it,
Greg (17:06):
Browsing.
Yeah.
Okay.
Run.
Okay.
This time it's at leastexecuting.
And it got the best 401K plans.
That's something.
Neil (17:23):
This is good.
Cause this is definitely in linewith it, you gotta look at
different areas and bring allthe information together
Greg (17:29):
and Yeah, and I'm definitely thinking that
comeback screen share it mighteven be helpful to use the
react.
Okay.
That's at least a reasonableplace to look.
Because boy, this is not gettinganywhere fast.
(17:51):
I'm drawing a blank on this.
I've never seen it.
Just keep dying.
In lieu of going deeper on that.
David or Rhonda, do you haveanything that you've been trying
to work on and wondering about?
David (18:05):
I've been working on something that I put together a
kind of a slideshow showing whatI've done and if you want, and
what better it is I thought Iwas reading this book about
(18:25):
mysteries where you're givenclues in the mystery.
And as a reader, you're able tofigure out the mystery on your
own.
But you're given the answer atthe end.
And I thought, I wonder if t cando something like this.
Greg (18:50):
That's an interesting idea.
The reason why I like
David (18:52):
this I'm teacher and I would like students to be able
to try to figure out theirmystery on their own and have
this book, it's only two pages.
Each mystery is only like twopages long.
But anyway it would be clueslike a guy comes back from the
desert and he claims that he hefound gold and that and he this
(19:18):
is, and he crazy found gold andhe came back.
He He says he was out in the sunand he's all canned and
everything, but he mentions thathe just shaved the day before.
And then the answer to the storyas well, if he'd been out in the
desert, if he had not shaved,then there wouldn't be a sunburn
(19:42):
because once the, once you shaveoff your beard, you wouldn't,
you, I don't think it would mix,but, so it's like there's a clue
in the story.
You can figure it out on yourown, kinda.
So I don't know how you are withtime, but I did a quick, if it's
(20:03):
okay, I did a quick kind ofsummary of the different
attempts I've made so far to tryto get ChatGPT to do it.
And what's interesting is toshow you.
How it's failed so far,
Greg (20:19):
we might have lost you.
Yeah, we lost him.
Okay.
Rhonda, it sounded like you hadsomething in mind, so go for it.
Rhonda? (20:28):
I just wanted to basically tell you why I'm here
and then also you might be ableto help solve some of the
problems that I'm having.
Building a cup and handlefinancial model to attach to, I
don't know if you've heard of Ib D.
It's international BusinessDaily.
It's a, it's an investor toolthat people use and has the 20
factors for how do you.
(20:48):
Check a stock and how do younot, you still have to do a lot
of research.
I want my tool to build out thatfinal research of a cup and a
handle, which is a pattern thatthe stock takes.
And when it gets to the handle,that's when you know it's gonna
shoot up.
You still have to check otherfinancials, but I wanna build an
app using AI that builds onthese apps that are already out
(21:11):
there.
I have shaken analytics, I haveoptometry.
I bought all these apps thathave lifetime, I spent$5,000 in
each and they're worthlessbecause I don't have the time
just in research cuz I have amillion projects I wanna do.
And I'm coming here to youbecause I like the perspective
and I like to learn about new
Greg (21:28):
tools.
Gotcha.
You could try dumping some ofthat information in and then,
asking ChatGPT to analyze it.
I don't know how I, not knowinganything about the cup and
whatever the name was, model.
I have no idea how much datathat is, if it's reasonable to
paste it in or not, but Right.
It sounds like it's somethingwhere you could analyze one
(21:49):
stock at a time.
Is that right?
You don't need to analyze like12 compared to each other?
No,
Rhonda? (21:54):
that's exactly right.
And what I'm thinking is I wouldbegin my own investor club and
everybody would join and theneverybody would take their 10
favorite stocks, if they couldbe mid-cap, whatever, and then
they run my program and whoeversees a cup and handle forming
would then put it into the notesfor our group and we'd all be
(22:15):
able to look at it each week oreach month that we meet.
We would get better and betterat identifying cup and cup and
handle and hopefully coldly toolenough that it would be
identified on its own.
Greg (22:28):
Yeah.
Thinking about the way ChatGPTworks, it's not great at
graphical things, and I don'tjust mean like analyzing an
image, graphical reasoning,which is what it sounds like
you're saying of I'm assumingthere's a curve this way and a
curve another way, and it needsto detect that.
I was thinking though that itcould actually even put it into
an analyzer.
(22:49):
Like you could use stablediffusion or mid journey and run
a describe on it and basicallyjust ask, is this describing an
upward curve or a downward curveor, I, I don't know the right
way to describe what you'retalking about, but there are
some ways you maybe could makethat work.
Rhonda? (23:04):
Okay.
So I love that.
Neil (23:09):
You can run algorithms like that, that are not really
in ChatGPT, but because you'retalking about like technical
analysis with the formation ofthe cup and the handle.
That might be something to lookat where, cuz there are a lot of
different, even larger scaleplatforms with major financial
institutions that allow you tobuild out models that detect
(23:29):
trading patterns and things likethat based on whatever your your
criteria are.
Rhonda? (23:33):
For the, so that's an open AI seek and run.
Neil (23:36):
It's not an open ai.
Is it
Greg (23:38):
open?
Open source?
Neil (23:39):
Open?
Yeah.
It, I mean it should be.
It should be.
Yeah.
Cuz there's lots, there, thereare ways to people build, it's
not exactly charting, but theybuild trading models to respond
to tweets and things like that.
And you can also do visualsetups where it's double
bottoming out or hitting a peakor something like that, and
then, execute a trade or send analert or something.
(23:59):
So that might be a route to lookat.
Rhonda? (24:02):
Love it.
Okay.
Thank you.
Greg (24:07):
Yeah, I'm not familiar with the, any of those, so I
can't give a suggestion of tryX, Y, or Z, but I have heard
those exist as well.
Okay.
All right.
I just got the file from David.
David (24:22):
It's a little different because you're not saying write
a mystery story.
It could do that very easily andit writes your typical, a bunch
of rich people with Britishthemes in a manner all together.
And, funny, it does it reallywell, it just, And it comes up
with somebody screams and thenthey all end up in the drawing
(24:43):
room and he talks to threesuspects.
And, but then what happens isit's always here's, oh, here's
why the person's guilty, butyou're not involved in trying to
figure out why.
So if you want, I can, I havehighlighted stuff, but I can
explain the process I've beenthrough so far.
(25:05):
It's been interesting.
Greg (25:07):
Got it.
So you're trying to get it.
Oh, I see.
Yeah.
So it's just not wow.
Blackwood manner.
You're, it's really leaning intothe Sherlock Homes experience
and clue for that matter.
Yeah.
It's
David (25:22):
funny, it'll, it can't put clues in the story and then
Put them in there subtly andthen hold off and let the reader
figure it out and then tell youat the end.
And it, so I tried telling it tostructure the story.
(25:43):
First I want an introduction,then I want there to be a a
crime.
Then I want the interviews, andthen I want the, what, the
reveal at the end.
And, but make it so that, thereader could use critical
thinking and
Greg (26:03):
yeah.
Okay.
And it,
David (26:06):
so it can tell you, it can tell you this is what you
should do, this is how to do it.
And it, cuz that's what I did.
I asked it to, what are theelements of a story like that?
And it writes'em all down.
It tells you everything you'resupposed to do and then you say,
okay, can you do that please?
And I can, I can't do it.
Greg (26:24):
That's funny.
That's so interesting.
Okay.
Let's see.
So let me see if ChatGPT four isactually working again.
Hey, it's working again.
All right, let me change thescreen share here.
But I get what you're sayingabout Yeah.
It just keeps, that's hilarious.
(26:44):
All right, so share this one.
Lost the video again.
There we go.
Okay.
Yay.
Finally.
Yeah.
So what I ended
David (26:56):
up doing was asking it just to give clues, like just to
give me ideas for clues and forcrimes that, like a crime, a
clue and a contradiction orsomething, and then put it in
yourself kind of thing.
But I was hoping it could domore
Greg (27:13):
than that.
Yeah.
So let's see.
Outline short mystery storyinclude the plot twists as well
as major plot points for eachchapter and hints towards the
plot twists subtle hints towardsthe plot twists.
(27:37):
Let's not even include anylength.
We'll just go with that.
I'm not even sure it's gonnafollow my outline quite near
enough.
I think I need to give it a shotprompt or two, but let's see.
David (27:51):
Yeah, the only problem with this kind of problem is
there's lots of reading.
Greg (27:55):
Yeah.
But I am thinking it'll probablybe something like every chapter
you generate by saying, here'sthe outline now generate for
chapter one and then generatechapter two and that kind of
thing.
Okay.
So there's a pi, aphilanthropist, a business
(28:16):
tycoon, and a brilliant butunassuming scientist.
Interesting though it didn'tactually include anything about
what the mystery is.
We've got a.
Little, it's doing the plottwist hint.
That's good.
So I almost think I need to justrestart this and let's just make
(28:38):
it a murder mystery, just tomake that easier.
Include the actual crime or thecrime, the plot twist, as well
as major points.
Stop generating for each chapterand subtle hints towards the
plot twist.
Oh, okay.
Then for each chapter, majorplot points, one plus subtle
(29:04):
hint towards the plot twist.
Go.
Let's see if that works better.
Suddenly it's very fixated onboats.
Got two boat stories in a row.
Okay.
So it's an elaborate suicide,but we don't actually know that.
Oh.
Okay.
It's doing it.
So the major plot point isMartel's body is discovered.
(29:27):
Detective is called to thescene, finds a new Van Gogh
painting.
Subtle hint is looking at thelast one.
Testament, align states The artworld will find my last piece
most breathtaking.
How does that connect to the Oh,cuz.
Cuz the plot twist is, he fakeda murder actually killed himself
(29:49):
cuz he wants to bring attentionto a counterfeit van Gogh.
That's a really weird plottwist, but Okay, sure.
Whatever.
We'll go with the ChatGPT likeit.
Okay.
So if we then run this as asecond prompt, which it's still
going.
So anytime you want to finish,that would be great.
There we go.
(30:09):
All right, so now actually forclarity, so this is the outline
generator, and now we're gonnamake another one that's going to
be the chapter generator, but Idon't need to name it that yet.
And write the, This chapter fora murder mystery.
(30:30):
Oh, actually I need to go backand grab the first prompt there.
It's okay.
New chat.
Here's the thing, here's thesecond thing.
Okay.
So acts as a murder, Mr.
Mess.
Best many bestsellers.
Write this chapter.
(30:52):
Of a murder mystery Short storyusing the outline, crime, plot
twist major points for chapter,and then subtle hints toward the
plot twist outline.
(31:13):
Actually, let's do this way justfor clarity.
Please write this chapter in afluoride entertaining way.
I don't know.
I did do, oh man.
I'm running this on three five.
All right.
I'll run it on four again in asecond.
At least three five's faster.
(31:34):
Okay.
One, one turn of phrase I like.
Her gaze was immediately drawnto the unfamiliar Van Gogh
painting laying nearby.
The strokes of the master artistseem to whisper secrets into her
ear.
Yeah.
Okay.
It's definitely going forfluoride.
All right, let's again, and runthis on four.
(31:55):
All right, so the main thingthat I'd be concerned about with
running it this way is then it'sgoing to start drifting away
from the details.
So what I would then do is say,for chapter one, summarize this,
and for chapter two, give it theoutline and then the summary of
chapter one.
(32:17):
I
David (32:17):
see.
To keep it on
Greg (32:19):
task.
Yeah.
So it doesn't suddenly add a newcharacter in chapter three and
pretend that they've been on theboat the whole time or whatever.
It can try and keep track ofwho's there, what are they
doing, that kind of stuff.
Yeah.
G P T four is much slower.
No,
David (32:35):
I'm not able to keep up as fast.
But is it writing in a stylethat the reader can figure out
Greg (32:46):
in?
Seems like it.
Guilty person.
Okay.
Yeah.
Wow.
So there's the hint right there.
Wait, most breathtaking, that'snot what it oh, no, it was Okay.
Nevermind.
That was the hint.
Little did Morro know then thatthe final chapters were to be
(33:07):
penned by the deceased himself.
Okay.
Yeah.
We're definitely getting thetone.
All right, so summarize thatchapter into a single paragraph.
All right, let's, that's closer,but let me try that again
differently.
Summarize that chapter into asingle paragraph to containing
(33:29):
all the pertinent details tocarry on into chapter two.
Let's try that.
That's better.
Okay.
It's including the importantbits.
It's got the quote.
Yeah.
No surprise there.
All right, so now the questionis, If I go back to, no, I think
(33:57):
it was this one.
Oh no, that's the three fiveversion.
Okay.
I thought I named it generator.
Maybe I'm wrong.
Come on.
There we go.
Okay, so now we'll take the sametitle and the chapter two info
(34:24):
and include the same stuff, butreplace chapter one with chapter
two.
And obviously don't need thatstill, or that.
There's chapter.
There we go.
Chapter two.
Okay.
Major plot points, blah, blah.
(34:44):
Okay, outline.
Now I also need to include usingthe outline and the summary of
the previous chapters.
So now, and where did I putthat?
(35:06):
There we go.
Not sure this is gonna work, butwe can at least try.
Well,
David (35:15):
I'm learning a lot about how you're.
Out writing it in for
Greg (35:19):
it so that the format Yeah.
Yeah.
The like including them as H T ML tags instead of using
variables, more common style.
I'm mostly just doing, I've seena lot of different ways of doing
it.
I don't think any of them workbetter than any others, with the
exception of this one.
You're not gonna run intoproblems with quotes cuz you
(35:41):
know, if you do a normal likevar chapter two summary equals
quote, then like, all of thequotes from people speaking are
gonna be messed up.
So it, it seems to do a prettygood job of working around that.
Wow.
It's immediately jumping to,huh?
(36:02):
Oh, okay.
So suggesting its counterfeit isalready in this chapter.
I missed that and is, Yep.
So they're bringing in the newcharacter of Claude describing
him.
Oh, I just noticed we're acouple minutes over time.
But this seems like it'sactually getting a decent story.
(36:23):
I'm curious enough that I'llprobably just run this a little
bit longer and then actuallyread some of it.
Cause I'm curious to see what itcomes up with.
It seems very I'm having Eric,but I'm having
David (36:35):
trouble.
Keep keeping up.
You're really fast and good atthis, but do you think it's
going to make it so that whenyou get to the end of the, at
the end before you.
You're told who's guilty, you'llbe able to use the clues in the
story to figure out yourself ifif you're able who did it.
Greg (37:00):
I think it's possible.
I think honestly we'd need togenerate all, yeah.
What was it?
Seven.
Oh, okay.
There's more than seven chaptersapparently, but at least the
seven chapters and kinda get asense of it.
Oh seven is the conclusion.
So then there's an epilogue, soyeah.
Yeah.
Won't take too much longer torun this process.
David (37:23):
It's a long process cuz you'd have to, go through the
story yourself and see whetheror not you can figure it out
before you read the
Greg (37:31):
last chapter.
And then, and
David (37:34):
then find out, one of the frustrating things was that it
kept revealing new informationat the end that you had no way
of knowing.
Kinda like the way they do thaton tv.
But anyway and then you're yeah,cuz my whole goal is that the
students, enjoy trying to figureit out on their own before
(37:57):
finding out the answer.
Greg (38:00):
Yeah.
Even just reading some of thesesubtle hints, it's not part of
the idea is that it's looks likea murder, but it's actually a
suicide.
I'm not seeing stuff that wouldreally super call that out.
There's one down here.
Yeah.
(38:21):
No, wasn't that one?
Painting is a new thing for him.
There it is.
He apparently was using a brandof paint that is toxic.
I'm not sure that's enough toreally help you realize.
It is a suicide.
So I guess in some ways I'd saythat the whole outline is a
little bit flawed.
(38:41):
This is much
David (38:42):
more sophisticated than what I was able to get in terms
of subtlety.
Cool.
Greg (38:49):
So I'll share the prompts that I'm using.
And actually I think open, Iadded a share.
Yep, it did.
Can you I'll share these aswell.
And
Rhonda? (38:59):
can you, is you're sharing the prompts and that is
the showing the syntax, theprompt syntax?
Okay.
All right.
That's, is that we're
Greg (39:07):
doing it in the chat.
How
David (39:08):
would I, sorry, it's in the chat.
Greg (39:11):
So that's the outline.
And that, come on.
There we go.
In fact, actually, let me renamethese just so I can keep them
straight too.
Outline for mystery and thengenerate chapter one and two.
(39:33):
All right.
I just put the first one in andthe second one should be one at
two.
Neil (39:42):
Greg, I gotta drop off, but I really appreciate this.
This was really great to watchyou in your element, go through
and watch your approach andstyle and everything and stuff,
and that was really fantastic.
Thank you.
Rhonda? (39:52):
It is really great.
David (39:54):
Well, thanks for taking the time to do that.
I really appreciate it.
I hope the you're welcome.
By the way, if I can ask you, doyou have, is there any book that
you recommend reading aboutprompt engineering or
Greg (40:07):
is it No, I haven't found any books yet.
I'm actually coming out with acourse probably next week or so.
I've been planning this week,but this has gotten away from
me.
Yeah.
Thank you all for coming.
David (40:19):
Yeah, thank you so much.
Really appreciate your approachso that people can see in real
time how things progress.
That's really great.
Yeah.
Thank you.
Thank you so much.
Greg (40:29):
Thank you for your time.
You're welcome.
Amazing.
Thank you.
Thank you.
Good rest of your week.
All right.
Okay.
Bye-Bye.
Bye.
Okay.
Bye-bye.
All right.
Looks like this should be goodto go.
Welcome...
Neil, in particular, I saw yourquestion.
A few other people had some, butwe'll start with yours.
Yeah, thank you.
I Actually I, rather than mereading it, why don't you just
explain in your own words, causeyou can do a better job.
Neil (00:31):
Yeah, of course.
So I've been trying to use thebasically the web search
functionality in ChatGPT to dosome kind of deeper research
into different topics.
And I'll, some of it has to dowith what sort of things are
trending and kind of predictivestuff.
And when I first saw that, thatwas an option I tried a couple
of different, techniques tobuild the prompts.
(00:54):
The prompts had, persona andknowledge and traits and like
steps to the task and all that.
And the first attempt that Itook at it, I really focus in on
the steps to the task.
And I would say start with thesesources that are more broad,
then dig into these sources,then dig into these sources.
Thinking like, how would I goabout researching something?
(01:16):
And then I put examples of URLsto look at.
And then, names of specificcompanies and things like that.
And it, it didn't seem to dovery good at all with the URL
examples.
So I pulled those out and thenit seemed to do better without
that information, but I almostlike by giving it maybe 11, 12
(01:38):
examples, yeah.
It didn't seem to do so well.
I narrowed it down a little bitmore.
It did better.
It was very odd.
And then eventually I just, Tookthe specific examples away, made
it more broad.
Only had three examples thatseemed to do the best, but it
seemed like it just didn'treally go super deep each time.
(01:59):
It would maybe hone in on two orthree websites and gather
information from that andpresent the findings.
And I was really looking for itto go a lot deeper and I
couldn't quite figure out, eeven try and continue exactly
where you left off or somethinglike that.
It would just do the task overagain and find the same thing.
So I wasn't sure what otherpeople had done and had success
(02:21):
with using that capability
Greg (02:23):
cuz it clearly
Neil (02:24):
was a lot better than being in barred and everything.
But just, I don't know, I washaving a hard time really
getting into it and getting itto really
Greg (02:34):
go to the depths
Neil (02:34):
That I wanted
Greg (02:35):
it to go to.
Got it.
Okay.
Yeah, I actually just did apodcast episode about this topic
interesting.
Although that was a little morebrainstorming and a little less
just flat out research.
But what's an example?
Just lost the video.
What's an example of the kind ofstuff you were researching?
Neil (02:57):
So it, it had to do with trends in 401K plans.
Okay.
Like risk trends.
Okay.
Yeah.
I was asking it to, take a lookat overall kind of trends.
And then I was directing it tospecific, like law firms and
things like that would typicallywrite about those trends or
(03:20):
industry organizations or thingslike that.
But, may I wish I could ch I notreally for compliance for, I
can't really share my screen oranything, but Okay.
Maybe something like similaralong those lines.
I'm not sure if someone else haslike a particular area that
they, I don't know, maybe evenAI trends or something like
that.
Something that we could try totrack down and see, like we
(03:41):
could all look at and go, oh,that's really good.
I don't know if there's anotherexample that we could use.
Similar along those
Greg (03:46):
lines.
Yeah, I'm not sure.
To me hearing, hearing like riskfor retirement funds sounds like
something that you'd be able to.
Maybe have to go into a PDF topull out, but at least it's
gonna be there as opposed tomore of an analytical, like what
directions are AI tools trendingin would be a much more thought
(04:08):
processy.
Yeah.
Yeah.
But I guess we could start withsomething like I don't know,
what are the return amounts forthe top 10, retirement plans?
And then start trying to digfrom there and see, can you give
me more data?
Can you explore?
Cuz one thing I have noticed isit seems like the browsing, at
(04:31):
least from my experience, ismuch more go get one webpage and
analyze it as opposed to go giveme 30 web pages and analyze
them.
Okay.
So I wonder if that might bejust part of it but let's try.
So I guess.
Starting off role wise act as aexperienced financial planner
(04:54):
and please go retrieve the rateof return for the top 10
investment plans retirementplans within the US Cause of
course that's worth clarifying.
And then I think we're gonnaneed to say yeah, ask me
(05:16):
questions to generate the bestoutput.
And actually I can alreadyimagine it asking what the top
10 retirement plans would whenthe US are.
So maybe just based on rate ofreturn.
It's a little recursive, but itmight be okay with that.
And then No, actually I justwanna run it from there and see
(05:40):
what it comes up with cuz.
Okay, good.
I'll look up the most currentdata specific types of
retirement plans.
Oh okay.
I feel like some of that'sobvious and implied, but Sure.
Yeah.
I guess let's go with type as a401k, historical rate of return.
(06:02):
Obviously not future.
That seems a little silly toeven ask, but I guess it.
Maybe makes sense.
Would you like That would bevery
Neil (06:11):
interesting to see what it comes up with cuz that's yeah.
Easy information to, there'spublicly available for some
plans, but it's not very easy toconnect all the dots to
determine
Greg (06:21):
something like that.
I think general plans,regardless of company.
And then any specific industriesor sectors you're interested in?
No, please.
Really just wanted to keep itbroad.
Okay.
It's browsing good.
What is it?
Browsing to historical rate ofreturn for 401k plans in the us.
(06:46):
I don't imagine that's gonnagive me a very good result, but
let's see what it does.
It's also just slow.
So this happens to me a lot.
It tried clicking on it andapparently failed.
It said click failed.
Why did the click fail?
I don't know.
Can I even open this link?
I can, it's a search on Bing forlots of stuff.
(07:06):
Okay.
Okay.
So now it's going off of apocket sense article and then it
went back to the search.
Neil (07:17):
This is a lot better than what I've most of the, so it
was, what I was looking in wasif you have, if you're like a,
an officer of a company, youare, you might be what is called
a fiduciary of a re you're likeresponsible for the retirement
plan.
So that was the lens I wastrying to get a sense for is
(07:39):
what are the how good would itdo?
Looking out there and getting asense for what the current
trending risk areas are, forthose individuals.
And then also trying to get itto think like of the stuff
that's happening out there,what's likely to become a more
meaningful risk in future yearsthat like what's trending and
(08:02):
just developing.
And anytime I would run it likethat, I definitely would not get
that many successfulclickthroughs on
Greg (08:10):
anything.
It would maybe just
Neil (08:12):
find one or two articles and then most of the time it
would say click failed retrying.
Even in that example, you lookin it's I don't know, seven,
eight websites or something.
Yeah.
Greg (08:21):
Yeah.
Honestly, this is the mostsuccessful clicks I think I've
gotten and it still didn't get avery helpful result.
Yeah.
It's particularly spending abunch of time on telling what a
401k is, which is strange, but,okay.
Even assuming nine and a halfpercent is accurate, who knows?
That's it says it's coming froma link.
(08:42):
It's then not well, okay.
Yeah.
It then even says, I ran out oftime before I could find
information about the average onsep.
Oh, okay.
But that's not what I was askingfor.
All right.
All right, then let's justre-ask the same-ish question.
Because what I want you to do isretrieve the rate of return for
(09:04):
the top, or actually let's evenjust say list the top 10
retirement plans based on rateof return, and then see if that
does a better job refining it.
It is a little strange to methat it was like, yes, I will go
answer that, and then it didn'tstay on track.
Neil (09:25):
Yeah.
I didn't realize there was atime component that it, it was
on that, maybe allotted only acertain amount of time to
Greg (09:32):
perform the search.
I am honestly not surprised.
You want it to be responsive.
Why is it loading airline creditcards, cashback, credit cards.
What in the world?
Okay, now it's failing to readanything, so I wonder what would
be a better way to, okay let'ssee if we could prompt it for a
(09:54):
second here.
If I just did top 10 retirementaccounts and I guess I should be
saying 401k.
All right, so here is an exampleof one who knows how accurate it
is, but does this then list the,okay, so it lists the average
(10:14):
annual return.
Great.
So let's just see if I give itstill on credit cards.
Okay we'll see what it comes upwith.
Based on the analysis at here,pick the top 10.
Interesting.
Still giving me some of the sametopic content as before.
(10:39):
Okay.
You can stop typing at any time.
ChatGPT.
Wow, that's a really weirddirection to take.
All right.
How about, I'm tell
Neil (10:53):
you to go here.
Curious in if anyone else is,from a strategy standpoint,
found, it, let's say it doeslook at eight websites.
A prompt that it's okay, nowlook at eight now.
Now find new information, butfactor the pri.
Like how many times can you haveit, do a search and collect
information and organize thatinformation and then apply it in
(11:17):
some way through a
Greg (11:18):
prompt somehow.
Yeah, I've definitely done dataanalysis where I just go get the
data and then I throw it in.
And there's a few tools thatI've seen that somewhat try and
work on that.
Wow.
It really doesn't like that USNews website.
I wonder out of curiosity ifit's not an H T G P thing.
(11:40):
All right.
So it'll work without that.
Stop typing.
Rhonda? (11:45):
It might be a financial thing.
I've noticed sometimes if I'vetried, I'm trying to build a
stock app and yeah, there aresome
Greg (11:55):
blocks.
Oh, that's interesting.
I am a little surprised, but Iguess there could be a lot of
things that, that open AI issaying, Nope, let's not go
there.
Okay.
Yeah, it's not htt p s that'sweird.
Rhonda? (12:16):
I think it looks like a good search because you're
asking historical, so it shouldbe able to pull that information
up.
It d it doesn't seem like itwould be proprietary in any way.
If you were, you're asking it todo a stock picking for future,
it won't do that.
But if you were to do it basedon history of 401ks, what's the
average, the best return rate?
(12:36):
I would think that would be ano-brainer.
Greg (12:39):
Yeah.
I have yet to have any goodexperiences with Bard, but let's
just go try that.
See if it does anything useful.
Okay.
That at least got.
That, that got something, butI'm not sure.
Hang on.
Okay.
It did correctly pull a bunch ofthese, it looks like.
(13:02):
Nice.
Okay.
Not so much.
So VTax is 19.1, but it doesn'tlist that.
It immediately then goes on toBaron Partners, which is 18.7,
which is the next one.
But clearly it has just failedat the task and actually skips
the next one.
Two, what?
(13:23):
That's weird.
Okay, hang on.
Why is it coming up with artisanmidcap, which is 18.4 and is not
mentioned in this article?
Okay.
All right.
So yeah this is probably justnot using that data at all.
It's pulling it from somewhereelse.
I wonder, does it say no?
(13:47):
Yeah, so it's not even, it's noteven loading it.
Use the article at, to pick,like I said, I have not had good
luck with Bard, so I'm not sure.
Yeah.
Okay.
That's about what I wasexpecting from Bard.
(14:08):
Okay.
I wonder if we take it the otherdirection.
So if we say retrieve the bestperforming funds from companies,
fidelity, Vanguard.
(14:28):
Baron.
I assume that's Baron Partners,but I don't actually know.
Is it gonna actually come upwith anything useful or is it
just gonna Okay, it's browsingthe web.
That's a good sign.
Oops.
Okay, that's a good one to run.
Still coming to a Forbes andthen Best Student credit cards.
(14:52):
It really likes looking atcredit cards.
Apparently.
I don't understand this.
This is weird.
That's really interesting.
I almost wanna try and dosomething else like financial,
but not related to this, just tosee if it decides to be like,
yeah, I'm gonna talk about homeloans, credit cards again or
something.
But we will wait on that.
Oh my God.
(15:13):
Bunch of bit credit card stuffand then it failed.
Huh?
Let's actually throw in a stepby step here, cuz I'm wondering
work step by step to retrieveooh, this might actually be a
(15:33):
good point to try react, but letme try this first one, retrieve
top companies two, retrieve fivefunds from each.
Three retrieve rate of returnfor each fund.
Four.
Compare to select top funds byrate of return.
(15:59):
And let's actually just give itto it as a bulleted list instead
of an in line.
All right.
See it.
One more try.
Wow.
No, it does not like the,something about the way I
provided that list is making itreally unhappy.
Let me just refresh the browser.
(16:21):
Wow.
I think it actually just lostthat entire pro.
It did.
It lost an entire prompt.
Okay, let's lemme see if I canrecreate that really quick.
That was this one.
And then work step by step.
Retrieve tall companies retrievefive of their funds, retrieve
(16:41):
the rate of return for eachfund, and then provide a table
sorted by rate of return withcompany, with columns company
fund, rate of return.
And I think that
Neil (17:01):
that'd be good cuz that, that's something, that's a good
example where it,
Greg (17:06):
Browsing.
Yeah.
Okay.
Run.
Okay.
This time it's at leastexecuting.
And it got the best 401K plans.
That's something.
Neil (17:23):
This is good.
Cause this is definitely in linewith it, you gotta look at
different areas and bring allthe information together
Greg (17:29):
and Yeah, and I'm definitely thinking that
comeback screen share it mighteven be helpful to use the
react.
Okay.
That's at least a reasonableplace to look.
Because boy, this is not gettinganywhere fast.
(17:51):
I'm drawing a blank on this.
I've never seen it.
Just keep dying.
In lieu of going deeper on that.
David or Rhonda, do you haveanything that you've been trying
to work on and wondering about?
David (18:05):
I've been working on something that I put together a
kind of a slideshow showing whatI've done and if you want, and
what better it is I thought Iwas reading this book about
(18:25):
mysteries where you're givenclues in the mystery.
And as a reader, you're able tofigure out the mystery on your
own.
But you're given the answer atthe end.
And I thought, I wonder if t cando something like this.
Greg (18:50):
That's an interesting idea.
The reason why I like
David (18:52):
this I'm teacher and I would like students to be able
to try to figure out theirmystery on their own and have
this book, it's only two pages.
Each mystery is only like twopages long.
But anyway it would be clueslike a guy comes back from the
desert and he claims that he hefound gold and that and he this
(19:18):
is, and he crazy found gold andhe came back.
He He says he was out in the sunand he's all canned and
everything, but he mentions thathe just shaved the day before.
And then the answer to the storyas well, if he'd been out in the
desert, if he had not shaved,then there wouldn't be a sunburn
(19:42):
because once the, once you shaveoff your beard, you wouldn't,
you, I don't think it would mix,but, so it's like there's a clue
in the story.
You can figure it out on yourown, kinda.
So I don't know how you are withtime, but I did a quick, if it's
(20:03):
okay, I did a quick kind ofsummary of the different
attempts I've made so far to tryto get ChatGPT to do it.
And what's interesting is toshow you.
How it's failed so far,
Greg (20:19):
we might have lost you.
Yeah, we lost him.
Okay.
Rhonda, it sounded like you hadsomething in mind, so go for it.
Rhonda? (20:28):
I just wanted to basically tell you why I'm here
and then also you might be ableto help solve some of the
problems that I'm having.
Building a cup and handlefinancial model to attach to, I
don't know if you've heard of Ib D.
It's international BusinessDaily.
It's a, it's an investor toolthat people use and has the 20
factors for how do you.
(20:48):
Check a stock and how do younot, you still have to do a lot
of research.
I want my tool to build out thatfinal research of a cup and a
handle, which is a pattern thatthe stock takes.
And when it gets to the handle,that's when you know it's gonna
shoot up.
You still have to check otherfinancials, but I wanna build an
app using AI that builds onthese apps that are already out
(21:11):
there.
I have shaken analytics, I haveoptometry.
I bought all these apps thathave lifetime, I spent$5,000 in
each and they're worthlessbecause I don't have the time
just in research cuz I have amillion projects I wanna do.
And I'm coming here to youbecause I like the perspective
and I like to learn about new
Greg (21:28):
tools.
Gotcha.
You could try dumping some ofthat information in and then,
asking ChatGPT to analyze it.
I don't know how I, not knowinganything about the cup and
whatever the name was, model.
I have no idea how much datathat is, if it's reasonable to
paste it in or not, but Right.
It sounds like it's somethingwhere you could analyze one
(21:49):
stock at a time.
Is that right?
You don't need to analyze like12 compared to each other?
No,
Rhonda? (21:54):
that's exactly right.
And what I'm thinking is I wouldbegin my own investor club and
everybody would join and theneverybody would take their 10
favorite stocks, if they couldbe mid-cap, whatever, and then
they run my program and whoeversees a cup and handle forming
would then put it into the notesfor our group and we'd all be
(22:15):
able to look at it each week oreach month that we meet.
We would get better and betterat identifying cup and cup and
handle and hopefully coldly toolenough that it would be
identified on its own.
Greg (22:28):
Yeah.
Thinking about the way ChatGPTworks, it's not great at
graphical things, and I don'tjust mean like analyzing an
image, graphical reasoning,which is what it sounds like
you're saying of I'm assumingthere's a curve this way and a
curve another way, and it needsto detect that.
I was thinking though that itcould actually even put it into
an analyzer.
(22:49):
Like you could use stablediffusion or mid journey and run
a describe on it and basicallyjust ask, is this describing an
upward curve or a downward curveor, I, I don't know the right
way to describe what you'retalking about, but there are
some ways you maybe could makethat work.
Rhonda? (23:04):
Okay.
So I love that.
Neil (23:09):
You can run algorithms like that, that are not really
in ChatGPT, but because you'retalking about like technical
analysis with the formation ofthe cup and the handle.
That might be something to lookat where, cuz there are a lot of
different, even larger scaleplatforms with major financial
institutions that allow you tobuild out models that detect
(23:29):
trading patterns and things likethat based on whatever your your
criteria are.
Rhonda? (23:33):
For the, so that's an open AI seek and run.
Neil (23:36):
It's not an open ai.
Is it
Greg (23:38):
open?
Open source?
Neil (23:39):
Open?
Yeah.
It, I mean it should be.
It should be.
Yeah.
Cuz there's lots, there, thereare ways to people build, it's
not exactly charting, but theybuild trading models to respond
to tweets and things like that.
And you can also do visualsetups where it's double
bottoming out or hitting a peakor something like that, and
then, execute a trade or send analert or something.
(23:59):
So that might be a route to lookat.
Rhonda? (24:02):
Love it.
Okay.
Thank you.
Greg (24:07):
Yeah, I'm not familiar with the, any of those, so I
can't give a suggestion of tryX, Y, or Z, but I have heard
those exist as well.
Okay.
All right.
I just got the file from David.
David (24:22):
It's a little different because you're not saying write
a mystery story.
It could do that very easily andit writes your typical, a bunch
of rich people with Britishthemes in a manner all together.
And, funny, it does it reallywell, it just, And it comes up
with somebody screams and thenthey all end up in the drawing
(24:43):
room and he talks to threesuspects.
And, but then what happens isit's always here's, oh, here's
why the person's guilty, butyou're not involved in trying to
figure out why.
So if you want, I can, I havehighlighted stuff, but I can
explain the process I've beenthrough so far.
(25:05):
It's been interesting.
Greg (25:07):
Got it.
So you're trying to get it.
Oh, I see.
Yeah.
So it's just not wow.
Blackwood manner.
You're, it's really leaning intothe Sherlock Homes experience
and clue for that matter.
Yeah.
It's
David (25:22):
funny, it'll, it can't put clues in the story and then
Put them in there subtly andthen hold off and let the reader
figure it out and then tell youat the end.
And it, so I tried telling it tostructure the story.
(25:43):
First I want an introduction,then I want there to be a a
crime.
Then I want the interviews, andthen I want the, what, the
reveal at the end.
And, but make it so that, thereader could use critical
thinking and
Greg (26:03):
yeah.
Okay.
And it,
David (26:06):
so it can tell you, it can tell you this is what you
should do, this is how to do it.
And it, cuz that's what I did.
I asked it to, what are theelements of a story like that?
And it writes'em all down.
It tells you everything you'resupposed to do and then you say,
okay, can you do that please?
And I can, I can't do it.
Greg (26:24):
That's funny.
That's so interesting.
Okay.
Let's see.
So let me see if ChatGPT four isactually working again.
Hey, it's working again.
All right, let me change thescreen share here.
But I get what you're sayingabout Yeah.
It just keeps, that's hilarious.
(26:44):
All right, so share this one.
Lost the video again.
There we go.
Okay.
Yay.
Finally.
Yeah.
So what I ended
David (26:56):
up doing was asking it just to give clues, like just to
give me ideas for clues and forcrimes that, like a crime, a
clue and a contradiction orsomething, and then put it in
yourself kind of thing.
But I was hoping it could domore
Greg (27:13):
than that.
Yeah.
So let's see.
Outline short mystery storyinclude the plot twists as well
as major plot points for eachchapter and hints towards the
plot twists subtle hints towardsthe plot twists.
(27:37):
Let's not even include anylength.
We'll just go with that.
I'm not even sure it's gonnafollow my outline quite near
enough.
I think I need to give it a shotprompt or two, but let's see.
David (27:51):
Yeah, the only problem with this kind of problem is
there's lots of reading.
Greg (27:55):
Yeah.
But I am thinking it'll probablybe something like every chapter
you generate by saying, here'sthe outline now generate for
chapter one and then generatechapter two and that kind of
thing.
Okay.
So there's a pi, aphilanthropist, a business
(28:16):
tycoon, and a brilliant butunassuming scientist.
Interesting though it didn'tactually include anything about
what the mystery is.
We've got a.
Little, it's doing the plottwist hint.
That's good.
So I almost think I need to justrestart this and let's just make
(28:38):
it a murder mystery, just tomake that easier.
Include the actual crime or thecrime, the plot twist, as well
as major points.
Stop generating for each chapterand subtle hints towards the
plot twist.
Oh, okay.
Then for each chapter, majorplot points, one plus subtle
(29:04):
hint towards the plot twist.
Go.
Let's see if that works better.
Suddenly it's very fixated onboats.
Got two boat stories in a row.
Okay.
So it's an elaborate suicide,but we don't actually know that.
Oh.
Okay.
It's doing it.
So the major plot point isMartel's body is discovered.
(29:27):
Detective is called to thescene, finds a new Van Gogh
painting.
Subtle hint is looking at thelast one.
Testament, align states The artworld will find my last piece
most breathtaking.
How does that connect to the Oh,cuz.
Cuz the plot twist is, he fakeda murder actually killed himself
(29:49):
cuz he wants to bring attentionto a counterfeit van Gogh.
That's a really weird plottwist, but Okay, sure.
Whatever.
We'll go with the ChatGPT likeit.
Okay.
So if we then run this as asecond prompt, which it's still
going.
So anytime you want to finish,that would be great.
There we go.
(30:09):
All right, so now actually forclarity, so this is the outline
generator, and now we're gonnamake another one that's going to
be the chapter generator, but Idon't need to name it that yet.
And write the, This chapter fora murder mystery.
(30:30):
Oh, actually I need to go backand grab the first prompt there.
It's okay.
New chat.
Here's the thing, here's thesecond thing.
Okay.
So acts as a murder, Mr.
Mess.
Best many bestsellers.
Write this chapter.
(30:52):
Of a murder mystery Short storyusing the outline, crime, plot
twist major points for chapter,and then subtle hints toward the
plot twist outline.
(31:13):
Actually, let's do this way justfor clarity.
Please write this chapter in afluoride entertaining way.
I don't know.
I did do, oh man.
I'm running this on three five.
All right.
I'll run it on four again in asecond.
At least three five's faster.
(31:34):
Okay.
One, one turn of phrase I like.
Her gaze was immediately drawnto the unfamiliar Van Gogh
painting laying nearby.
The strokes of the master artistseem to whisper secrets into her
ear.
Yeah.
Okay.
It's definitely going forfluoride.
All right, let's again, and runthis on four.
(31:55):
All right, so the main thingthat I'd be concerned about with
running it this way is then it'sgoing to start drifting away
from the details.
So what I would then do is say,for chapter one, summarize this,
and for chapter two, give it theoutline and then the summary of
chapter one.
(32:17):
I
David (32:17):
see.
To keep it on
Greg (32:19):
task.
Yeah.
So it doesn't suddenly add a newcharacter in chapter three and
pretend that they've been on theboat the whole time or whatever.
It can try and keep track ofwho's there, what are they
doing, that kind of stuff.
Yeah.
G P T four is much slower.
No,
David (32:35):
I'm not able to keep up as fast.
But is it writing in a stylethat the reader can figure out
Greg (32:46):
in?
Seems like it.
Guilty person.
Okay.
Yeah.
Wow.
So there's the hint right there.
Wait, most breathtaking, that'snot what it oh, no, it was Okay.
Nevermind.
That was the hint.
Little did Morro know then thatthe final chapters were to be
(33:07):
penned by the deceased himself.
Okay.
Yeah.
We're definitely getting thetone.
All right, so summarize thatchapter into a single paragraph.
All right, let's, that's closer,but let me try that again
differently.
Summarize that chapter into asingle paragraph to containing
(33:29):
all the pertinent details tocarry on into chapter two.
Let's try that.
That's better.
Okay.
It's including the importantbits.
It's got the quote.
Yeah.
No surprise there.
All right, so now the questionis, If I go back to, no, I think
(33:57):
it was this one.
Oh no, that's the three fiveversion.
Okay.
I thought I named it generator.
Maybe I'm wrong.
Come on.
There we go.
Okay, so now we'll take the sametitle and the chapter two info
(34:24):
and include the same stuff, butreplace chapter one with chapter
two.
And obviously don't need thatstill, or that.
There's chapter.
There we go.
Chapter two.
Okay.
Major plot points, blah, blah.
(34:44):
Okay, outline.
Now I also need to include usingthe outline and the summary of
the previous chapters.
So now, and where did I putthat?
(35:06):
There we go.
Not sure this is gonna work, butwe can at least try.
Well,
David (35:15):
I'm learning a lot about how you're.
Out writing it in for
Greg (35:19):
it so that the format Yeah.
Yeah.
The like including them as H T ML tags instead of using
variables, more common style.
I'm mostly just doing, I've seena lot of different ways of doing
it.
I don't think any of them workbetter than any others, with the
exception of this one.
You're not gonna run intoproblems with quotes cuz you
(35:41):
know, if you do a normal likevar chapter two summary equals
quote, then like, all of thequotes from people speaking are
gonna be messed up.
So it, it seems to do a prettygood job of working around that.
Wow.
It's immediately jumping to,huh?
(36:02):
Oh, okay.
So suggesting its counterfeit isalready in this chapter.
I missed that and is, Yep.
So they're bringing in the newcharacter of Claude describing
him.
Oh, I just noticed we're acouple minutes over time.
But this seems like it'sactually getting a decent story.
(36:23):
I'm curious enough that I'llprobably just run this a little
bit longer and then actuallyread some of it.
Cause I'm curious to see what itcomes up with.
It seems very I'm having Eric,but I'm having
David (36:35):
trouble.
Keep keeping up.
You're really fast and good atthis, but do you think it's
going to make it so that whenyou get to the end of the, at
the end before you.
You're told who's guilty, you'llbe able to use the clues in the
story to figure out yourself ifif you're able who did it.
Greg (37:00):
I think it's possible.
I think honestly we'd need togenerate all, yeah.
What was it?
Seven.
Oh, okay.
There's more than seven chaptersapparently, but at least the
seven chapters and kinda get asense of it.
Oh seven is the conclusion.
So then there's an epilogue, soyeah.
Yeah.
Won't take too much longer torun this process.
David (37:23):
It's a long process cuz you'd have to, go through the
story yourself and see whetheror not you can figure it out
before you read the
Greg (37:31):
last chapter.
And then, and
David (37:34):
then find out, one of the frustrating things was that it
kept revealing new informationat the end that you had no way
of knowing.
Kinda like the way they do thaton tv.
But anyway and then you're yeah,cuz my whole goal is that the
students, enjoy trying to figureit out on their own before
(37:57):
finding out the answer.
Greg (38:00):
Yeah.
Even just reading some of thesesubtle hints, it's not part of
the idea is that it's looks likea murder, but it's actually a
suicide.
I'm not seeing stuff that wouldreally super call that out.
There's one down here.
Yeah.
(38:21):
No, wasn't that one?
Painting is a new thing for him.
There it is.
He apparently was using a brandof paint that is toxic.
I'm not sure that's enough toreally help you realize.
It is a suicide.
So I guess in some ways I'd saythat the whole outline is a
little bit flawed.
(38:41):
This is much
David (38:42):
more sophisticated than what I was able to get in terms
of subtlety.
Cool.
Greg (38:49):
So I'll share the prompts that I'm using.
And actually I think open, Iadded a share.
Yep, it did.
Can you I'll share these aswell.
And
Rhonda? (38:59):
can you, is you're sharing the prompts and that is
the showing the syntax, theprompt syntax?
Okay.
All right.
That's, is that we're
Greg (39:07):
doing it in the chat.
How
David (39:08):
would I, sorry, it's in the chat.
Greg (39:11):
So that's the outline.
And that, come on.
There we go.
In fact, actually, let me renamethese just so I can keep them
straight too.
Outline for mystery and thengenerate chapter one and two.
(39:33):
All right.
I just put the first one in andthe second one should be one at
two.
Neil (39:42):
Greg, I gotta drop off, but I really appreciate this.
This was really great to watchyou in your element, go through
and watch your approach andstyle and everything and stuff,
and that was really fantastic.
Thank you.
Rhonda? (39:52):
It is really great.
David (39:54):
Well, thanks for taking the time to do that.
I really appreciate it.
I hope the you're welcome.
By the way, if I can ask you, doyou have, is there any book that
you recommend reading aboutprompt engineering or
Greg (40:07):
is it No, I haven't found any books yet.
I'm actually coming out with acourse probably next week or so.
I've been planning this week,but this has gotten away from
me.
Yeah.
Thank you all for coming.
David (40:19):
Yeah, thank you so much.
Really appreciate your approachso that people can see in real
time how things progress.
That's really great.
Yeah.
Thank you.
Thank you so much.
Greg (40:29):
Thank you for your time.
You're welcome.
Amazing.
Thank you.
Thank you.
Good rest of your week.
All right.
Okay.
Bye-Bye.
Bye.
Okay.
Bye-bye.
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!