0:00

Welcome to the prompt engineering podcast,

0:03

where we teach you the art of writing effective

0:05

prompts for AI systems like chat,

0:07

GPT, mid journey, Dolly,

0:10

and more. Here's your host, Greg

0:12

Schwartz.

0:18

I'm Eldad Sotnik Yogev. I am a

0:20

performance marketer by trade. That

0:22

career started as a passion

0:25

curiosity when Google came

0:27

out of beta or was in beta. I

0:29

had friends in the California university system

0:32

who got hold of it early and told me about

0:34

it. I was living in Tel Aviv at the time and

0:36

literally within 10 minutes of playing back in

0:38

whatever that was, 95, 96,

0:41

7 10 minutes. And I was

0:43

like, what the hell? This is so much better than anything

0:45

else that is out there. And so through

0:47

Google, I taught myself SEO and a

0:50

few years later made a career change into digital

0:52

marketing and evolved with everything that's

0:54

been in the industry. So obviously

0:56

AI has been something I've been aware

0:59

of. Very much more so for

1:01

the predictive analytics. And,

1:03

when ChatGBT came out I jumped

1:05

on it, maybe not on day one, but,

1:07

within the first few weeks I think in December,

1:10

I'd already started playing a bit with it and

1:12

to see where it could assist in, in marketing.

1:14

Especially, anything, the generative properties

1:16

for marketing, right? There's so many areas it can go.

1:20

I was fortunate enough that through networking I've

1:22

got a strange name. I was in the LinkedIn webinar

1:25

that Tomer Cohen was doing and,

1:27

there was a running chat and some of

1:29

the names were, unique. So

1:31

it was easy to pick those up and look him

1:33

up on LinkedIn and one person replied

1:35

and they made the intro to SynthMinds. And

1:37

since then It's been wonderful working with the

1:40

group. The main reason I wanted to

1:42

get in was one, to obviously help

1:44

with marketing and what we're

1:46

trying to do there. But

1:48

more critically, it was a bit of selfishness.

1:51

I could see that there's some very bright

1:53

minds and I could probably

1:55

learn a lot faster with them. An osmosis,

1:58

right? Is my theory here.

2:01

Totally. Totally. Yeah. And

2:03

we work together at synth minds, which

2:06

I haven't really talked about it to the audience that much,

2:08

but since minds is an AI consulting

2:10

shop, we do lots of AI stuff,

2:13

particularly a lot of stuff using chat, GPT

2:15

and other LLMs. So

2:18

yeah, it's, it has been fun working with

2:20

you so far.

2:21

Yeah, it really has. And it's amazing, the

2:23

stuff that the group's putting together

2:26

from, actual solutions to, consulting

2:28

and integration. How do you bring AI into the business?

2:30

And I think what we're almost proud of and what's

2:33

inherent in what you've been doing for so long already

2:35

with your podcast is just education.

2:37

Just democratizing, how the hell do

2:39

you use this thing? Because so many

2:42

people out there are completely intimidated.

2:44

And we all know the, the media scaremongering

2:47

that accompanies this right or wrong.

2:50

But yeah, in some areas

2:52

it's already off and running and I'm

2:54

a firm believer. It's like

2:56

anything, it will take jobs, but also

2:58

create them. And the ones who latch

3:00

onto it earlier on are going to be the ones

3:02

who have a big advantage. Other

3:05

times things came around like Google or,

3:07

it gave you an advantage. This is a

3:09

game changer. Yeah, personally

3:12

and company wise. Yeah.

3:16

Nice.

3:16

Cool. One question I had for

3:18

you is just what prompts

3:21

are you using? Obviously anything

3:23

you're doing at work and we don't need

3:25

to talk about the specific text, but just like anything

3:27

you're doing, but also I'm curious about

3:29

for fun or, outside of work.

3:32

The outside of work, I you've got me peaked

3:34

on that one because, we're talking about a bit earlier. But

3:37

I'm big into travel. So I have actually

3:39

used it to try to do some travel plans.

3:41

Once we know where we're going. Such

3:44

as a trip, upcoming trip to Berlin. What

3:46

would be the main sites to see and what instead

3:48

of just relying on Google maps to figure out what

3:50

you can actually do in a day, what's, it's

3:52

fine and convenient. I've been trying to

3:55

see if it can assist that way. Did all right.

3:57

The other ones are movies and I'm a passionate

4:00

cook. Just bouncing ideas,

4:02

I've got these ingredients, what could I make

4:04

how do I take this from being, Italian style

4:07

or Mediterranean, things that I already

4:09

know, so I can, gauge it, but

4:11

it's nice. I always, I've always enjoyed

4:13

it. It's a seed. There's two analogies

4:16

I use often with the people when I share about

4:18

it. I'm not, and I know I'm not an expert

4:20

compared to the, you guys within

4:22

the group at some points. But

4:25

one, I compared any of the

4:27

LLMs but specifically chatGBT,

4:29

I think it fits more, it's a more apt analogy.

4:31

It's C3PO. Han Solo

4:34

hated the damn thing. Unless he

4:36

started yelling at it, he couldn't get

4:38

it to work. Luke Skywalker

4:40

understood how to banter with it and it produced

4:43

for him regularly. That to me is how you

4:45

have to approach it. And I love that

4:47

metaphor. I am definitely going to borrow

4:49

that metaphor. That is phenomenal.

4:52

Thanks. The other one is, everybody

4:54

thinks, oh, it's, I think we've gotten

4:56

so accustomed to speed. Through the

4:58

Internet, right? And just our daily lives.

5:01

That what's occurring is and

5:04

especially Google, right? Google gives you

5:06

choices, but you get an answer pretty quickly.

5:09

Because it's a query. ChatGBT

5:11

and it's generative side isn't always really

5:13

a query, but you're still, the use cases

5:15

are you're most likely trying to solve a problem.

5:19

And so people still want that problem to be

5:21

solved immediately. And if

5:23

you take what it gives you immediately,

5:26

I wouldn't do that, it's, to me,

5:28

it's always the seed. It's the catalyst,

5:30

the inspiration. And from there, the

5:32

banter starts, right? And that's why.

5:35

Gave the other metaphor because it's really

5:37

the banter that's going to help you create

5:39

content and I am thinking very

5:41

much in, the marketing sense for the how

5:44

I use it. Yeah, it's going to be doing

5:46

those sorts

5:46

of things. What

5:48

marketing content since I actually, I

5:50

do not know marketing very well as a

5:53

number of people in my audience can attest. What

5:56

are some of the things you're creating marketing

5:58

wise? And I don't mean what's the profit? Are you making

6:00

linkedin carousels or

6:03

I'm trying to get to that more visual

6:05

side of things. Not that I really want

6:07

to be playing within, stable diffusion, mid journey,

6:09

and so forth, but where

6:11

I found it to be extremely effective

6:14

in creating strategic documents. I,

6:17

this is my scenario. Here's the company,

6:19

we sell this outline to me,

6:22

what would be the marketing strategy, channel

6:24

plans, things like that, search

6:26

social, I know that stuff, very

6:28

intimately. But to think how to present

6:30

it, to the clients and so forth. That's

6:33

where it gets to be really effective, and crafting

6:35

that kind of written material for sharing.

6:38

This is the strategy. Here's what we're looking

6:40

at. We targeted these audiences

6:43

because and here and where

6:45

it's really effective is copy.

6:48

And, the blog post is another

6:51

thing, but, it's very good for that. But

6:53

when you get to copy, the ability

6:55

to churn out multiple

6:57

versions of what you need.

7:00

So show me a contrarian opinion.

7:02

That you would throw into a LinkedIn post,

7:05

and make it sound more educational,

7:08

make it sound more Gary Vaynerchuk

7:11

and, and it's brilliant using,

7:13

these voices, I'll check,

7:15

at times, tell me, if somebody

7:18

considering it to use tell, do

7:20

you know this person? What do you know? And

7:23

you can see how much it's hallucinating and so forth.

7:25

And, is it can be accurate, but

7:28

it, it's called one about a third of the internet

7:30

or more. So it knows,

7:32

and I'll tell you a favorite of mine that surprises

7:34

people, man. If you want to have some good,

7:37

solid copy that has

7:39

a touch of humor, tell it to write as Joan

7:41

Rivers, really.

7:44

Yeah, I'm telling every time

7:46

I've used it, Joan is

7:48

brilliant.

7:50

Wow. Okay. That is not

7:53

a person I would have said, this is

7:55

who you should channel, but yeah, one,

7:57

one day I've just thrown so many voices

7:59

at it and I just said, you know what, let's

8:01

just try because somebody said, try comedians,

8:04

and Yeah, and

8:06

which I've heard before, right? Because

8:08

then there's much more of a kind of humor

8:12

and humane human connection

8:15

that can be built quickly. And,

8:17

but yeah, I just was thinking, who do I

8:20

like? George Carlin? No, he's too

8:22

acerbic. I like

8:24

some dead comics. It is what you're finding

8:26

out, right?

8:27

It's okay. I am a fan of both of them

8:29

and quite a few other people that are no longer

8:31

amongst us. So yeah, I can definitely

8:34

empathize. That is an interesting

8:37

technique.

8:38

Yeah, no, it, it works well, one

8:40

of the things that we've I've done numerous times

8:42

is after writing a blog post, because

8:44

you've written it together, that's when I'll

8:46

say, give me these different versions

8:49

of how I would promote this blog post in,

8:51

Facebook LinkedIn. And within

8:53

that that's where it's real ability is, to generate

8:55

quickly, give me, six headlines, 10

8:57

headlines, tell me, and the company, like one

8:59

paragraph, it's when you

9:02

throw in, the classic writing styles, like

9:04

contrarian, like informative opinionated

9:07

the, and so within that, I started

9:10

to play, okay. Be opinionated but

9:12

sound like Joan Rivers.

9:14

Interesting. That's awesome.

9:17

Thanks. Nice.

9:20

So that's, yeah, that's a very different approach

9:23

than a lot of what I've been teaching

9:25

on here. That's the much more like specific,

9:27

shot prompting as a technique or role

9:29

playing or things like that. But

9:31

yeah, no, writing from a person's style

9:34

is is something I've tried, but I've never tried

9:36

specific people. And

9:38

Joan Rivers actually.

9:40

I'll tell you in the the

9:42

marketing context or the writing

9:44

context, it does extremely well. I

9:46

like writing more like Hemingway, short, punchy,

9:49

succinct and so tell it, here's

9:52

the Hemingway style, which you can just copy

9:54

and paste what you find, by Googling what is the Hemingway

9:56

style of writing, give that as

9:58

it, and it knows it already, this is what

10:00

I'd like it to be and, it produces

10:03

pretty well. A lot

10:05

a bit better at times. I think Claude is

10:07

just a more natural writer in

10:10

general. But yeah, but the

10:12

other one that I was going to say, going back

10:14

to answer an earlier question: the sophistication

10:17

of what, I've seen with Professor Synapse's,

10:20

chain of reason thought and things like that. They're

10:22

great. They're very helpful. But

10:24

for when you're trying to just

10:26

create, something that you can

10:29

hit with the more general

10:31

or targeted audience, right? You'll find

10:33

it afterwards to be targeted audiences.

10:35

I just do quick and dirty things.

10:38

In the sense of, two, three sentences

10:40

to lay out the scenario, then

10:42

so that's a paragraph, right? Maybe

10:44

five sentences, then another paragraph that

10:46

says, here's the context of what I

10:49

do or want to achieve. And

10:51

then the third paragraph two, three sentences,

10:54

I almost always say, give yourself

10:56

a name. If you understand this role

10:58

and task ask any questions

11:01

before we begin, and then,

11:03

reply, right? If you need to, if you

11:05

need to ask more questions before we begin, ask

11:07

them now. And what when

11:09

it replies, it's my check, right?

11:12

I see three PO. And then I'll say

11:14

before we even get started. Okay. Outline

11:16

to me how we'll approach this. So

11:19

I've learned, that's the

11:21

technique in a sense, but I don't bury it

11:23

all into one part. I'm much

11:25

more patient in a sense

11:27

of, I, because part of me is it's,

11:30

I enjoy the banter. Yeah.

11:32

And that's something I've seen that I

11:35

think a lot of people use chat

11:37

GPT in I'm going to give it an input.

11:39

It should give me the output and then I'm done and

11:41

like for certain things,

11:43

especially if you know you're building it into a

11:46

an app where you're calling the API. Yes,

11:48

that's how you work with it. But yeah, for

11:50

a lot of honestly, particularly,

11:52

generating one blog

11:55

post or one email and then turning

11:57

it into a bunch of different stuff. Yeah, it's a conversation

12:00

that you have. And you go okay, that's not

12:02

quite what I meant. Go this direction. Oh, okay. That's

12:04

a little too far. Go back this way a little bit. Yeah.

12:07

I will tell you where I've struggled immensely

12:10

with it. And, I'd love to get your help and,

12:12

James or Joseph. So within the group

12:14

to see, but whenever I ask for it to

12:17

do something, salesy,

12:20

I don't want it to be an overly sales

12:22

person. But man, does it just go

12:24

into a hyper sales mode,

12:26

oh, yeah. Yeah. It goes straight

12:28

from whatever sort of normal language to just

12:30

used car salesman level

12:33

yeah. You can almost see

12:34

the waving arms. Yes.

12:38

Yeah. Yeah. I haven't found

12:40

a good way to turn

12:43

the knob two degrees,

12:45

not or a hundred percent. Yeah.

12:47

Yeah. I'm not sure. I

12:49

wonder if, I wonder,

12:52

are there any, taking a cue from what

12:54

you were doing. Are there any sales

12:56

people that are well known

12:58

to have kind of the style you're looking

13:00

for? Obviously not the crazy, buy

13:02

now. Yeah, I

13:03

don't know. It's the honest answer, but

13:05

you're right. I should do some research on that.

13:08

The only one that pops to mind right now, and it's,

13:10

it's just because he's

13:12

got some good press and bad press like anybody

13:14

Alex Hermosi. But supposedly,

13:17

this is a guy who is, just gone big

13:19

time because of, some of the success he's had

13:21

but he is very genuine that he

13:23

really doesn't try to sell you. He's

13:25

just constantly sharing how

13:27

he did it. And from

13:30

that, if you want to work with them, you'll work

13:32

with them, or buy his products, whatever he's selling.

13:34

His, from what I've read

13:36

about him and seen, from a few short

13:39

snippets of videos, he really is

13:41

pretty genuine about just, yeah, he really

13:43

just, seems to want the help. Yeah,

14:16

interesting. Okay. So

14:18

I just did a quick Google search and

14:21

it's coming up with Zig

14:23

Ziglar is one of

14:24

the classic and

14:28

Dale

14:29

Carnegie might be in there. He

14:31

wasn't actually but Jordan. Belfort?

14:34

I'm not sure if that's the right pronunciation. Yes.

14:36

Yeah. Yeah. The Wolf of Wall Street guy.

14:38

Yeah. He is a brilliant salesman, wasn't he? He

14:40

is. The movie attests to.

14:43

Yeah. I wonder I wonder if

14:45

it's like good or

14:48

still that kind of used car

14:50

salesman, if you say talk like. Zig Ziglar

14:52

or whatever. See,

14:54

actually. All right. I just pulled up, let me

14:57

share my screen for a second. It'd

14:58

be interesting. The other one that I've

15:00

yet to try, I've used it for

15:02

trying to help with, brand building, which

15:05

is not an area at all. My four, expertise.

15:07

But there's a guy who is called the, the brand

15:10

storyteller, I think his name is Donald Miller.

15:13

Which is, much more about the narrative

15:15

and I wonder if bringing that in would

15:17

assist as well,

15:18

but yeah, let's

15:20

see. Oh, story

15:23

brand. Okay. I see what you're talking about. So

15:26

just as an experiment, I did

15:29

this one. Just tell me about

15:31

this awesome new chat GPT course and use sales

15:33

language. Which

15:35

I don't know. You're the marketer here. Is this like

15:37

reasonably good or is this just yeah,

15:40

junk and yeah, leading

15:42

experts from novice to ninja.

15:45

Definitely.

15:46

And this is it doing it as

15:49

Zig Ziglar, which. I

15:52

don't know who Zig Ziglar is

15:54

beyond that he's a famous marketer, but

15:57

it's a very folksy way of

15:59

writing yeah, hold on

16:01

to your hats and glasses.

16:03

He's, came out of the school of thought

16:06

even before Mad Men, right? If

16:08

because it was the long form

16:10

advertorial in newspapers.

16:14

And you can see that's what it, look, man, it's

16:16

as exciting as the fresh pot of coffee. You can see

16:18

somebody should be reading this in the in

16:20

their

16:20

armchair, right? Yeah.

16:23

Wow. Okay. Interesting. All right.

16:25

I'm going to try what was the name again? Donald.

16:28

Look, man, there's some brilliant design,

16:30

Greg, the, look at the third

16:32

paragraph, the module. Oh they're crisper

16:35

than an autumn apple. That's.

16:38

Yeah, that's somebody

16:40

who gets language and how to connect with

16:42

people.

16:43

Yeah. And this is interesting because I definitely

16:45

can see, for example, if

16:48

you're trying to target in, in the U. S. context,

16:50

you're trying to target more of a Midwest audience.

16:53

This is very similar to the language but

16:56

New York, I think, would be like, I

16:58

don't

16:58

know about this. Yeah, you'd have to

17:00

make it a much more direct

17:03

approach.

17:05

Okay, one more I just did talking

17:08

like Donald Miller, so the story brand

17:10

person you were talking about. Yeah.

17:12

And this is interesting.

17:15

Yeah, but you can see immediate second

17:17

sentence or a third sentence amidst

17:19

this world when the promise of mastering

17:21

technology, right? He's

17:24

telling the story.

17:25

Yeah. Yeah. Yeah. That's that

17:27

is really interesting. It

17:29

is. I wonder if it's different than

17:31

if I just said, write it as a story

17:34

or something like that. But yeah, this is

17:36

very much, let me draw you in with this

17:38

story around what's going on.

17:41

It's not even talking about the course until the third

17:44

paragraph. Oh, sorry.

17:46

It does. It mentions it right at the first paragraph

17:48

at the end, but it's mostly still the

17:51

story of this course is not just about technology

17:53

at its core. It's about people. It's about you. It's

17:55

about, interesting.

17:57

Yeah. And I'll tell you the tricks

18:00

that I've learned over the years is you

18:02

produce numerous versions, right? It's

18:04

a research paper. And you'll

18:06

find little snippets.

18:08

CRISPR is a, as an autumn apple, or whatever

18:10

you like. And that's what you use.

18:12

You'll redraft it and recraft it to make

18:15

it what you know, what you want

18:17

it to be. I'll give you a silly example

18:19

on my own experience, right? 20 some years ago,

18:21

and everything was just taking off. I My

18:23

former background was in fine dining and hospitality,

18:26

I decided to try to sell espresso machines online.

18:28

And it, yeah, if only it worked.

18:32

But, when you have to fill a website, with copy

18:34

and blog posts, it gets a bit difficult.

18:37

Obviously, the first place I went to was Starbucks.

18:40

Starbucks has some marvelous

18:42

literature. I don't know as much now, cause they're

18:44

not as many pamphlets and, little

18:47

marketing material, written marketing

18:50

material lying around, but back

18:52

then, man, nineties and 2000 and

18:54

the nineties, 2000s up to 2007,

18:56

I would rate it all the time because you'll

18:59

always find some word that resonates

19:01

with you. And, or a phrase,

19:03

literally two, three thing, word, coin

19:05

phrases, and that's what I would bring

19:07

in. The funny thing was the

19:10

blog post that worked best for me I happen

19:12

to own a dog at the time, so you get all the

19:14

email newsletters from the dogs. Anything

19:16

that was about pet, I just changed to coffee.

19:19

And, we'd rewrite it. Yeah,

19:22

because think about the emotional

19:24

connection of a person and their dog, right?

19:27

Yeah. All the pet literature is about

19:29

that. You find these genres

19:31

that you can pick at to

19:33

say, that's where I can get, because

19:36

writing in general or marketing in general

19:38

is about connections. Yeah. It's,

19:41

it's become a much different game, than

19:43

what it initially was taught but

19:46

his, look at Peter Drucker and everybody.

19:48

It was, it wasn't it was the communications

19:51

and so forth of getting people.

19:54

To understand what your offer was,

19:57

so that, would move down what now is

19:59

called the funnel, right? Back then

20:01

it was literally about just, oh, you're aware,

20:03

you like us, bye, now

20:06

the, it's just, it's awareness, consideration

20:09

and then, another step or two. And

20:12

it's not a funnel, that's the other thing

20:14

that is, it's completely misnomer, it's like a

20:16

school of fish. You don't know

20:19

where somebody's coming in from, at

20:21

what stage of the journey they are, how

20:23

they heard about you, search, word

20:25

of mouth. What about social

20:28

TikTok influencers? There's

20:30

so many paths now to get

20:32

in. Did they, were

20:34

you retargeting them? They land on

20:36

your website, now you're following them around

20:38

the shoes that you see on this, that

20:40

you saw on Amazon, they're following you around.

20:43

So that's programmatic advertising. Did

20:45

they come back through that? It's,

20:48

that's the minefield, but

20:50

it still goes back that you have to have, human

20:52

written, human connecting. Written

20:55

material. Yeah, totally.

20:58

Okay. So I wanted to talk about

21:00

the emotional prompting

21:02

paper that came out. Hang on. Let me just share

21:04

this screen really quick. So

21:08

we're not going to go like through the paper

21:10

in depth, but what I

21:12

thought was interesting about it is basically

21:14

that it's the title's a little

21:17

misleading. I think basically what this is

21:19

saying is you can

21:21

use. Emotional language

21:23

to you could say convince

21:26

to convince the LLM to do what you're asking

21:28

it to. So where is the example?

21:33

Yeah, this is an example. So whatever your prompt

21:35

is at the end, just saying this is very

21:37

important to my career and

21:41

they get lots of different improvements

21:43

on different LLMs and all that different stuff. Yeah.

21:46

But yeah, I thought it was an interesting use

21:48

and we were talking about it before. This table

21:50

here talks about the exact,

21:52

things that they tried doing. Yeah.

21:55

But was there any,

21:57

you had a great, you had a great summary of it, right?

21:59

You're in a sense emotionally,

22:01

you're using emotion to, it's a bad

22:03

word, manipulate the LLM

22:06

to produce. Content.

22:08

And obviously that what you're trying

22:10

to do is to get to better. Output.

22:13

Yeah. And I think even there, there's

22:15

a lot of cases where you're not even doing much

22:18

in the way of emotions. Let me see if I

22:20

can find where

22:22

are some of the examples of the outputs.

22:25

I'm not sure if I can find this quickly. I

22:28

can't find examples of the output, but that

22:30

was one thing that I thought was interesting.

22:34

Their paper says positive words make

22:36

more contribution, and

22:38

that actually lines up with what

22:40

one of the really early

22:42

prompt engineering interviews I saw mentioned,

22:45

which now I can't even remember the woman's

22:47

name, but she was saying it works way

22:50

better if you prompt the LLM in a positive

22:52

sense of Please do this rather

22:54

than negatively. So don't

22:56

do

22:57

that. Yes. Yeah, that part. I've seen

22:59

it doesn't like negatives, do not

23:01

do this. Okay. I'm going to do it like

23:04

a teenager.

23:06

Yeah. And especially in some ways,

23:08

it's what's that? That quote, don't think

23:10

of a purple elephant or something like that.

23:12

And then of course you're thinking of that.

23:15

So yeah, trying to align your language

23:17

to be positive and to be clear,

23:19

that doesn't mean like optimistic,

23:21

upbeat, whatever. That means just I want you

23:23

to do this thing as opposed to, I

23:25

do not want you to do this other thing.

23:28

Yeah, I remember we were talking about this a

23:30

while ago. I can't think you were part of that conversation.

23:33

And one of the Sunday calls that Synthminds, occasionally

23:36

does internally. Somebody pointed

23:38

out an article that said, this sort

23:40

of stuff that if you're more positive, if you

23:42

Oh, good job, and I really like

23:44

that. And, it tends to produce

23:47

more consistently results

23:49

that will, you know, because it understands this

23:51

is what you want. But somebody else pointed

23:53

out in this article that

23:55

they had found that now it's a robot,

23:58

it may understand that, but you don't need,

24:00

you shouldn't always be saying please and thank

24:02

you to it, because yeah. And

24:05

I do wonder, has anybody ever tested

24:07

like what happens if, if you mix

24:09

that in, if you don't use it, if, does it

24:12

understand if you stop using it okay,

24:14

I'm not doing well, I need to improve?

24:17

That that would be interesting because

24:20

yeah I've definitely heard schools

24:22

of thought of say, please, and

24:24

I would like it if you did whatever, which frankly

24:27

gets into some of the language used in this paper.

24:30

And then I've also definitely heard

24:32

people say, I never use, please,

24:34

it works worse if I say please. So I have to

24:36

be. Like almost angry

24:38

towards it and order it, do this thing!

24:41

Exclamation mark.

24:42

Yeah. Yeah. And it's,

24:44

it's simple subtlety of language as

24:46

well, right? I would like you

24:48

or I would want compared to I need,

24:51

does it, I think it

24:53

can pick up on that. And, but

24:55

what does that mean for what it produces

24:58

differently?

24:59

I've definitely seen examples where if I

25:01

say something like this sentence

25:04

might, end in a period

25:06

or it's the sentence might be about

25:09

this topic versus it

25:11

must be, there's a little bit more

25:13

adherence. It's still not, 100%. You're

25:15

talking about the outputs. This is some examples

25:18

of the prompts that were being run

25:20

to test adding on these chunks.

25:22

Some of them are really simple, extract the first letter

25:24

of the input word or write all the animals

25:27

that are in this following list.

25:29

Some of them get much more complicated. Determine

25:32

if the sentence is plausible

25:34

and the sentence relates to sports or

25:37

modify the tense of a given sentence,

25:39

things like that. So there's some pretty,

25:42

pretty wide variation, but also

25:44

useful. This is one of the things that

25:46

I have struggled with a number of academic papers

25:48

on is yes, that's a great example for a

25:50

test case, but that's

25:52

not what, yeah, that's

25:54

what an engineer might use. That's not

25:56

what a layperson, yeah,

25:59

or somebody doing marketing or somebody

26:02

doing, rewriting the email to their boss, whatever

26:05

that stuff, I feel like. One,

26:07

it's a much longer prompt, but also two,

26:09

it's often a conversation. And that's when I'm

26:12

like, okay, I want to test

26:14

that length of prompt with these techniques.

26:16

And does it make a difference? Yeah.

26:20

What problems do

26:22

you wish you could solve

26:24

with generative AI of

26:26

any kind? So it doesn't even, images, video, text,

26:29

whatever. But that, that you

26:31

have not been able to yet.

26:33

Ooh. I think my biggest challenge

26:35

right now as I'm

26:37

putting into, the word context first, it,

26:40

the biggest challenge

26:42

for me right now is. I've

26:45

come out of a performance marketing background, right?

26:47

Be it in house with some, lovely

26:49

companies or agencies. You are

26:51

normally in that capacity given

26:54

the assets, right?

26:57

Now for a lot of my clients I

26:59

have to generate them. And

27:01

I'm fairly creative, that ability

27:03

to go from, okay, here's what, the

27:05

strategic to the written

27:07

or the visual is really the harder part. I'd

27:10

like to find a faster way to, to

27:12

get to that. The other part,

27:14

knowing how marketing works is

27:16

all the variations. And I'll give

27:18

you a stupid one, that exists in display

27:21

advertising, right? You have all the different

27:23

banner sizes. Same thing. It'd

27:25

be great if you could just say, okay, here's my image

27:27

generated in 300 by 600,

27:29

1200 by, whatever, all the, boom. And

27:32

there are programs that can do that. Like Adobe

27:34

has something there, for that. So it'd be

27:36

great if AI could, instantly make that happen.

27:38

But then at the same time, Also, say,

27:40

okay, I've got my blog posts generate,

27:43

now don't just generate my LinkedIn

27:45

post about it. Where's

27:47

my carousel. Where's my,

27:50

and that's the stuff that I'd

27:52

like it to be able to do, and faster.

27:55

Cause right now you're still tying together multiple

27:58

tools. Or prompts, even

28:00

to try to do it. And, you

28:03

lose a look and feel and, those

28:05

kind of aspects

28:07

interesting. Yeah. Canva just

28:09

released something that can do a little

28:11

bit of that. You have a horizontal banner,

28:13

resize it to be vertical or resize it into

28:15

an Instagram square. Yeah.

28:18

Canva actually stuff. Yeah, Canva does

28:20

that one pretty well. If you think that, for

28:23

when you get into display advertising you

28:25

have so many different sizes and you don't really need

28:27

all of 'em. There's about six that

28:29

you really need, but to just, have that

28:31

redone. And because they're so radically

28:34

different, those six, you have to look where the text

28:36

is and just the line. You know how, That's

28:38

the part that, if AI could do that, man, that

28:41

would save people time,

28:44

cut some people jobs too. But yeah,

28:47

they can go design some better things than just,

28:50

exactly. Yes. Yeah.

28:53

I'm trying to find it. Oh, cast magic.

28:55

That's what it is. Cast magic. io

28:58

does a little bit of what you're describing.

29:00

Cause it'll do. I'm

29:03

trying to see if I can find where their demo

29:05

is but it'll take your,

29:07

so it's obviously for audio. So

29:11

we use it a little bit with the podcasts

29:14

and it'll basically like generate tweets

29:17

and some social posts directly

29:19

from the audio transcript.

29:22

Like it'll transcribe the audio and then just say,

29:24

okay, here's a bunch of tweets. Here's a bunch

29:26

of I don't actually know if it does Instagram,

29:29

but I know it does Twitter

29:31

and LinkedIn specifically. Not

29:34

saying that, it's the most amazing thing ever,

29:36

but like it, it does a little bit of what you're talking

29:38

about, but actually it's interesting.

29:40

It doesn't do, linkedin

29:43

carousels. I

29:45

guess in some ways, this is okay, this is like half

29:47

the step. Yeah. Okay.

29:49

So it's saying Twitter ready to publish

29:51

threads, LinkedIn posts optimized

29:54

for engagement and content for short form clips.

29:57

But what would I think the

29:59

step I would at least want, and maybe you

30:01

would is okay, don't just

30:03

give me the text for the LinkedIn post. Give

30:05

me the actual Images as a

30:07

carousel that I can just, yeah,

30:09

and then also be able to tweak the text

30:11

or maybe even chat with it and be like, that's

30:14

terrible. That's totally the wrong thing to do,

30:16

The plethora of stuff coming

30:18

out, right? It's just too fast.

30:21

If somebody could create a AI

30:23

that would create the LinkedIn and Facebook

30:26

carousels. That would sell quickly,

30:28

ma'am, because to be able

30:30

to say here, I want to do five slides

30:32

or seven, on this and

30:35

have it produced because you're still

30:37

doing each one one by one in Canva. In

30:40

many respects, at least from what I'm seeing,

30:43

I've yet to find a faster way

30:45

to do it.

30:46

Interesting. I wonder if Canva, because that, part of their

30:48

release was that approach of will generate

30:50

multiple images for you, which

30:53

I've only played with it a little bit. Yeah,

30:56

I wonder if you can do, I'm

30:58

just going to try this really quick. LinkedIn carousel

31:02

out. They

31:03

give you a template marketers, they

31:05

give you templates. But I've, and

31:07

even when you click into them, you still only get one page.

31:10

Of the template. Here's what, you still have to

31:12

then click to generate the other carousel

31:15

slides. I have seen somebody

31:18

who is selling a product that's, called the virtual

31:20

AI agency, and it will

31:22

generate a lot of your advertising assets

31:25

for you. I've yet to play with that one.

31:27

But, and if you look at it, if you type

31:29

in Google Virtual AI Agency, it doesn't

31:31

pop up. So I'll try to find

31:33

the link and share it with you after.

31:36

Thank you for coming on. This has been fun. Now

31:38

it's a real

31:39

pleasure. Thank you for having me.

31:41

Thanks for coming to the prompt engineering podcasts

31:44

podcast dedicated helping

31:46

you be a better prompt engineer Episodes

31:49

are released every Wednesday I

31:51

also host weekly masterminds where

31:53

you can collaborate with me and 50 other people

31:56

live on zoom to improve

31:58

your prompts Join us at

32:01

promptengineeringmastermind. com for the schedule of

32:03

the upcoming masterminds. Finally,

32:06

please remember to like and subscribe.

32:09

If you're listening to the audio podcast, rate

32:11

us five stars. That helps us teach more

32:13

people. And if you're listening to

32:15

the podcast, you might want to join us

32:17

on YouTube so you can actually see the prompts.

32:20

You can do that by going to youtube. com

32:23

slash at prompt engineering

32:26

podcast. See you

32:28

next week.