Episode Transcript
Transcripts are displayed as originally observed. Some content, including advertisements may have changed.
Use Ctrl + F to search
0:15
All right. Looks
0:17
like this should be good to go. Welcome... Neil,
0:20
in particular, I saw your question.
0:22
A few other people had some, but we'll
0:24
start with yours. Yeah,
0:26
thank you. I Actually I, rather than me reading
0:28
it, why don't you just explain in your own words, cause you
0:30
can do a better job.
0:31
Yeah, of course. So I've been trying to use
0:33
the basically the web search functionality
0:35
in ChatGPT to do some
0:38
kind of deeper research into different topics.
0:41
And I'll, some of it has to do with
0:43
what sort of things are trending and
0:45
kind of predictive stuff. And
0:47
when I first saw that, that was
0:49
an option I tried a couple of
0:51
different, techniques to build
0:54
the prompts. The prompts had, persona
0:56
and knowledge and traits and like steps
0:58
to the task and all that. And the first
1:01
attempt that I took at it, I
1:03
really focus in on the steps to the task.
1:05
And I would say start with these
1:08
sources that are more broad,
1:10
then dig into these sources, then dig
1:12
into these sources. Thinking like, how
1:14
would I go about researching something? And
1:17
then I put examples of URLs
1:19
to look at. And then, names of
1:21
specific companies and things like that. And
1:23
it, it didn't seem to do very
1:26
good at all with the URL
1:28
examples. So I pulled those
1:30
out and then it seemed to do better
1:33
without that information, but
1:35
I almost like by giving it maybe 11,
1:38
12 examples, yeah. It didn't
1:40
seem to do so well. I narrowed it down a little
1:42
bit more. It did better. It
1:45
was very odd. And then eventually
1:47
I just, Took the specific
1:49
examples away, made it more broad.
1:51
Only had three examples that
1:53
seemed to do the best, but it seemed like
1:55
it just didn't really go
1:57
super deep each time. It would
1:59
maybe hone in on two or three websites
2:02
and gather information from that and present
2:04
the findings. And I was really
2:06
looking for it to go
2:08
a lot deeper and I couldn't quite figure out, e
2:11
even try and continue exactly where you left
2:13
off or something like that. It would just do
2:15
the task over again and find the same thing. So
2:18
I wasn't sure what other people had done
2:20
and had success with using that capability
2:23
cuz it clearly
2:24
was a lot better than being
2:26
in barred and everything. But just,
2:29
I don't know, I was having a hard time really getting into
2:31
it and getting it to really
2:34
go to the depths
2:34
That I wanted
2:35
it to go to. Got it.
2:37
Okay. Yeah, I actually just
2:39
did a podcast episode
2:42
about this topic interesting.
2:44
Although that was a little more brainstorming
2:46
and a little less just flat out research.
2:48
But what's an example? Just
2:51
lost the video. What's an example of
2:53
the kind of stuff you were researching?
2:57
So it, it had to do with trends
2:59
in 401K plans. Okay.
3:02
Like risk trends. Okay.
3:05
Yeah. I was asking it to, take
3:09
a look at overall kind of
3:11
trends. And then I was
3:13
directing it to specific, like
3:15
law firms and things like that would typically
3:18
write about those trends
3:20
or industry organizations or things
3:22
like that. But, may I
3:25
wish I could ch I not really for compliance for,
3:27
I can't really share my screen or anything, but
3:29
Okay. Maybe something like similar along
3:31
those lines. I'm not sure if someone else has like a
3:33
particular area that they,
3:36
I don't know, maybe even AI trends or something like
3:38
that. Something that we could try to track down and
3:40
see, like we could all look at and go, oh, that's really
3:42
good. I don't know if there's another example that we could
3:44
use. Similar along those
3:46
lines. Yeah, I'm not
3:48
sure. To
3:50
me hearing, hearing like
3:53
risk for retirement funds
3:55
sounds like something that you'd be able to. Maybe
3:58
have to go into a PDF to pull out, but at least it's gonna
4:00
be there as opposed to more
4:02
of an analytical, like what directions
4:04
are AI tools trending in would be a much
4:06
more thought processy.
4:08
Yeah. Yeah. But I guess
4:11
we could start with something like
4:13
I don't know, what are the return
4:16
amounts for the top 10,
4:19
retirement plans? And then start trying
4:21
to dig from there and see, can you give
4:23
me more data? Can you explore?
4:26
Cuz one thing I have noticed is it
4:29
seems like the browsing, at
4:31
least from my experience, is much more go get
4:33
one webpage and analyze it
4:35
as opposed to go give me 30 web
4:37
pages and analyze them. Okay.
4:40
So I wonder if that might be just
4:42
part of it but let's try. So I guess.
4:44
Starting off role wise
4:47
act as a experienced
4:50
financial planner and
4:55
please go retrieve the rate
4:58
of return for the top 10
5:01
investment plans retirement
5:03
plans within
5:06
the US Cause of course that's
5:08
worth clarifying. And
5:10
then I
5:13
think we're gonna need to say yeah,
5:15
ask me questions to generate
5:17
the best output. And
5:22
actually I can already imagine it asking
5:24
what the top 10 retirement plans would
5:26
when the US are. So maybe
5:28
just based on rate of return.
5:32
It's a little recursive, but it might be okay with
5:34
that. And then No,
5:38
actually I just wanna run it from there and see what it comes
5:40
up with cuz. Okay,
5:43
good. I'll look up the most current data
5:47
specific types of retirement plans.
5:53
Oh okay. I feel like some of that's obvious
5:55
and implied, but Sure. Yeah. I
5:57
guess let's go with type as a 401k,
6:01
historical rate of return. Obviously
6:04
not future. That seems
6:06
a little silly to even ask, but I guess it. Maybe
6:08
makes sense. Would you like That would be
6:11
very
6:11
interesting to see what it comes up with cuz
6:13
that's yeah. Easy information
6:15
to, there's publicly available for some plans,
6:17
but it's not very easy to connect all
6:19
the dots to determine
6:21
something like that. I
6:23
think general plans,
6:25
regardless of company. And
6:30
then any specific industries or sectors
6:32
you're interested in? No, please.
6:35
Really just wanted to keep it broad.
6:38
Okay. It's browsing good. What
6:40
is it? Browsing to historical rate
6:42
of return for 401k plans in the us.
6:46
I don't imagine that's gonna give me a very
6:48
good result, but let's see what it does. It's
6:50
also just slow. So this happens to me a lot.
6:53
It tried clicking on it and apparently failed.
6:55
It said click failed. Why did the click fail? I
6:57
don't know. Can I even open this link?
7:00
I can, it's a search
7:03
on Bing for lots of stuff.
7:06
Okay. Okay.
7:09
So now it's going off of a pocket
7:12
sense article and
7:15
then it went back to the search.
7:17
This is a lot better than what I've most
7:20
of the, so it was, what I was
7:22
looking in was if you have, if
7:25
you're like a, an officer of a
7:27
company, you are, you
7:29
might be what is called a fiduciary
7:32
of a re you're like responsible for the retirement
7:35
plan. So that was the lens I was trying
7:37
to get a sense for is what
7:39
are the how good would it do? Looking
7:42
out there and getting a sense for what
7:44
the current trending risk
7:47
areas are, for those individuals.
7:50
And then also trying to get it to think like
7:53
of the stuff that's happening out there,
7:55
what's likely to become
7:58
a more meaningful risk in future years
8:01
that like what's trending and just developing.
8:03
And anytime I would run it like that, I
8:05
definitely would not get that many successful
8:09
clickthroughs on
8:10
anything. It would maybe just
8:12
find one or two articles and then most
8:14
of the time it would say click failed retrying.
8:16
Even in that example, you look in it's I don't know, seven,
8:18
eight websites or something. Yeah.
8:21
Yeah. Honestly,
8:23
this is the most successful clicks I think
8:25
I've gotten and it still didn't get a very
8:28
helpful result. Yeah. It's particularly
8:30
spending a bunch of time on telling what a 401k
8:33
is, which is strange, but,
8:35
okay. Even assuming nine
8:37
and a half percent is accurate, who knows? That's
8:39
it says it's coming from a link. It's
8:42
then not well, okay. Yeah.
8:44
It then even says, I ran out of time
8:47
before I could find information about
8:49
the average on
8:51
sep. Oh, okay. But that's not
8:53
what I was asking for. All right. All right, then let's
8:56
just re-ask the same-ish question.
8:59
Because what I want you to do is
9:02
retrieve the rate of return for the top,
9:05
or actually let's even just say list the
9:07
top 10 retirement plans based on rate
9:09
of return, and then
9:12
see if that does a better job
9:15
refining it. It
9:18
is a little strange to me that it was like, yes,
9:21
I will go answer that, and then it didn't stay on track.
9:25
Yeah. I didn't realize there was a time component
9:29
that it, it was on that, maybe allotted only
9:31
a certain amount of time to
9:32
perform the search. I
9:34
am honestly not surprised. You want it to be
9:38
responsive. Why is it loading
9:40
airline credit cards, cashback,
9:43
credit cards. What in the world? Okay,
9:45
now it's failing to read anything, so
9:49
I wonder what would be a better way
9:51
to, okay let's see if we could prompt
9:54
it for a second here. If I just did top
9:56
10 retirement accounts
9:59
and I guess I should be saying 401k.
10:05
All right, so here is an example
10:07
of one who knows how accurate it is, but
10:09
does this then list the,
10:13
okay, so it lists the average annual return. Great.
10:15
So let's just see if I give it
10:19
still on credit cards. Okay
10:21
we'll see what it comes up with. Based on
10:23
the analysis at here,
10:26
pick the top 10. Interesting.
10:33
Still giving me some of the same
10:35
topic content as before.
10:39
Okay. You
10:41
can stop typing at any time.
10:43
ChatGPT. Wow,
10:47
that's a really weird direction to take. All
10:51
right. How about, I'm tell
10:53
you to go here. Curious in if
10:55
anyone else is, from a
10:57
strategy standpoint, found, it,
10:59
let's say it does look at eight websites.
11:02
A prompt that it's
11:04
okay, now look at eight now. Now find new
11:07
information, but factor
11:09
the pri. Like how many times can you have it,
11:11
do a search and collect information
11:13
and organize that information and then apply
11:17
it in some way through a
11:18
prompt somehow. Yeah,
11:22
I've definitely done data
11:24
analysis where I just go get
11:26
the data and then I throw it in. And
11:28
there's a few tools that I've seen that somewhat
11:31
try and work on that. Wow.
11:33
It really doesn't like that US News
11:35
website. I wonder out
11:37
of curiosity if it's not an H T G P thing.
11:40
All right. So it'll work without that. Stop
11:43
typing.
11:45
It might be a financial thing. I've
11:48
noticed sometimes if I've
11:50
tried, I'm trying to build a stock app
11:53
and yeah, there are some
11:55
blocks. Oh,
11:57
that's interesting. I
12:01
am a little surprised,
12:03
but I guess there could
12:05
be a lot of things that, that open AI
12:07
is saying, Nope, let's not go there. Okay.
12:10
Yeah, it's not htt
12:12
p s that's weird.
12:16
I think it looks like a good search
12:18
because you're asking historical, so
12:20
it should be able to pull that information up. It
12:23
d it doesn't seem like it would be proprietary
12:25
in any way. If you were, you're asking it to do a stock
12:28
picking for future, it
12:30
won't do that. But if you were to do it based
12:32
on history of 401ks, what's
12:35
the average, the best return rate? I would think
12:37
that would be a no-brainer.
12:39
Yeah. I have yet to have any good experiences
12:42
with Bard, but let's just go try that. See
12:44
if it does anything useful. Okay.
12:48
That at least got. That,
12:51
that got something, but I'm not
12:54
sure. Hang on. Okay.
12:58
It did correctly
13:00
pull a bunch of these, it looks like. Nice.
13:03
Okay. Not so much. So
13:05
VTax is 19.1, but
13:07
it doesn't list that. It immediately then goes
13:10
on to Baron Partners,
13:12
which is 18.7, which is the next
13:14
one. But clearly it has just failed at the task
13:18
and actually skips the next one. Two,
13:20
what? That's
13:23
weird. Okay, hang on. Why is
13:25
it coming up with artisan
13:27
midcap, which is 18.4
13:31
and is not mentioned in this article? Okay.
13:34
All right. So yeah this is probably
13:36
just not using that data at all.
13:39
It's pulling it from somewhere else. I
13:42
wonder, does it say no?
13:47
Yeah, so it's not even, it's not
13:49
even loading it. Use the
13:52
article at, to
13:56
pick, like
13:59
I said, I have not had good luck with Bard,
14:01
so I'm not sure. Yeah. Okay. That's
14:04
about what I was expecting from Bard. Okay.
14:08
I wonder if we
14:10
take it the other direction. So
14:12
if we say retrieve
14:17
the best performing
14:19
funds from
14:21
companies, fidelity,
14:26
Vanguard. Baron.
14:29
I assume that's Baron Partners, but I don't actually
14:32
know. Is it gonna actually
14:34
come up with anything useful or is it just gonna
14:37
Okay, it's browsing the web. That's a good sign. Oops.
14:43
Okay, that's a good one to run. Still
14:47
coming to a Forbes
14:51
and then Best Student credit cards. It really likes
14:53
looking at credit cards. Apparently. I don't understand
14:55
this. This is weird. That's really interesting. I
14:59
almost wanna try and do something else like financial,
15:02
but not related to this, just to see if it decides
15:04
to be like, yeah, I'm gonna talk about home loans,
15:07
credit cards again or something. But we
15:09
will wait on that. Oh my God.
15:13
Bunch of bit credit card stuff and then it failed.
15:18
Huh? Let's
15:20
actually throw in a step by
15:23
step here, cuz I'm
15:25
wondering work
15:28
step by step to retrieve
15:32
ooh, this might actually be a good point to try react,
15:34
but let me try this first one,
15:37
retrieve top companies
15:42
two, retrieve five
15:45
funds from each. Three
15:47
retrieve rate of return
15:50
for each fund. Four.
15:52
Compare to select
15:55
top funds by rate
15:57
of return. And let's actually
15:59
just give it to it as a bulleted list
16:02
instead of an in line. All
16:05
right. See it.
16:09
One more try. Wow.
16:12
No, it does not like the, something
16:14
about the way I provided that list is making
16:17
it really unhappy. Let me just refresh the browser.
16:21
Wow. I think it actually just lost that entire
16:25
pro. It did. It lost an entire prompt. Okay,
16:27
let's lemme see if I can recreate that really
16:29
quick. That was this one.
16:32
And then work step
16:35
by step. Retrieve
16:38
tall companies retrieve five of their funds,
16:40
retrieve the rate
16:42
of return for each fund,
16:46
and then provide a
16:48
table sorted by rate
16:50
of return with
16:52
company, with columns company
16:56
fund, rate of
16:58
return. And I think that
17:01
that'd be good cuz that, that's
17:03
something, that's a good example where
17:05
it,
17:06
Browsing. Yeah. Okay. Run.
17:13
Okay. This time it's at least executing.
17:17
And it got the best 401K plans.
17:20
That's something.
17:23
This is good. Cause this is definitely in line with
17:25
it, you gotta look at different areas
17:28
and bring all the information together
17:29
and Yeah,
17:32
and I'm definitely thinking that
17:34
comeback screen share it
17:36
might even be helpful to use
17:39
the react. Okay.
17:43
That's at least a reasonable place to
17:45
look. Because boy, this is
17:48
not getting anywhere fast. I'm
17:51
drawing a blank on this. I've never seen
17:53
it. Just keep dying.
17:56
In lieu of going deeper on that.
17:58
David or Rhonda, do you have anything that you've
18:00
been trying to work on and wondering
18:02
about?
18:05
I've been working on something that
18:09
I put together a kind of a
18:12
slideshow showing
18:14
what I've done and
18:18
if you want, and what better
18:20
it is I thought
18:22
I was reading this book about
18:25
mysteries where
18:29
you're given clues in the mystery.
18:32
And as
18:35
a reader, you're able to figure
18:37
out the mystery on your
18:39
own. But you're
18:41
given the answer at the end. And
18:44
I thought, I wonder if t
18:47
can do something like
18:49
this.
18:50
That's an interesting idea. The reason why I like
18:52
this I'm teacher and I
18:54
would like students to be able
18:57
to try to figure out their
18:59
mystery on
19:01
their own and have this book, it's only
19:03
two pages. Each
19:05
mystery is only like two pages long.
19:08
But anyway it would be clues like
19:10
a guy comes back from the desert
19:14
and he claims that he he found gold
19:17
and that and he this is, and
19:21
he crazy found gold and he came back.
19:23
He He says he was out in the sun
19:25
and he's all canned and everything,
19:30
but he mentions that he just shaved
19:32
the day before. And then the answer
19:34
to the story as well, if
19:36
he'd been out in the desert, if
19:39
he had not shaved, then
19:41
there wouldn't be a sunburn because
19:43
once the, once you
19:45
shave off
19:48
your beard, you wouldn't, you,
19:52
I don't think it would mix, but, so
19:55
it's like there's a clue in the story. You
19:57
can figure it out on your own, kinda. So
20:00
I don't know how you are with time, but I did a quick,
20:03
if it's okay, I did a quick kind
20:05
of summary of the different attempts
20:07
I've made so far to
20:10
try to get ChatGPT to
20:12
do it. And what's interesting
20:14
is to show you. How
20:16
it's failed so far,
20:19
we might have lost you. Yeah,
20:22
we lost him. Okay. Rhonda,
20:24
it sounded like you had something in mind, so go
20:26
for it.
20:28
I just wanted to basically tell you why I'm here
20:30
and then also you might be able to help
20:33
solve some of the problems that I'm having. Building
20:35
a cup and handle financial model
20:38
to attach to, I don't know if you've heard of I b D.
20:40
It's international Business Daily. It's
20:42
a, it's an investor tool
20:45
that people use and has the 20 factors for
20:47
how do you. Check a stock and
20:49
how do you not, you still have to do a lot
20:51
of research. I want my tool
20:53
to build out that final research of a
20:55
cup and a handle, which is a pattern
20:58
that the stock takes. And when it gets
21:00
to the handle, that's when you know it's
21:02
gonna shoot up. You still have to check other
21:04
financials, but I wanna build an app using
21:07
AI that builds on
21:10
these apps that are already out there. I have shaken
21:12
analytics, I have optometry.
21:14
I bought all these apps that have lifetime, I spent
21:17
$5,000 in each and they're worthless because
21:19
I don't have the time just
21:21
in research cuz I have a million projects I wanna do.
21:24
And I'm coming here to you because I
21:26
like the perspective and I like to learn
21:28
about new
21:28
tools. Gotcha. You could
21:30
try dumping some
21:32
of that information in and then, asking
21:34
ChatGPT to analyze it. I don't know how
21:37
I, not knowing anything about the cup
21:39
and whatever the name was, model.
21:41
I have no idea how much data that
21:43
is, if it's reasonable to paste it in or
21:45
not, but Right. It sounds
21:48
like it's something where you could analyze one stock at
21:50
a time. Is that right? You don't need to analyze like 12
21:52
compared to each other? No,
21:54
that's exactly right. And what I'm thinking
21:57
is I would begin my own
21:59
investor club and everybody would join
22:01
and then everybody would take their
22:03
10 favorite stocks, if they could be mid-cap,
22:05
whatever, and then they run my program and
22:08
whoever sees a cup and handle forming would
22:11
then put it into the notes
22:13
for our group and we'd
22:15
all be able to look at it each week or each
22:17
month that we meet. We would get better and better at
22:19
identifying cup and cup and handle and
22:21
hopefully coldly tool enough
22:24
that it would be identified on its
22:26
own.
22:28
Yeah. Thinking about
22:32
the way ChatGPT works, it's not great
22:34
at graphical things, and I don't just mean like analyzing
22:36
an image, graphical reasoning, which is what it sounds
22:39
like you're saying of I'm assuming there's a
22:41
curve this way and a curve another way, and it
22:43
needs to detect that. I was
22:45
thinking though that it could actually even put it into
22:48
an analyzer. Like you could use
22:51
stable diffusion or mid journey and run
22:53
a describe on it and basically just ask,
22:55
is this describing an upward curve or
22:57
a downward curve or, I, I don't know the right
22:59
way to describe what you're talking about, but
23:02
there are some ways you maybe could make that work.
23:04
Okay. So
23:07
I love that.
23:09
You can run algorithms like that, that
23:11
are not really in ChatGPT, but
23:14
because you're talking about like technical analysis
23:16
with the formation of the cup and the handle.
23:18
That might be something to look at where,
23:20
cuz there are a lot of different, even larger
23:23
scale platforms with major
23:25
financial institutions that allow you to build out
23:27
models that detect trading
23:29
patterns and things like that based on whatever your
23:32
your criteria are.
23:33
For the, so that's an open AI seek
23:35
and run.
23:36
It's not an open ai. Is it
23:38
open? Open source?
23:39
Open? Yeah. It, I mean it should be. It should
23:41
be. Yeah. Cuz there's lots, there, there are ways
23:44
to people build, it's
23:46
not exactly charting, but they build trading
23:48
models to respond to tweets and things like that. And
23:51
you can also do visual setups
23:53
where it's double bottoming out
23:55
or hitting a peak or something like that, and
23:57
then, execute a trade or send an alert or something.
23:59
So that might be a route to look at.
24:02
Love it. Okay. Thank
24:04
you.
24:07
Yeah, I'm not familiar with the, any of those, so
24:09
I can't give a suggestion of try
24:11
X, Y, or Z, but I have heard
24:13
those exist as well. Okay.
24:18
All right. I just got the file from
24:20
David.
24:22
It's a little different because you're not saying
24:24
write a mystery story. It could do that very
24:26
easily and it writes your typical,
24:29
a bunch of rich people with British themes
24:32
in a manner all together.
24:34
And, funny, it does it really
24:36
well, it just, And it comes
24:38
up with somebody screams
24:41
and then they all end up in the drawing room
24:43
and he talks to three suspects. And, but
24:46
then what happens is it's
24:48
always here's, oh, here's
24:52
why the person's guilty, but you're not
24:54
involved in trying to figure out why.
24:57
So if you want, I can,
25:00
I have highlighted stuff, but I can explain
25:03
the process I've been through so far. It's been
25:05
interesting.
25:07
Got it. So you're trying to get it.
25:10
Oh, I see. Yeah. So
25:13
it's just not wow. Blackwood
25:15
manner. You're, it's really leaning into the Sherlock
25:18
Homes experience and clue for that matter.
25:20
Yeah. It's
25:22
funny, it'll, it can't put
25:26
clues in the story and then
25:30
Put them in there subtly and
25:33
then hold off and let the reader figure
25:35
it out and then tell you at the end.
25:38
And it, so
25:41
I tried telling it to structure the story.
25:43
First I want an introduction,
25:46
then I want there to be
25:48
a a crime. Then I want
25:51
the interviews, and then I want
25:53
the, what, the reveal at the end. And,
25:56
but make it so that, the
26:01
reader could use critical thinking and
26:03
yeah. Okay. And
26:06
it,
26:06
so it can tell you, it can tell
26:08
you this is what you should do, this is
26:10
how to do it. And it, cuz
26:12
that's what I did. I asked it to, what are the elements
26:15
of a story like that? And it writes 'em all
26:17
down. It tells you everything you're supposed to do
26:20
and then you say, okay, can you do that please?
26:22
And I can, I can't do it.
26:24
That's funny. That's so interesting. Okay.
26:27
Let's see. So
26:31
let me see if ChatGPT four is
26:33
actually working again. Hey,
26:35
it's working again. All right, let me change the screen
26:38
share here. But I get what you're saying
26:40
about Yeah. It just keeps,
26:43
that's hilarious. All right, so
26:46
share this one. Lost
26:51
the video again. There we go. Okay.
26:54
Yay. Finally. Yeah. So what I
26:56
ended
26:56
up doing was asking it just to give
26:58
clues, like just to give me ideas
27:01
for clues and for crimes that, like
27:03
a crime, a clue and a contradiction
27:06
or something, and then put
27:08
it in yourself kind of thing. But
27:12
I was hoping it could do more
27:13
than that. Yeah.
27:15
So let's see. Outline short mystery
27:17
story include the
27:20
plot twists as well
27:22
as major plot
27:24
points for each chapter
27:29
and hints towards the
27:31
plot twists subtle
27:34
hints towards the plot twists.
27:37
Let's not even include any length.
27:40
We'll just go with that. I'm
27:42
not even sure it's gonna follow my
27:44
outline quite near enough. I think I need to give
27:47
it a shot prompt or two, but let's
27:49
see.
27:51
Yeah, the only problem with this kind of problem is there's
27:53
lots of reading.
27:55
Yeah. But I am thinking
27:58
it'll probably be something like every
28:00
chapter you generate by saying, here's
28:02
the outline now generate for chapter
28:04
one and
28:07
then generate chapter two and that kind of thing. Okay.
28:13
So there's a pi, a
28:15
philanthropist, a business tycoon, and a brilliant
28:17
but unassuming scientist. Interesting
28:20
though it didn't actually
28:23
include anything about what the
28:25
mystery is. We've
28:28
got a. Little, it's
28:30
doing the plot twist hint. That's good. So
28:34
I almost think I need to just restart this
28:36
and let's just make it a murder
28:38
mystery, just to make that easier.
28:41
Include the
28:44
actual crime or the crime,
28:47
the plot twist, as well as major points. Stop
28:50
generating for each chapter and subtle hints
28:52
towards the plot twist. Oh,
28:54
okay. Then for each chapter,
29:00
major plot points, one
29:04
plus subtle hint towards the plot twist. Go.
29:06
Let's see if that works better. Suddenly
29:11
it's very fixated on boats. Got
29:13
two boat stories in a row. Okay.
29:17
So it's an elaborate suicide, but we don't actually
29:19
know that. Oh.
29:22
Okay. It's doing it. So the major
29:24
plot point is Martel's body is
29:26
discovered. Detective is called
29:28
to the scene, finds a new
29:31
Van Gogh painting. Subtle hint is
29:34
looking at the last one. Testament, align states
29:36
The art world will find my last piece most breathtaking.
29:39
How does that connect to the Oh,
29:42
cuz. Cuz the plot twist is, he
29:44
faked a murder
29:47
actually killed himself cuz he wants
29:49
to bring attention to a counterfeit van Gogh. That's
29:51
a really weird plot twist, but Okay, sure. Whatever.
29:54
We'll go with the ChatGPT like it. Okay.
29:57
So if we then
30:00
run this as a second prompt,
30:03
which it's still going. So anytime
30:05
you want to finish, that would be great. There
30:09
we go. All right, so now
30:11
actually for clarity, so this
30:14
is the outline generator,
30:16
and now we're gonna make another one that's going to be
30:18
the chapter generator, but I don't need to name
30:21
it that yet. And write
30:23
the, This
30:26
chapter for a murder
30:28
mystery. Oh,
30:31
actually I need to go back and grab
30:33
the first prompt there.
30:36
It's okay. New chat. Here's
30:41
the thing, here's the second thing.
30:44
Okay. So acts as a murder, Mr. Mess.
30:46
Best many bestsellers. Write
30:49
this chapter. Of
30:52
a murder mystery Short story using
30:56
the outline,
30:58
crime, plot twist
31:01
major points for
31:04
chapter, and then subtle
31:07
hints toward the plot twist outline.
31:13
Actually, let's do this way just
31:15
for clarity. Please
31:18
write this chapter in a fluoride entertaining
31:21
way. I don't know. I
31:25
did do, oh man. I'm running this
31:27
on three five. All right.
31:29
I'll run it on four again in a second. At
31:31
least three five's faster. Okay.
31:34
One, one turn of phrase I like. Her
31:37
gaze was immediately drawn to the unfamiliar
31:39
Van Gogh painting laying nearby. The strokes
31:42
of the master artist seem to whisper secrets
31:44
into her ear. Yeah.
31:46
Okay. It's definitely going for fluoride. All
31:49
right, let's again,
31:52
and run this on four. All
31:55
right, so the main thing that I'd
31:57
be concerned about with running it this way
31:59
is then it's going to start
32:02
drifting away from the details.
32:05
So what I would then do is say, for
32:08
chapter one, summarize this,
32:10
and for chapter two, give it the outline
32:13
and then the summary of chapter one. I
32:17
see. To keep it on
32:19
task. Yeah. So it doesn't
32:21
suddenly add a new character in chapter
32:23
three and pretend that they've been on the boat
32:25
the whole time or whatever. It
32:27
can try and keep track of who's there, what
32:30
are they doing, that kind of stuff. Yeah.
32:33
G P T four is much slower. No,
32:35
I'm not able to keep up as fast. But is
32:38
it writing in a style
32:40
that the reader can figure
32:44
out
32:46
in? Seems like it. Guilty person. Okay.
32:49
Yeah. Wow. So there's the
32:51
hint right there. Wait,
32:55
most breathtaking, that's not what it oh,
32:58
no, it was Okay. Nevermind. That
33:00
was the hint. Little
33:04
did Morro know then that the final
33:06
chapters were to be penned by the deceased himself.
33:08
Okay. Yeah. We're definitely getting the tone.
33:11
All right, so summarize
33:14
that chapter into a single
33:17
paragraph. All
33:20
right, let's, that's
33:22
closer, but let me try that again differently. Summarize
33:25
that chapter into a single paragraph to
33:28
containing all the pertinent
33:31
details to carry on
33:34
into chapter two. Let's
33:38
try that. That's
33:41
better. Okay. It's
33:43
including the important bits. It's got the
33:45
quote. Yeah. No surprise there. All
33:48
right, so now the question is, If
33:52
I go back to,
33:56
no, I think it was this one. Oh
34:00
no, that's the three five version. Okay. I
34:05
thought I named it generator.
34:09
Maybe I'm wrong. Come
34:11
on. There we go. Okay, so
34:13
now we'll take
34:15
the same title
34:17
and the chapter
34:20
two info and
34:27
include the same stuff, but
34:31
replace chapter one with chapter
34:33
two. And
34:36
obviously don't need that still, or
34:39
that. There's chapter. There
34:41
we go. Chapter two. Okay. Major
34:43
plot points, blah, blah. Okay,
34:46
outline. Now
34:48
I also need to include using
34:51
the outline and
34:53
the summary of the
34:56
previous chapters. So
34:58
now, and
35:02
where did I put that? There
35:07
we go. Not
35:09
sure this is gonna work, but we can at least try.
35:13
Well,
35:15
I'm learning a lot about how you're. Out
35:17
writing it in for
35:19
it so that the format Yeah.
35:22
Yeah. The like including
35:25
them as H T M L tags instead
35:27
of using variables, more common style.
35:30
I'm mostly just doing, I've seen a
35:32
lot of different ways of doing it. I don't think any
35:34
of them work better than any others, with
35:36
the exception of this one. You're not
35:38
gonna run into problems with quotes cuz
35:41
you know, if you do a normal like var
35:44
chapter two summary equals quote,
35:46
then like, all of the quotes from people
35:48
speaking are gonna be messed up. So
35:51
it, it seems to do a pretty good job of working
35:53
around that. Wow.
35:56
It's immediately jumping to,
36:00
huh? Oh, okay. So
36:03
suggesting its counterfeit is already in this chapter.
36:05
I missed that and
36:07
is, Yep.
36:10
So they're bringing in the new character of Claude
36:13
describing him. Oh,
36:17
I just noticed we're a couple minutes over time. But
36:20
this seems like it's actually getting a decent
36:22
story. I'm curious enough that I'll probably just
36:25
run this a little bit longer and then actually read
36:27
some of it. Cause I'm curious to see what it comes
36:29
up with. It seems very
36:32
I'm having Eric, but I'm
36:35
having
36:35
trouble. Keep keeping up. You're really fast
36:37
and good at this, but do you think
36:39
it's going to make it
36:42
so that when
36:45
you get to the end of the, at the end before
36:48
you. You're told who's
36:50
guilty, you'll be able to use
36:53
the clues in the story to figure out yourself
36:55
if if you're able who
36:57
did it.
37:00
I think it's possible. I think honestly
37:02
we'd need to generate all, yeah. What
37:05
was it? Seven. Oh,
37:07
okay. There's more than seven chapters apparently, but
37:10
at least the seven chapters and kinda get a sense
37:12
of it. Oh
37:14
seven is the conclusion. So then there's
37:16
an epilogue, so yeah.
37:19
Yeah. Won't take too much longer
37:21
to run this process.
37:23
It's a long process cuz you'd have to,
37:26
go through the story yourself and
37:28
see whether or not you can figure it out
37:31
before you read the
37:31
last chapter. And then,
37:34
and
37:34
then find out, one
37:36
of the frustrating things was that it kept
37:39
revealing new information
37:41
at the end that you had no way of knowing.
37:45
Kinda like the way they do that on tv. But anyway
37:47
and then you're yeah,
37:50
cuz my whole goal is that the students,
37:53
enjoy trying to figure
37:55
it out on their own before finding
37:57
out the answer.
38:00
Yeah. Even just reading some of these
38:03
subtle hints, it's not part
38:06
of the idea is that it's looks
38:09
like a murder, but it's actually a
38:11
suicide. I'm
38:13
not seeing stuff that would really
38:15
super call that out. There's
38:17
one down here.
38:20
Yeah. No, wasn't that one?
38:22
Painting is a new thing for him. There it is. He
38:25
apparently was using a brand of paint
38:27
that is toxic. I'm
38:30
not sure that's enough to really help
38:33
you realize. It is a suicide. So I
38:36
guess in some ways I'd say that the whole outline
38:38
is a little bit flawed. This
38:41
is much
38:42
more sophisticated than what I was able to
38:44
get in terms of
38:46
subtlety. Cool.
38:49
So I'll share the prompts
38:51
that I'm using. And actually I think open,
38:53
I added a share. Yep, it did. Can
38:56
you I'll share these as
38:58
well. And
38:59
can you, is you're sharing the prompts and that
39:01
is the showing the syntax, the prompt syntax?
39:03
Okay. All right. That's,
39:06
is that we're
39:07
doing it in the chat. How
39:08
would I, sorry, it's in the chat.
39:11
So that's the outline. And
39:14
that, come
39:18
on. There we go. In fact,
39:20
actually, let me rename these just so I can keep
39:22
them straight too. Outline
39:25
for mystery and then
39:29
generate chapter one and two.
39:33
All right. I
39:36
just put the first one in and the second one
39:38
should be one at
39:40
two.
39:42
Greg, I gotta drop off, but I really appreciate this.
39:45
This was really great to watch you in your element,
39:47
go through and watch your approach and style
39:49
and everything and stuff, and that was really fantastic.
39:52
Thank you.
39:52
It is really great.
39:54
Well, thanks for taking the time to do that. I
39:56
really appreciate it. I hope the you're
39:58
welcome. By the way, if
40:00
I can ask you, do you have, is there any
40:02
book that you recommend reading about
40:05
prompt engineering or
40:07
is it No, I haven't found any
40:09
books yet. I'm actually coming out with
40:11
a course probably next week or
40:13
so. I've been planning this week, but this has
40:15
gotten away from me. Yeah. Thank you all for
40:18
coming.
40:19
Yeah, thank you so much. Really appreciate
40:21
your approach so that people can see
40:23
in real time how things
40:25
progress. That's really great. Yeah. Thank
40:27
you. Thank you so much.
40:29
Thank you for your time. You're welcome. Amazing. Thank
40:31
you. Thank you. Good rest of your week. All
40:33
right. Okay. Bye-Bye. Bye. Okay.
40:36
Bye-bye.
Podchaser is the ultimate destination for podcast data, search, and discovery. Learn More