0:30

So my name is Lee Chazen and

0:32

I was originally a teacher.

0:34

I've worked in a lot of different professions. I

0:36

got into content strategy because

0:39

a friend told me about seven or eight years ago

0:41

that what I was doing is actually content

0:43

strategy and I wasn't familiar with the term back

0:45

in 2013 or 14. Did

0:48

some work in Silicon Valley

0:50

as a chief content strategist and

0:52

now I do consulting on my own. And

0:55

when ChatGPT came around

0:57

about two months ago, I said, oh my

0:59

God, this is gonna change everything. And so I pivoted

1:02

pretty hard into prompt

1:04

engineering and I I think

1:06

I read some article in the New York Times

1:09

that let me know that this was going to be huge

1:11

and it was okay to be a creative

1:14

liberal arts sort of person and

1:16

do this work like you didn't

1:18

have to be necessarily technical

1:21

or have a background in computer science or

1:23

programming, and it's working just

1:26

fine.

1:27

That's awesome. What

1:29

would you say were the first steps

1:31

that you took to learn prompt

1:33

engineering?

1:35

I think it was two and a half months ago I spent

1:37

the next three or four days just experimenting

1:39

with prompts, any type of prompt

1:41

I could think of. I realized

1:44

some people were going a weird direction with

1:46

this. Let's try to undermine

1:48

the system in some way. Let's let's

1:50

try to fool it into being something that it's not.

1:53

And with me, I was mainly

1:55

trying to get it to finish written

1:57

products that I had started,

2:00

but I just needed that extra kind of boost,

2:02

that personal editor. So

2:05

my first, yeah, my first few

2:07

prompts were I just need help

2:09

with my content. And

2:13

from there I just thought, wow, I don't have enough

2:15

hours in the day. I wanna do this all day long.

2:18

That's amazing. Just give us a little more sense

2:20

of, when you say helping with content,

2:23

what kind of content? Like books or podcasts

2:25

or poems, song lyrics. I

2:27

don't know.

2:28

Yeah all of that. And also scheduling,

2:30

for example, today I wrote down

2:33

all the major categories

2:36

of things I need to get done in a day, And

2:38

then I could easily just create a prompt saying,

2:41

All right. I need some creative time, some administrative

2:43

time, some networking time

2:46

and then some promotional time for my consulting

2:48

operation. I went on and on and I gave each

2:50

one a percentage. I could create a prompt

2:53

now that would say

2:55

divide my day, and divide my week into

2:57

segments according to these percentages.

2:59

And tell me. How I can accomplish

3:02

all these things in a day, and I will get that

3:04

as a response. So there's a lot of just

3:06

real practical, functional stuff

3:09

that I was doing initially, So

3:11

if you're like a content strategist

3:13

a teacher, anyone who puts up website

3:15

content or social media content that,

3:18

those were like the first things that I started doing.

3:20

For example, as a as a former teacher

3:23

who became what they call an preneur,

3:25

so I produce educational products. And

3:28

things, and I needed

3:30

to finish a lot of that. And

3:32

one day sitting there, and this is the prompt

3:34

that I wanted to share with people I thought

3:37

what is the, just the coolest,

3:40

most amazing, prompt I can come

3:42

up with that will solve. Like

3:45

a teacher's problems. And

3:47

what I came up with was, now it's over 500

3:49

words, but initially it was 447

3:51

words. And it was designed

3:54

based around this game that I

3:56

had invented with my students.

3:58

Back when I was teaching social studies, we invented this game

4:01

called Global Challenge. The idea was

4:03

to learn the contents of an entire world

4:05

history textbook, all 800 pages

4:07

in the form of a game. But in order

4:09

to do that, you need game questions.

4:11

So the prompt was and

4:14

I'm not gonna read the whole thing now because it would take

4:16

too long, but it was,

4:19

in the role, you're this omniscient author

4:22

master of all content related

4:24

to world history, current events, high school curriculum.

4:26

So I gave it a role, that, that meta promptt.

4:30

And then I got into, you're gonna design

4:32

20 questions, seven categories,

4:34

six different levels of learning to cover

4:36

the span of recorded human history.

4:40

And it's, at first it told me that

4:42

it can't do this all in the

4:44

character limit, but I would have to just

4:46

keep hitting continue

4:48

and let me pause you for a second. That is actually

4:50

one helpful trick for people

4:53

when it runs out of character output, as

4:55

it so often does. You

4:57

can tell it, 18 different ways,

4:59

but the simplest one is just continue

5:02

and it will usually mind you pick

5:04

up where it left off.

5:06

Yeah, and I just thought, how

5:08

much time do I have to keep hitting continue

5:10

because this is going to create 400

5:14

multiple choice and short answer questions.

5:17

Across the span of different intelligence

5:19

types. Cause I built that into the prompt too.

5:22

So there'd be something for a math logic

5:24

kind of person, a visual spatial

5:26

sort of person. And,

5:29

this could revolutionize everything. I

5:31

don't know what textbook publishers are

5:33

thinking right now, but this

5:36

could replace the textbook

5:38

because the corpus text, the body of all

5:40

content that. Goes into ChatGPT

5:43

will cover pretty much

5:46

everything. It's all been

5:48

if it's digital, if it's put out there in some form

5:50

on the internet apparently ChatGPT

5:52

will find it

5:53

Particularly with this prompt, have you run into

5:56

problems with hallucination?

5:57

Yeah. That can be a problem. And

6:00

so you have to be

6:02

somewhat knowledgeable to

6:05

do this to begin with because you're

6:07

gonna need to fact check things and.

6:12

I don't know what the percentage is, the accuracy

6:14

percentage, but my initial guess

6:16

is like 95% or 97%

6:19

accurate, just based on what I've seen so far.

6:22

But here's the kind of the flaw of

6:24

being a human being is that when

6:26

something sounds authoritative

6:28

and. Is using a certain kind of

6:30

vocabulary, you just automatically think

6:33

it is correct,

6:38

and so there's a possibility that disinformation

6:40

misinformation can

6:43

happen. You just gotta be on the lookout

6:45

for it.

6:47

That makes sense. Walk us through this output.

6:49

So I had to create a point

6:51

value based on the

6:54

the level of learning so that if it was like, so

6:56

there's this thing in education called Bloom's Taxonomy

6:59

of Learning, where if you do the

7:01

simplest thing where you just recall some information,

7:04

That's like a level one. When

7:06

you're up at the top and synthesizing

7:08

it and applying it, evaluating

7:10

the information, manipulating it into something different,

7:13

that should be the highest point value because now you're

7:15

really doing some heavy thinking. You'll

7:18

see in the right column there, it'll say

7:20

Bloom's level. And the number

7:22

of points they're gonna get. So automatically

7:25

this thing has, I can't even

7:27

tell you how long this would take a teacher to

7:29

do. And this took when I was initially

7:31

doing with the students and students wrote the questions

7:33

and we created the point values and the charts and everything.

7:36

This was like a two month long project.

7:39

Wow, two

7:41

months.

7:42

To go through and

7:44

curate or cull whatever

7:46

the word is, all of the knowledge

7:48

from the textbook and turn it into questions.

7:50

But here's the cool thing, this is why

7:52

I think teachers shouldn't be afraid of ChatGPT

7:55

because I think the question is more

7:57

important than the answer. And if students can form

7:59

good questions and do this themselves and

8:01

create great prompts, then

8:04

that means they're learning. If

8:06

you know what to ask about a subject, it means

8:08

you know the subject. If you don't know what to ask,

8:11

then you haven't read enough.

8:15

That totally makes sense, and that's a nice

8:17

way to flip that discussion people have

8:19

of, every time people freak out

8:21

about a technology and education, it's

8:24

that the answers are provided. It's calculators,

8:26

it's Wikipedia, it's ChatGPT,

8:29

but thinking about you

8:31

need to be able to ask the right question.

8:34

That's really a good point cuz with a calculator,

8:36

yeah, it'll tell you whatever you want and it'll

8:38

do it more accurately than ChatGPT will, but

8:41

if you don't know how to ask the right question, it doesn't matter.

8:44

Yeah. And what the effect that I think this is going to

8:46

have on people using ChatGPT and other

8:49

LLMs is that

8:51

it's going to improve our thinking.

8:53

It really will because it'll make you ask

8:56

better questions, which means you're gonna be thinking

8:58

at a higher level. You're

9:00

gonna want to be refining these

9:02

prompts. To get better

9:04

answers. So you're not gonna be caught up in the technical

9:06

side, the coding, the getting the

9:09

technology to work for, you're gonna be caught up in the

9:11

how do I take advantage of this superior

9:13

technology to ask Most

9:16

fascinating purposeful,

9:19

meaningful, whatever question. And

9:21

so I did another one like this for just general

9:24

content for any,

9:27

anyone like running a business. Or

9:29

a content strategist or people working in

9:31

social media that I

9:34

generated a question, so that

9:37

ChatGPT would ask the user

9:39

a series of questions. And in getting

9:42

all those answers, it

9:44

will then create a prompt. So the purpose

9:46

is to, you're writing a prompt to

9:48

create a prompt and

9:50

so based on that, they're gonna have a prompt that they can use

9:52

forever whenever their

9:55

company or the idea changes, all

9:57

they have to do is answer the questions that ChatGPT

9:59

has generated from that original prompt.

10:03

And it will give you all

10:05

of the answers or all of the

10:07

content you need for a website, for social media,

10:09

for a book, whatever it is you're working on. The

10:12

idea was once you have all this

10:15

knowledge, now create content

10:18

for the website. Give me a list of the

10:20

top 50 key words, for seo

10:22

so I can use that in social media, use that on

10:24

the site. Or

10:27

the actual product itself, because

10:30

I could have it build out all those different sections

10:32

into the entire book that explains

10:36

that, that explains the whole concept. And

10:38

so in about 30 days time, like that business

10:40

is formed and ready to go, maybe

10:44

sooner. If I imagine if I had a team

10:46

of three or four prompt engineers, we could finish

10:49

any. Major product

10:51

like this and just, all right guys, we've got two

10:53

weeks to do a book, a website,

10:56

and put a product out.

11:00

Very cool. So specifically,

11:03

let's go back to this first prompt.

11:06

Can you call out some of the techniques

11:08

that you're using in this prompt, just

11:11

so the audience can understand and see how they're

11:13

being applied.

11:14

I don't have to speak perfectly

11:17

as I'm speaking to a person. I

11:19

just have to get it all of the information

11:21

I want in there. So up top I say, this

11:23

is for a world history game called Global Challenge

11:26

2.0 Metamorphosis.

11:28

Then I explained what the game is. Then

11:31

I said, all right, now we're gonna move on to the

11:33

questions for this grade level, and

11:37

they should move progressively. Meaning I wanted

11:39

to start at the beginning of the book and

11:42

go 20 questions per chapter

11:44

all the way through. So

11:47

I gave it the order of operations. That,

11:50

so each thing you're drilling down what do I need

11:52

next in this process?

11:54

Now I need them to be divided into seven

11:56

categories. And here's

11:58

what those categories are like.

12:00

It needs to divide it into major

12:03

events, vocabulary, people, geography,

12:05

government, current events and trivia.

12:08

And then I said within those 20 questions,

12:10

so I keep drilling down generate

12:13

five questions in category one. Three

12:15

in categories two, three, and four.

12:19

So I break it down like, how do I want

12:21

those 20? And the remarkable thing is that

12:23

it did it perfectly on the first

12:25

run. And I did not expect that.

12:27

It really blew my mind. I've

12:29

had my mind blown so many times in

12:32

the last two months by things. And

12:34

Oh, and then I gave it specific instructions regarding

12:36

the current events because it's not enough to

12:39

just ask a student a current event question

12:41

about what's happening now. I wanted them

12:43

to, wherever possible,

12:45

I wanted ChatGPT to relate that

12:47

question to something that was happening during

12:50

that period of history that we're studying. And

12:52

then I gave it an example, which

12:55

surprisingly helps, if

12:57

you give ChatGPT, an example

12:59

of what you want, it

13:02

somehow models that.

13:09

Yeah, so that technique is called

13:12

shot prompting and not

13:14

doing it is called Zero shot Prompting cuz there's

13:16

no shot. But one shot

13:18

prompting is giving a single example and

13:20

few, or n there's a lot of different

13:22

terms for it, but either of those is when

13:25

you give it multiple examples where

13:27

in this Wherein this is the example.

13:29

I see where you said connect

13:31

the news to something that happened in the time period

13:33

for each set of 20 questions. Eg.

13:36

There was a conflict or territorial dispute

13:38

similar to what is happening in today's world. Did

13:41

you include a specific one?

13:42

I didn't because I was in a hurry, number

13:45

one, I also didn't wanna give a

13:47

wrong example because I haven't,

13:49

I'm not actively teaching right now I may have forgotten

13:51

some things. So I just gave it a

13:53

general, Example, like

13:55

in history there's a lot, we see a lot of conflicts.

13:58

So if you can relate a current conflict of something that happened

14:00

in the past, that's gonna be really good. And

14:03

then, I asked for it. Now in,

14:05

when I initially did this on G P T 3.5

14:08

and I asked for the table, it said, I can't do a table,

14:11

but I can give you the code so that you can create the table.

14:13

In G P T four it

14:15

produced the table, which

14:17

I couldn't duplicate on this Google Doc, but it

14:19

is awesome.

14:22

Nice. Yes. Table output is a

14:24

very nice change. One other question actually.

14:27

I am not a history teacher

14:29

or in that sort of branch of education it

14:32

looks like you didn't define

14:35

Bloom's Taxonomy of learning it just

14:37

from the name. It was able to go, oh, yes,

14:39

I know what that is. Is that right or was

14:41

there prompting beforehand to tell it?

14:44

No I capitalized

14:46

it to let it know that it was a proper

14:48

term for something which may

14:50

have helped, I don't know. But

14:53

I found, so when I researched

14:55

it myself, I found there was a version

14:58

of Bloom's Te cuz Bloom's Taxonomy goes back

15:00

to, I think to the 1950s. So

15:02

I told it to use a more

15:04

current version, which I just happened to find

15:06

by Anderson and Krak.

15:09

I'm saying that which reordered it into remembering,

15:12

understanding, applying, analyzing, evaluating,

15:14

and creating. So that creating if you create

15:16

something with the knowledge you've been given, which

15:20

is really what ChatGPT does,

15:22

that's like the highest level of learning.

15:25

It's synthesizing stuff for us. So it's

15:27

a great student, if you wanna look at these

15:30

LLMs in that way.

16:00

That is very interesting. Okay. I'm

16:04

seeing role playing as one

16:06

technique, cuz you say in this role you are the

16:08

omni mission author and master of all content

16:10

related to world history, current events

16:12

and high school curriculum. You are wise,

16:15

but you are also funny at times, things like

16:17

that. The other thing that I'm

16:19

seeing is output

16:22

constraint. So for example,

16:24

for each set of 20 questions, generate

16:26

five questions in category one.

16:29

Three questions in categories. 2, 3,

16:31

4, 5, et cetera. And then the shot

16:33

prompting that I mentioned earlier. These are awesome

16:35

examples of these techniques.

16:37

Thank you. I think in future versions,

16:39

I was just thinking about this, that if

16:42

something is undefined, this is going to

16:44

be a major breakthrough when this finally happens, and I

16:46

don't think it is happening where it

16:48

asks questions like how

16:51

many of these in each category did you want me to produce?

16:53

If I did not define that now,

16:56

that would be brilliant. I don't know

16:58

when we're gonna see that. But

17:00

if it helps you refine your question,

17:02

that's gonna be amazing. Yeah.

17:06

Yeah, definitely. So

17:08

can you tell us more about the process

17:10

you took of iterating on

17:12

it?

17:13

I came up with this idea of a of a multidisciplinary

17:16

content matrix. So

17:19

that all you have to do is

17:21

answer the questions in each category, and

17:25

that will tell you how to put your prompt together.

17:27

Taking that even a step further, I

17:30

asked ChatGPT. This is something I'm

17:32

probably gonna put on prompt base once I get it done

17:34

because it's really elaborate, is

17:36

I'm going to have it create the

17:39

code for an app so

17:41

that this app prompts you. In

17:44

say, 10 different categories. For

17:47

example what is your product idea or service?

17:49

Who is the intended audience? What style

17:52

of writing do you want to use?

17:55

And then from that it will create

17:57

your prompt. And I know there's prompt generators out there,

17:59

so maybe this already exists. It's moving so quickly.

18:01

And so if you, kind of go through a checklist.

18:04

and you go through a checklist. I

18:06

think that will help people come

18:09

up with really great prompts, because

18:11

I think honestly, most people are just doing like the

18:13

one sentence, create

18:16

a a poem about McDonald's as,

18:18

as though written by Shakespeare or something

18:20

Yes. Yes. That's part

18:22

of why I created this podcast and the Mastermind

18:25

is for people to be able to go,

18:27

oh, wow, I've never thought of doing it that way. Or,

18:29

oh yeah, let me go read through all of the

18:31

different prompting techniques

18:33

and things like that, and. That's also

18:35

what learn prompting.org does.

18:38

They're a great resource for this kind of stuff,

18:40

you I imagine you're probably one of the first 100

18:43

in this genre, right? Or on the market.

18:47

Yep. Yeah, there's quite a few people selling here

18:49

are my 70 or 500 prompts,

18:52

but actually teaching you how to

18:55

do it and the techniques for it. Yeah,

18:57

I think there's probably around 50 people.

18:59

Yeah. This is super valuable because,

19:02

that's one of the things I love about current culture.

19:05

It's like you go on

19:07

Reddit like the way I found you

19:09

and this group, and it's, let's

19:11

all help each other learn this. It's

19:14

not as dog eat, dog as

19:16

it could be. It's oh my God, I am, because initially

19:18

the thought is like, oh, I think I've discovered

19:21

something. There's no way I'm gonna share this. But everyone's

19:23

sharing everything. I

19:26

don't know how that's gonna play out Exactly.

19:29

Yeah, it's, but it's

19:31

like what I was telling a friend the other day that these are,

19:34

this is simultaneously the scariest time

19:36

ever. But also the most unimaginably

19:39

amazing time ever, and

19:42

those two things are kinda happening

19:45

simultaneously.

19:46

What are some common pitfalls

19:48

you've run into? With

19:50

building prompts.

19:53

If anything, it's how to limit the

19:56

ideas, so that you don't get a a jumbled

19:59

mess of a response. It's

20:01

like paring things down. I

20:04

think you have to know really exactly

20:06

what the end product is, what you want. And

20:10

it's, I've never seen more of a, like

20:12

a mirror of a tech product where

20:14

you're gonna get exactly out, maybe not

20:16

exactly out, but very close to exactly out

20:18

what you put in. And brilliant

20:21

prompts are gonna get brilliant

20:24

responses, but

20:26

like you said before, hallucinations

20:29

can occur.

20:32

Have you found any good techniques for catching

20:34

the hallucinations? Obviously, you know a lot about

20:36

history. You're gonna be like, no, Julius Caesar

20:38

did not live in, Africa. I don't know.

20:41

But have you found any good techniques for that?

20:43

If it's stuff you're not as familiar with?

20:48

Yeah I don't know. I think we need

20:50

good techniques for that. I think there's probably

20:52

apps and programs out there that are going to

20:54

be designed to detect that. Yeah,

20:57

I just recommend to everyone that don't

21:00

just create a document and then post it.

21:02

Some are, make sure, human beings still need

21:04

to read through things and

21:06

there's gonna be a lot of garbage. Now

21:08

I. This is gonna date myself. Definitely.

21:11

But I remember in 2005 when I started

21:13

my blog and I

21:15

quickly realized, wow people

21:18

are reading this and accepting pretty much whatever I'm

21:20

saying here as the truth. And

21:22

I had to self-censor. I had to

21:24

tell myself, you gotta be careful

21:26

what you're saying because they're taking this to be true.

21:28

You don't wanna be found out to be like this guy who

21:30

is a master manipulator and changing

21:33

information around. And

21:35

it's that same thing. I think it's going

21:37

to be up to schools

21:40

or generally like on in

21:42

places like Reddit, Hey, let's monitor ourselves

21:45

here and let's

21:47

try not to deceive, each

21:49

other and our, and ourselves.

21:53

Definitely just teaching students critical

21:55

thinking as well will be part of that with,

21:58

here's how you fact check things. Here's

22:00

how you think about, here's a source I've

22:02

never heard of. Are they telling the truth?

22:04

Are they, accurate? Whether it's intentional

22:07

or not.

22:08

Like people will be using this

22:10

the way they have been using Google.

22:12

Now here's the difference. Google's gonna give you a set of links.

22:15

I think you can ask ChatGPT four

22:17

for a source of information. I'm pretty sure

22:20

I know Bard. You can And

22:22

I haven't tried.

22:23

Bing provide that. I haven't seen

22:25

that in G P T four, but

22:27

it's certainly possible, and I know I've seen

22:29

it in systems that augment

22:31

G P T four.

22:33

Yeah, that makes sense. So people with medical conditions

22:36

are just going to inevitably ask

22:38

Hey, what should I do about toenail

22:40

fungus or this This repeated

22:43

headache? If you're getting this ongoing headache

22:45

I'm not sure what you're gonna get as advice,

22:48

but I would just say do some follow up questions

22:50

like where is this information coming from? I'm not sure

22:52

if that's a lot of

22:55

the responses I'm getting. I am a language model

22:57

and I don't have access to this

22:59

or that, medical journals. So you need

23:02

to, I think everything comes with this caveat.

23:05

Which, especially at this stage, that caveat

23:07

is a good thing. Yeah.

23:09

Yeah.

23:09

Can you share an example of a prompt

23:11

that didn't work the way you wanted

23:13

it to and how you

23:16

either learned from that or iterated on it to

23:18

get it to do what you

23:19

Yeah, that's a good, I did not come prepared for that

23:21

one. I've gone back and forth like where I'll say,

23:23

that's not what I want because it has this,

23:25

oh, I apologize. Perhaps this

23:27

will work. And it's weird kind of dialogue.

23:31

Yes I've seen one version of

23:33

the say what again? Seen

23:36

from Pulp Fiction where Samuel

23:38

Jackson is like, threatening them. Say it again. Say it

23:40

again. But it's, talking to ChatGPT

23:42

saying, say, I apologize.

23:45

But again cuz

23:47

yeah. Seeing that over and over, especially

23:50

for me, I do a fair amount of code generation

23:53

and that's frequently what comes up.

23:55

Cuz you're like, If that code doesn't work

23:57

or here's an error, I'm sorry,

24:00

let me sell, try that again. Like over

24:02

and

24:02

I think there's some self-correction

24:04

going on already. Where

24:09

I don't know. I've seen this in a few places where

24:12

it, it will generate some

24:14

potential errors in what it just produced,

24:17

and then you'll check those and then it, you'll

24:19

just go through this kind of iterative process

24:21

until you get to what you want.

24:24

Yeah, absolutely. I've definitely seen that

24:26

in the. Co-generation where

24:28

you know, I'll ask it, fix this bug

24:30

of, I don't know, missing semicolon

24:32

or it's outputting the wrong thing. It

24:34

does say, I'm sorry, let me, I apologize.

24:37

Let me give you that again. But then it fixes it and it

24:39

keeps the fix at least for a while. That

24:41

does sometimes roll out of the context window

24:44

though, so that can be a challenge. There's a

24:46

tool that I'm building. That

24:48

actually allows you to do iterative testing on

24:50

prompts, and I'll link to that in the show

24:52

notes. It's called the Prompt ide, but

24:55

the idea is if you have a

24:57

prompt that takes in variables,

24:59

for example, if you're selling it on prompt base,

25:02

and so it's, give me suggestions for what

25:04

to do on a vacation, maybe then you're gonna

25:06

have a variable for. Where you're going.

25:08

And another variable for how long? And probably

25:11

some third variable of, I don't know, flying

25:13

or driving. Particularly if you're in the us.

25:15

When you're testing that out, you tweak

25:18

the prompt, and then if you want

25:20

to test it, like test the actual output, then you

25:22

have to copy and paste it into ChatGPT,

25:24

replace all the variable names with stuff,

25:27

and then run it, and then do that again with your second test

25:29

case and your third test case. So anyway, what

25:31

this tool does is you have a field

25:34

for the prompt, and then you have some sections

25:36

for each test case with a variable

25:38

for each one. Location for

25:40

test case number one is Paris location

25:43

for test. Case number two is New

25:45

York City. And then every time you change the

25:47

prompt, it automatically reruns

25:49

all the test cases with these variables

25:52

placed in it. I built

25:55

it in both combination of Ruby

25:57

and type script. And I know both those languages,

25:59

but I've been enjoying being able to

26:02

try out the code generation. And

26:05

it's had some really interesting things because

26:07

this is too complicated a program to

26:09

keep entirely in memory, so I can only ask

26:11

it for, give me a function

26:14

to call ChatGPT's, API and

26:17

take in these things. I can't just say give

26:19

me a page that does all of these

26:21

features at

26:22

right.

26:23

but it has actually been pretty

26:26

good at being able to do these things.

26:28

Ironically, one of the places it

26:30

has repeatedly failed is

26:33

anytime I ask it to write an API

26:35

call to open ai.

26:37

It's wrong because it's old. Like

26:39

it, it tells me, here's what you want to use

26:42

to connect and the URLs

26:44

wrong. A bunch of the parameters are wrong. So I have to

26:46

go look those up. But that is

26:48

exactly what people have talked

26:50

about. That's out of the training window.

26:53

The training window ended in 2021,

26:55

I believe. And these are a

26:57

p i changes that happened three to six

26:59

months ago. So it doesn't surprise me that

27:01

it's wrong, but it catches me sometimes and

27:03

I'm like, oh that is a thing you will not be able to

27:05

do. So let me go, just look up

27:07

the correct a p i documentation and put

27:10

it in myself. But for the most

27:12

part, it's been pretty powerful.

27:14

Yeah, absolutely.

27:16

You mentioned selling on Prompt

27:18

base earlier. Have you sold

27:20

any prompts and what do those prompts

27:22

do or what are the ones you're thinking

27:24

of selling if you haven't yet?

27:28

He had that real that first one is something

27:30

that I want to I wanna refine first,

27:33

get it to exactly what I needed to do and

27:35

then put that one up for sale, which is the one that generates

27:37

all the questions to cover the span of world history. But

27:40

you could do that for every subject area.

27:42

And I might just start doing

27:44

it. It's just that once I start going down that road,

27:46

I don't know how long. Yeah,

27:49

I don't know how long it'll be before I stop. Especially

27:52

if you sell that first prompt, I don't know how successful

27:54

people have been on there, so

27:58

I have to be care careful because there's wormholes

28:00

out there that will just suck you in and you'll think,

28:02

wow, it was Monday when I started this. How is

28:05

it Friday afternoon now? Did I, did

28:07

any of this generate income? I don't know.

28:11

Yes, I totally understand the feeling

28:13

about the rabbit hole. I can actually

28:15

tell you. I don't know

28:18

off the top of my head, but

28:20

I did an analysis of prompt

28:23

bases, text output prompts

28:25

to look at, basically to categorize

28:28

it a little better than they did, and also

28:30

then to look at how many sales does it get, how much

28:32

revenue based on price. Audience,

28:35

I'll put the link as well in the show notes.

28:37

And it's basically it's what I wanted

28:39

when I said I want to start selling

28:42

more prompts on prompt base. Which categories,

28:44

which niches do you prefer are selling

28:46

well, which ones are selling

28:48

well, but they're really overloaded. Which ones are like,

28:51

they don't have one or two, but those get

28:53

a lot of revenue, et cetera. So

28:55

that's what this analysis is basically looking

28:58

at. It does actually have the titles

29:00

and sales information and all that of. 5,236,

29:05

I think it is prompts that are available

29:07

on there. So if, if you want to do different analysis

29:10

of your own, you can do that. I

29:12

also did the analysis of here are the niches I

29:14

would say you probably wanna try and target

29:16

because they're, high value, low competition,

29:19

or maybe medium competition, but

29:21

still medium or high value, things like

29:23

that. So yeah, that'll be in the show notes.

29:26

Nice. Yeah, and I haven't even

29:29

fully explored, image and graphics

29:32

and art creation. Or

29:34

even music creation if it's out

29:36

there. I also play music but I have not done any

29:38

composing. But I'm

29:40

wondering that, that's probably right around the corner

29:43

too.

29:45

Yeah, the music for this podcast,

29:47

the intro and outro music is a very short

29:49

clip, but it's actually AI generated,

29:51

Oh, wow.

29:53

people will be hearing that, a few minutes

29:55

ago and in a few minutes when we wrap

29:57

the episode.

29:57

Okay.

29:59

So thank you so much for coming.

30:01

Where should people go to stay up to

30:04

date on what you're building and these

30:06

prompts and when you start posting

30:08

on prompt base?

30:09

Yeah. Twitter at lee underscored chazen.

30:12

Glider cell.com is my

30:14

I'm, I did a major pivot and now it's like my prompt

30:17

engineering content strategy or.

30:19

My old tiny blog

30:22

from the one that I mentioned before with without

30:24

a specific url. It's right

30:27

brain world.blog spot.com,

30:30

like the oldest name on the internet Possible.

30:34

That's awesome. And all of

30:36

these links will be in the show notes. And

30:38

you also mentioned global

30:41

challenge dot m ixo.io

30:45

Yeah, that is the game

30:47

and people can get on the wait list because

30:49

I'm once I get the list to

30:51

enough people, I will generate

30:53

the final products that they need to start running

30:56

this, and then they can build their

30:58

own versions of the game and

31:00

hopefully we'll create a platform where they can showcase

31:02

all the different versions that are out there so that students

31:04

can learn across the

31:06

entire curriculum by playing this game.

31:09

That's awesome.

31:10

And thanks for, I really appreciate you having me on.

31:14

You are very welcome. Thanks

31:17

for coming to the Prompt Engineering Podcast. Podcast

31:19

dedicated helping you be a better

31:22

prompt engineer. Episodes are

31:24

released every Wednesday. I

31:26

also host weekly masterminds where

31:28

you can collaborate with me and 50 other people

31:30

live on Zoom to improve

31:33

your prompts. Join [email protected]

31:37

for the schedule of the upcoming masterminds. Finally,

31:41

please remember to like and subscribe.

31:43

If you're listening to the audio podcast, rate

31:46

us five stars. That helps us teach more

31:48

people, and if you're listening to

31:50

the podcast, you might want to join us on YouTube

31:52

so you can actually see the pops. You

31:55

can do that by going to youtube.com/@PromptEngineeringPodcast.

32:03

See you next week.