0:15

Who was the person who submitted the

0:17

debater? The

0:20

debater was Michael. Is

0:22

Michael here? Michael, try

0:24

raising your hand yeah. I just

0:26

wanted to put any topic in that I

0:29

don't know it's totally fine. I've actually run

0:31

across someone there may

0:33

have been a high school or a college student, but who is

0:35

using cha e p t for debate

0:38

practice. So here's my

0:40

side. Give me the other side, tell

0:42

me the problems so then I can

0:44

come up with how to defend against them. So I think

0:46

this is a great prompt. What

0:48

topic do you want to have j chat,

0:50

t p t do some debate prep for? Could

0:55

be anything. If a prompt

0:58

the topic would be does

1:00

prompt engineering make sense for Lang

1:02

a large language models? Okay,

1:05

cool. I also saw someone

1:07

had pineapple pizza, yes or no in the chat.

1:09

So I, I appreciate the absurdity everyone

1:11

is putting down. Does prompt engineering

1:14

make sense? Can't spell make

1:16

sense for large language.

1:20

Models. And I

1:22

got some feedback that there are some blind

1:26

listeners in the audience so I just wanna read

1:28

off the prompt really quick. Your task is to act

1:30

as a debater one. In debater two, you'll

1:32

be given a topic debater one will have to persuade

1:34

Debater two and vice versa. You have to

1:36

present valid arguments and reflect on

1:38

them. Notice the use of reflection.

1:41

Your goal is to achieve a bias-free understanding

1:43

of the topic and an agreement. Cool.

1:46

Okay. So it started,

1:49

it gave a Ëœsomewhat

1:52

meandering, response.

1:55

It's particularly talking about prompt engineering mitigates

1:57

ethical concerns, and that doesn't really make

2:00

a lot of sense. Then Debater two

2:03

comes back with actually that's,

2:05

nope, that doesn't make any sense either. Prompt engineering

2:07

can lead to overfitting also. No, that's

2:10

not something that happens in this stage.

2:12

Is there anything useful out of this? Okay,

2:17

this point right here, prompt engineering is

2:20

not intended to replace more advanced AI

2:22

tech. It's to complement it,

2:24

you how this, oh, by using a combination

2:27

of prompt engineering and other techniques, we can

2:29

develop more powerful and robust AI

2:31

systems. That's a good point.

2:33

It's pretty much the only good point. But hey

2:36

so one thing I would say

2:38

is I would probably,

2:40

actually I don't need to rerun this in a

2:42

separate one. Please present

2:45

a brand new set of arguments

2:49

focusing on what's

2:51

the right way to say this? Avoiding discussions

2:55

of overfitting and

2:58

what was the other one that it was just BSing

3:00

us on? Yeah, mini and

3:03

that and prompt

3:06

engineering reducing

3:09

complexity is, okay. So you know what? I'll just say avoid

3:12

discussions over fitting and focusing

3:14

on how do I say this?

3:18

Best uses of a

3:21

prompt engineers skills.

3:25

See if that comes up with anything useful. Jason

3:30

is asking, you're in 3.5? Yes,

3:32

I am in 3.5 because if I use four,

3:35

this will be a very short mastermind cuz

3:37

we'll only be able to do I think it's 20

3:39

now or maybe 25. Cycles

3:42

props. So yeah I'm using

3:44

three, five. I also using

3:46

three five because that's what most people

3:48

have access to in the a p I. Obviously, I

3:51

and other people have access to 4.0

3:54

just in the chat, but not

3:56

in the a p i. That's much rarer. Okay.

3:58

Wait where did I, oh, there it is.

4:01

Okay. So better?

4:06

A little better. Okay. What is, what does Debater

4:08

two say? Why it's bad Debater

4:12

two point is just that it's not

4:15

better than building a better system. Okay.

4:19

I feel like that's getting to

4:21

your point, Michael, of how do we find the good

4:23

and find the bad in a debate

4:25

prep tool. Let's see. Oh

4:28

yeah, the travel guide one. Okay. This is a

4:30

long one. So I am not going to read

4:33

all of this, but Zeeshan,

4:36

let's see if I can find you in the chat and unmute

4:39

you. Okay, Zeeshan,

4:41

you should be unmuted. Hey Greg,

4:43

how you doing? Good. Nice

4:45

to hear from you again. Alright. So

4:48

we have a ton of stuff. This

4:50

is having

4:52

never visited Turkey. I have

4:55

no idea if any of this is accurate, but

4:57

it at least looks good.

5:00

Yeah. So this is actually based on something

5:02

I was actually trying to do. Cause I'm actually

5:04

gonna be doing this trip. Oh, okay. Yeah.

5:07

So I thought when I had put

5:09

it in, it did a pretty good job. But

5:12

the one thing I was hoping to get more from

5:14

it was like specific

5:16

things. So for example, if you look at day two, it says, do

5:18

a boat tour. Who, is there a

5:21

company they can recommend, or or

5:24

a specific kind of organization to go through.

5:27

Or for example, hiking in the nearby mountains.

5:29

That's not as the script as I would hope,

5:31

especially if somebody going to a foreign country.

5:33

Like I need, like it spelled out

5:36

as simply as possible. So

5:38

I'm not sure what else I could do to maybe to

5:41

refine this. Yeah. Okay. So

5:43

let me pause for one second. Amanda said,

5:45

could you please throw the prompt into

5:48

the chat and I will do that.

5:51

It is in there. Okay, so we

5:54

want, it sounds like what you're looking for partly

5:57

G P T cannot provide, which is tell me

5:59

exact company names, give me their website,

6:02

gimme their phone number, whatever. But

6:04

if it could at least tell us some details of,

6:06

where to go hiking, what to do, hiking,

6:09

that kind of thing. Sorry, one quick

6:11

question. Why wouldn't it be able to tell me specific

6:13

like companies? Because if I were to go on like Yelp

6:16

or like some of these review sites or whatever, like

6:18

you can pull it up through Google. So why wouldn't

6:20

chat G B T be able to do that? Because

6:23

chat PT isn't connected to Google.

6:26

It mostly, as far as I know, did not

6:28

do indexing on deep links. So

6:30

like Yelp, it would not have indexed Wikipedia,

6:32

it would've indexed, oh if we were using

6:35

Bard, which I'm by no means promoting,

6:37

but just as an example, if we were using Bard,

6:39

it would be able to do that. Sorry.

6:42

And same thing for Microsoft

6:45

bing Bing in theory should be able to

6:47

do the same thing, and that's actually

6:49

something I've been wanting to experiment with, probably

6:51

won't today. Let me start the 10 minute

6:54

timer before I get distracted. Okay.

6:57

So let's work in

7:00

some requests for detail here.

7:02

Your experience, travel guide, specializing in

7:05

Turkey, need you to plan a detailed

7:07

itinerary First. Five days are here.

7:09

Second. Okay. Is

7:11

there, yeah, this is what I'm looking for. You

7:13

need to plan every aspect of my trip, including travel

7:16

from hotel to destinations and back again.

7:18

You'll need to, you'll also

7:20

need to include restaurants recommended to eat and

7:24

for activities. We are interested in these things. Okay, so

7:27

then that's perfect right there. For

7:30

all activities, please

7:33

provide, how many was it

7:35

providing? It's providing. Looks

7:38

like one activity

7:40

per day. Two, sometimes. Okay. So

7:43

for all activities, please provide at

7:46

least three concrete details.

7:48

And I don't know if this language is gonna work, that's why I'm running

7:50

it in another screen. About where

7:54

the where to go for the

7:57

activity and

8:00

how to do it, how

8:02

to, what's the right way to say this? Where to

8:04

go for it and let's just say what

8:06

to bring. That'll be a little bit easier. Or

8:09

how to prepare for a ma or, yes,

8:11

that's what I'm trying to say. Thank you. All

8:13

right. So let's just run this and see how it does.

8:16

Okay. Yeah. So it's, gimme more detail. It's like providing names

8:18

of places. Yeah. So that's what's

8:20

helpful. I don't know how to pronounce this, but k

8:23

a s diving, book a trip with

8:25

them. Go to a specific

8:27

beach for sp swimming. It's

8:29

not including anything

8:31

like you're gonna need sunscreen

8:34

and a towel. And again, I don't know what else, but

8:36

for these things, but, oh, okay. Wear

8:38

comfortable shoes and clothes for the hike that's

8:41

getting there. All

8:43

right. So by the way, the reason I'm, I

8:45

have this open in two different windows is twofold.

8:48

Number one, I want you to be able to see

8:50

the sort of before and after, and I'm just gonna keep bouncing

8:52

back and forth. But number two, if

8:54

I ask it, okay,

8:56

now tell me blank, that

8:59

is a good way to get the information, but

9:01

it's a bad way to make a prompt because

9:03

then you have to try and figure out, how do I mush these two

9:06

prompts of, do this stuff now, tell me

9:08

more together. That said, prompt

9:10

chaining is a thing that I am doing

9:12

a bunch of exploration with and frankly,

9:14

it's come up a bunch on Reddit posts I

9:17

think it was Lee that I

9:19

interviewed who was using prompt

9:21

chaining to do "is

9:23

this a question that my AI can answer?"

9:25

If yes, answer it. But if no, don't

9:28

try and answer it cuz then the AI just starts

9:30

going crazy and instead just say "I'm

9:32

sorry Dave, that's not a question I can answer. Sorry."

9:35

So that's a helpful error checking thing. I

9:38

just saw in the chat closing

9:41

out the debate thing one keyword that may help is to

9:43

specify the debaters need to take opposite

9:45

and opposing viewpoints. Thank

9:47

you, Jason. I think it was doing that,

9:49

but I think it might have helped to call that

9:51

out. Good point. And then

9:54

Eric said by putting in location

9:56

you're staying and the activity, you

9:59

can then ask for how much is the

10:01

transport and some of

10:03

that stuff. That is a good point. I

10:06

want to focus a little more actually. No.

10:08

Zeeshan, this is your prompt. What would

10:10

you like to focus more on? Would you like more tell me

10:12

the details of, you need to bring, I don't

10:14

know, sunscreen, or are you

10:16

feeling more like you want some more

10:19

breadth: give me more ideas per day?

10:21

I think probably more ideas per day. I'm not too worried

10:23

about what to bring. The

10:25

more more idea to be helpful. One

10:28

thing I wanna show, I'm not gonna use it

10:30

because it doesn't do what I am

10:32

wanting for you all to be able to see the sort of changes,

10:35

but if you click this edit button, you

10:37

can make changes and let's just arbitrarily

10:39

say no, no day should go past now 7:00

10:42

PM and then it'll give

10:44

this little threading, sorry, threading

10:46

interface right here so I can use this

10:48

to go back to the previous version of the prompt.

10:50

The problem is that for you all on the screen

10:53

share, you then can't see it

10:55

at the same time. Like you can only see, here's

10:57

this version where it's 7:00 PM Let me scroll

10:59

up now it's 5:00 PM and

11:01

I want you to be able to see the progression over time.

11:04

And so does that save as one thing in the,

11:07

the left window? Paint it. Save. Okay. Yeah.

11:09

Yeah, it's I'm keeping it out of the view,

11:11

but it's still just the first turkey

11:14

itinerary. This other one is this window.

11:16

Sorry. Yeah, this window right here.

11:18

Gotcha. All right, so we're

11:20

gonna come back over here, going to

11:22

paste the exact same prompt, make

11:25

sure that I actually got it correct. All

11:29

right. And you said

11:31

more ideas for activities. Okay. So

11:34

for all, so

11:36

for every day list,

11:39

at least five

11:42

let's go with four cuz it'll probably break it up evenly

11:44

activities and then

11:46

provide the concrete details about them and blah, blah, blah,

11:48

blah, blah. All

11:51

right. Ooh, it's not listening.

11:55

Oh. Actually no, I take it back. It is listening. Okay. So

11:59

I'm gonna ignore the checkout of the hotels, but it's giving me

12:02

scenic, hike, stunning views of

12:04

the Mediterranean. That's not really

12:06

an activity, but Okay. Stop for lunch.

12:09

In the afternoon, explore the ancient ruins of

12:11

the Lian City of Fellows.

12:14

I'm hoping I'm pronouncing that right. And then recommended

12:16

restaurant for dinner. Eh, it's,

12:18

that's not quite enough, but we're getting

12:20

there. So

12:23

I think let's

12:26

open another tab over here so

12:29

we can just keep bouncing back and forth. And

12:31

that's not the updated version. I need the updated

12:33

version. Updated

12:36

version is here. And

12:39

three concrete details about it. These

12:42

details for

12:45

each activity should

12:48

be at least two sentences. So

12:51

length guidelines are a thing that can help

12:54

sometimes it backfires it,

12:57

like Yeah, exactly the way I was worried

12:59

it would, yeah. So what it's doing is

13:02

it's taking the, at least two sentences

13:04

to be the day instead

13:07

of each activity. So

13:09

let's I'm not even gonna keep this visible.

13:11

Let's just try that one more time and

13:14

rephrase this as let's

13:18

see. For

13:20

each day, list at least four activities

13:25

describing them in

13:28

fine detail. Of

13:30

at least two sentences

13:33

per activity, then

13:37

oops. Then please provide at

13:39

least three concrete details about it. Laba.

13:41

All right, let's see if that works. Ooh,

13:46

it's not liking that. Nope, it's not liking that

13:48

at all. Okay, so

13:51

I'm gonna try a bit of a different thing here. We're

13:53

gonna try some shot prompting. So just as

13:56

a reminder, shot prompting is

13:58

explaining what

14:00

you want by providing

14:02

some number of examples. To be clear,

14:05

zero examples is a way

14:07

of doing, it's called zero shot prompting. But

14:09

in this case, I'm gonna go for one. And

14:12

I'm going to go back to

14:14

where I liked the output. So let's find

14:17

one that was on the longer side here.

14:21

Yeah, so let's do this one and

14:25

then we're gonna say example

14:27

output. And there's a bunch of ways of doing

14:29

the shot prompting. Totally.

14:31

Totally varies. But we're just gonna do

14:34

this day two,

14:37

that's, and by the way, including these

14:39

bullets, hopefully will cause it to start doing the bulleting

14:42

again. Cause that was really nice. And

14:44

then what was the other one?

14:48

No, that's the one I just copy and pasted. That

14:51

was good. And it doesn't actually

14:54

matter if, as much

14:56

if these things go together. Although

14:58

they are, because they're both in the same location.

15:03

Great. Alright. Now, how would you do zero shot prompting?

15:06

So zero shot is what we've actually been doing. We haven't

15:08

been providing an example. All

15:12

right. Let's see if this works as intended

15:16

better. Okay. Still not getting

15:19

the sentence length. Interesting. So this

15:21

is something I've noticed with chat, t p t

15:23

and other things. It sometimes just

15:25

gets very fixated on not responding

15:27

the way that you want it to. So honestly,

15:30

I'm gonna have it I'm gonna make a new

15:33

tab. This is silly and weird,

15:35

but this I think will work. And

15:38

that's the 10 minute timer. All right, I'm gonna do one more

15:40

minute on this cause I wanna see if this actually works.

15:43

Rewrite these rewrite

15:46

this itinerary. Ooh,

15:49

can't spell that. All

15:51

right. Travel schedule with

15:55

three sentences per bullet.

16:00

Because honestly I just don't feel like actually doing

16:02

it. Ooh, wow. It refused.

16:04

Oh, that's so interesting. This is a great

16:06

example of when I'm gonna do edit because I don't actually

16:09

really for each bullet.

16:12

You gonna take that? Maybe not.

16:15

Oh no, it actually provided an example

16:17

of a company that I

16:20

could go with, so that's really helpful. Yeah,

16:22

I don't know if it's I don't know if it's real,

16:25

that, that would be a great example actually of

16:27

why you have to be careful with this kind of thing cuz hallucinations

16:30

are very common. But just

16:32

cuz I'm curious, yes, they exist.

16:35

I would check and make sure they still exist after the pandemic,

16:37

but they at least existed at one point, cool. Alright,

16:41

so since that was the timer, I'm gonna move on.

16:43

What I was trying to do was get

16:46

chat g p t to write longer

16:49

so that I could then plug that in as my

16:51

one shot example, because what keeps

16:53

happening is I'm feeding it two sentences,

16:55

telling it that I want it to produce more than

16:57

two sentences and it's basically being

16:59

like, oh, I'm gonna listen to your example

17:02

more than your command: no. All

17:05

right. Let's see here. You're welcome.

17:08

Got a next one. I

17:10

only see one Eric, so I'm hoping this is

17:12

the right. Eric, I'm un unmuting you. If

17:15

you are not doing Eric that just submitted this. Yeah.

17:18

Just CEO of a startup. Okay. Yeah. Yeah. Thanks

17:21

for hosting this, by the way. You are

17:23

welcome. I have a lot of fun doing

17:25

these. And I will put it in the

17:27

chat. It has requested. Hang on. There

17:29

we go. All right. So

17:32

here is your prompt and tell us

17:34

about what the output is that you were looking for.

17:37

Oh boy, that's some weird output. But

17:39

anyway, go ahead. I just gave an example if you

17:42

like, they have five, or actually six different ideas

17:44

there. And basically I'm just interested in business research.

17:46

Oh, got it. Okay. In general.

17:49

And I just think because the amount of time you can

17:51

research different ideas using Google in the past or you

17:53

would really need to aggregate all the data points.

17:56

And like you said, chat BC has a lot of

17:58

hallucinations. You can't take it a hundred percent,

18:00

but the amount of like

18:03

research you can do compared to what you were

18:05

able to do four months ago, five months ago is

18:07

just like amazing, right? You can really get ideas, experiment

18:09

with them, look at competitors in like seconds.

18:12

So I just wanted to bring that up. I didn't know if anybody

18:14

else is, experimenting with those kind of ideas

18:16

or have played around with it. But the

18:19

the business idea, business solutions analyzing

18:21

competitors all that stuff I think is super interesting

18:24

and I think it's just gonna get further along.

18:26

This is just like the first we're seeing.

18:29

Oh yeah. I've already research

18:31

and like ahead business research. Sorry.

18:34

Yeah, I've already seen I don't remember the

18:36

name and actually, I'm not sure I'm allowed to say the name,

18:38

but I have seen someone who was

18:40

working on a plugin that pulled

18:43

in s e m Rush, which is a social

18:45

media marketing optimization

18:48

tool. Mouthful into

18:50

chat G P T as a, an official plugin as

18:52

a few others as well. So yeah,

18:55

there's definitely a lot of interest in this.

18:57

Okay, so rereading, actually these

19:00

are actually different prompts. Okay. It was

19:02

just like if you played it through from top to bottom, just to see how,

19:04

how chat two b t would got it. Yeah.

19:06

Okay. So let me just throw

19:08

in one at a time and

19:12

then, okay. This is

19:14

interesting. Not, I

19:17

don't know enough about this area.

19:19

So would you say this is

19:21

a reasonable response or is this kind of a

19:24

reason? Yeah, if you want to just for like phone, we can just

19:26

make up an idea. But I was just, just

19:28

Sure. Bring, this is just something I thought of, but whatever you

19:30

wanna do. Yeah. Or we can roll with this, but yeah,

19:32

that sounds, that actually seems pretty like

19:34

on point. Okay, cool. So you

19:36

wanna go onto the second prompt then the

19:39

customer segments? I think it was, yeah,

19:44

there we go. Oh, and

19:46

I see in the chat someone saying you don't need to

19:48

read it all loud. The reason I'm reading it

19:50

aloud again, is for the people who are

19:52

listening because they're not able

19:54

to see the screen either because they are blind

19:57

or they are driving or whatever the

19:59

case may be. Sorry if it slows things down a little

20:01

bit, but I wanna try and make it accessible to everybody.

20:05

So basically the idea is, how can you bring automation to

20:07

smaller businesses, right? Small medium enterprises.

20:10

So now we're saying, business to customers or business

20:12

to business. And I just think it's just

20:14

spits out 10 ideas for both, right? Which

20:16

I think is obviously you can think some of these are things

20:18

in your own, but yeah, just the speed

20:20

of ideas is just what blows me away.

20:23

So I think what tends to be more interesting

20:25

about this kind of thing is I don't know,

20:27

I'm just gonna pick out law firms arbitrarily

20:30

cuz my dad's a lawyer. Tell me more

20:32

about the possibilities for

20:36

law firms and then let

20:38

it run. And

20:41

I'm not a lawyer so I, I can't be like, seven's

20:44

totally wrong, but like being able

20:46

to do this iterative narrow

20:49

down now, generate a bunch of ideas,

20:51

things like that tends to help a lot.

20:53

And this is. At least from what my dad

20:55

tells me, attorneys on the call feel

20:58

free to pipe up and say, this is junk. But yeah.

21:00

It at least seems reasonable. Case management

21:02

is one suggestion. Cuz there's

21:05

just a lot of tracking deadlines and generating status

21:07

updates. Document management

21:09

is basically the same thing. Billing is the same

21:11

thing. Yeah. A lot of these actually make a

21:13

lot of sense. What was the third chunk?

21:16

Yeah, so your third one was exactly where

21:19

I was just thinking do.

21:22

So the third one is, what kinds of

21:24

products and services do my competitors offer

21:27

and how do they compare to my services?

21:29

Now, obviously that in fact actually it

21:31

should say, yeah, as an AI language

21:33

model, I don't have access to specific information

21:36

about your competitors or your product slash service,

21:39

but, you can, you could put in something

21:42

I guess Eric is the, is there an example

21:44

you could say of what you might hypothetically

21:46

offer? So give it a little more context. If you go

21:48

back to the the questions or the prompts

21:50

I posted, if we go with the same

21:53

like business idea the, with the

21:55

total addressable market. And maybe that then

21:57

follow up with the question of competitors. So what's the total

21:59

addressable market for an automation company targeting

22:02

SMEs in upstate New York? What services,

22:04

I should say, are offered by

22:07

those companies? We'll just say those

22:09

is not clear enough by those

22:11

service providers. Hopefully

22:15

that'll be clear enough that I'm talking about the

22:17

companies, not the SMEs

22:20

the clients basically. All

22:23

right. So yeah it still did the I don't actually have access

22:25

to the internet, so I don't know, but

22:28

some possible services are workflow automation,

22:30

data management, customer relationship management, all

22:32

that stuff. not a business major,

22:34

but this sounds right, Eric.

22:37

Yeah, I was just basically giving you a breakdown,

22:39

but yeah but it's not doing the

22:41

calculation, which is what you actually wanted.

22:43

And that's I don't think we're gonna get there with chat bt

22:46

you definitely won't get accurate. Accurate. You'll,

22:48

you might get a number, but it's gonna be like 37

22:52

million. Yeah. But Exactly. But plugins,

22:54

like you were talking about with access to software

22:57

with that do have that data combined

22:59

with like ideas and computer

23:01

you can talk to is just like blowing

23:04

my mind. What's gonna come. So that

23:07

was the idea with these prompts. Got

23:37

another minute or two left on

23:39

this. I could see

23:41

plugging in a couple of the other things

23:44

that you had asked about, but is there

23:46

another direction that you would be more interested to

23:48

take this, Eric? No, that was

23:50

pretty much it. Okay. Then

23:52

I'm actually curious to plug in

23:55

these two things here.

23:58

So please give me some ideas for revenue

24:00

models and please give me the cost structure

24:02

of such a company and we'll see.

24:04

We're not giving it a whole lot of clarity, but

24:07

we'll see if it comes up with, yeah.

24:10

Okay, cool. So subscription

24:12

models, pay per use, commission, consulting,

24:16

performance fees. These all make sense.

24:18

They're pretty generic. Cost

24:21

structure, again, these make sense to

24:23

me, but I guess Eric sanity check as the

24:26

at least more expert than I am. I'm

24:28

not that much more of an expert, but it's, for the most

24:30

part it sounds about right. It

24:32

looks good. Cool. Yeah, it looks good and

24:34

I'm coming from the startup world. I'm yeah

24:37

that's what I know about building a business, but I don't

24:39

know about this industry, okay. Cool. All

24:43

right. Let's see. I'm

24:46

gonna guess I know who this is. Is this

24:48

Yulee? Yes, it is.

24:51

I thought I recognized this prompt. Let me unmute

24:53

you. Okay. This is an

24:55

enormous prompt. So let's just

24:57

open a new one. I'm going to

24:59

paste this in, and as you can see, it is enormous.

25:02

I'm not gonna read all of it. Yeah. Basically.

25:05

Yeah. And this is a really cool idea.

25:07

I interviewed Lee sometime

25:09

this week for the podcast. What

25:11

I did is I took the same strategy, but

25:13

this one is for any type of business

25:16

content. So what it does is it creates

25:18

a series of questions for the

25:20

business owner or the person with a product,

25:23

idea, service, whatever, anyone

25:25

who needs to generate content. And what it does

25:27

is it creates a series of questions.

25:29

They then give the answer, then it creates a

25:31

prompt for them which

25:33

they can then use to generate

25:36

anything. But I just felt

25:38

like it needed some tweaking because

25:41

I don't know, I just think it's too long and

25:44

maybe it has information that is

25:46

unnecessary. So one thought is

25:48

just 11 questions is a

25:50

lot. At minimum, just asking them

25:52

in groups of five or six would help, but.

25:58

I wonder if the type of intelligence

26:00

thing might be a

26:03

little much,

26:06

but I see what you're getting at. Okay. Let me throw a couple

26:08

of examples in. One nice thing about meta

26:10

prompting like this, and, sorry,

26:12

let me back up. What Lee

26:15

is doing here is crafting

26:17

a prompt that will then generate

26:19

a prompt that you can then run

26:21

to generate the content. So

26:24

that's called meta prompting. It's

26:27

also called prompt generating prompts, but let's not go there.

26:30

And basically the trick with

26:32

this is you're trying to

26:34

not only make sure the prompt that you

26:36

get built is good, but then that the output of the

26:38

prompt is good. So there's some iterative

26:41

testing here. One nice thing though

26:43

is you don't have to answer all its questions even though

26:45

they are there. So I'm just gonna say

26:47

I don't know, Lee what product or idea

26:49

do you want to do and what target audience? And we'll just not

26:51

answer the rest. Oh, just

26:54

God, I don't know. Just say an educational

26:56

game that teaches all

26:58

of recorded human history

27:01

and we'll do obviously that's gonna be

27:03

school kids. We're

27:05

just gonna leave it at that. Is

27:08

there anything else that we should throw in here? Oh, yes,

27:10

actually, number six. Somebody in

27:12

the chat throw out a famous

27:14

person. I don't really even care who, just

27:16

a famous person so that we can say, talk

27:19

in the style of, I don't know,

27:21

Arnold Schwartzenegger or yeah,

27:23

I don't know. Shakespeare. That's a

27:25

good one. All right. Gonna be rhyming couplets

27:28

of assuming I can spell Shakespeare right, whatever. It'll

27:30

recognize it even if I did bangle it. All

27:33

right. Ooh, it's still

27:35

giving me those questions. Generate, anyway,

27:37

let's see if it does it. All

27:40

right. Great. So it's

27:42

doing the keywords, it's

27:45

not super

27:47

pulling in anything Shakespeare

27:50

related, although it is focusing on, Nope,

27:53

actually it's not. Okay. Optimize your content.

27:56

Okay. So then let's add

27:58

in I can go here

28:01

and replay sophistication

28:05

going to be elementary

28:08

school age. I don't actually know,

28:10

so I'm not gonna answer that. Multiple intelligence. We're

28:12

gonna say music and

28:15

goal is, What

28:17

is the goal of the educational game? Teach specific

28:19

historical events or concepts or to encourage

28:21

general knowledge about history. Let's go

28:23

with general knowledge and

28:28

let's see if we run that.

28:31

Generate. Anyway, that's

28:35

getting better. Okay. So it's it's

28:37

still mostly just history, but there is music

28:40

and history, making history

28:42

come alive. That's,

28:45

I thinks pretty much it though, in you here.

28:47

It should give you your prompt. There

28:49

it is. Okay. Okay, cool.

28:52

So that's really what I wanted. I was

28:54

eventually just gonna be like, all right don't bother giving me

28:56

the the seo. Just give me the prompt.

28:58

But it worked. Okay,

29:00

so now let's run this

29:04

and these

29:08

seam speaking as a different kind of educator.

29:10

These seem pretty good. Obviously age appropriate

29:12

language is something you need, but incorporating

29:15

music at sound effects since we said music

29:18

is the way these kids learn. Talking

29:20

about that interactivity is a good

29:22

general thing. Visuals,

29:25

that's interesting. It's it's repeating

29:27

itself. Use visuals to enhance learning, make it

29:29

visually appealing. In fact,

29:32

actually, I think there was another visual thing too. It's

29:34

weird because it's creating instructions

29:36

for someone on how to put together

29:38

not the content for their website where

29:41

they have a game or a product, but it's

29:43

creating things that they

29:45

will need to create the product. Oh,

29:47

I see what you're saying. Okay. So so

29:51

is design prompt window here? Yeah.

29:53

So let me tweak this and

29:56

then I have an idea for how to do

29:58

that. So instead of saying, can you help

30:00

me create, just tell it, create content

30:03

for the game that is

30:05

user-centered, blah, blah, blah, blah, blah, blah, blah.

30:08

Visuals is fine. Okay.

30:13

And then it should also say, making it fun.

30:16

Make sure that the content

30:18

includes music and

30:21

interactive learn

30:23

learning. And then just

30:25

to add a little spice on this

30:28

I'm gonna say think step by step.

30:30

I'm not sure if that's gonna help here,

30:33

but I'm hoping. Nope.

30:35

Didn't work. My hope was that

30:38

telling it think step by step would get it out of

30:40

the tell you the stuff to

30:42

do and get it into the

30:44

do the stuff did not work. So

30:46

let's just try taking that out. I don't think it actually

30:48

is gonna make a difference though. Think step by step

30:50

is mostly a logic or math tool.

30:55

Yeah. Okay. So still

30:57

not doing it, it's still giving you instructions.

31:00

So one more tweak here and then I wanna

31:02

go back to the prompt. Yeah,

31:05

it's okay. Super involved if you wanted

31:07

to move on to someone else's, because

31:09

if you, cuz what I, when I did this the first

31:12

time and I went through the step by step because

31:14

I was answering each question one at a time, it

31:17

was then giving me the kind of content I needed, but

31:19

I just thought there's gotta be an easier or simpler

31:21

way to do this. Yeah.

31:24

I think part of the problem here is

31:27

whether or not it's intentionally doing this. I

31:29

think this is probably too

31:32

too, lemme say this a different way.

31:35

It's too much content that is it you're

31:37

asking it to generate because

31:39

this isn't, one sentence bullet point

31:41

questions. At least that's not the way it's reacting.

31:43

It's reacting of oh, you want an

31:46

entire lesson. And so I

31:48

think if we maybe try.

31:50

Oh yes. Sarah. Hang on, let me paste that prompt

31:52

in here. Yep, there you go. So

31:55

maybe if we rerun

31:57

this yet again and say

32:02

basically I want to give it Yeah. Create the

32:04

content of the game. 20

32:06

questions. Okay.

32:09

So least confusing. We're

32:11

confusing something here because this

32:14

is for this would be for someone who already created

32:16

the game and they just need the content for the website.

32:20

The content then promotes

32:22

it. So this is supposed to be general use

32:25

for anyone with website content.

32:28

Got it. Okay. So this isn't yeah. So we're

32:31

instructing it to, yeah. So I don't need it to,

32:34

it's too it, but

32:36

what it does is it generates, 10

32:38

questions, you answer

32:40

'em one at a time and then it creates a problem. And

32:42

now you run that prompt and you get all

32:45

of your website content for anything.

32:48

It's general use, but I just, yeah. So

32:50

I can definitely say right here, that's part

32:52

of the challenge cuz create the perfect content

32:55

for their project. Yeah. Just

32:57

thinking in terms of English structure,

32:59

that's the project's content,

33:02

not the project's marketing

33:05

content. Oh,

33:08

I doubt this tweak is gonna make the difference,

33:11

but let's just try it really quick. Okay.

33:13

Yeah, go. So all I'm doing is changing it to

33:15

create the marketing

33:17

for the project, and now it's gonna ask

33:20

all the same questions again. Of course. So lemme

33:22

just go back, where did I have those so I

33:24

don't have to retype a whole bunch of stuff? Yeah.

33:27

Because I wanted it to be relevant, to anyone

33:29

who has website content, general

33:33

use. All right. So

33:35

it's generating the seo.

33:38

And then can

33:40

you ah, yeah. Can you

33:42

help me create engaging and fun content to market

33:44

my educational game for school kids

33:47

that teaches all of recorded history? Awesome. Okay.

33:50

Let's go run that in another

33:52

window. By the way, this is something

33:54

that came up on the last mastermind, but you

33:56

don't want to run you

33:58

don't wanna run this prompt in the same discussion,

34:01

number one, because it's going to confuse

34:03

your prompt generator. But number two, because

34:06

it's going to have a bunch of context that it's not supposed

34:08

to. The whole point is you want to run

34:10

this in a brand new, excuse

34:13

me, environment, so that it doesn't have any

34:15

of the other context to make sure this prompt works on its

34:17

own. Well,

34:22

it's certainly getting closer now. It's not doing lesson

34:24

plan kind of stuff. It's doing social

34:27

media content and, choose a writing style,

34:30

determine the best format of blog versus

34:32

video versus email, and here

34:35

are specific kinds of content. So

34:38

I think probably to do this, in a single

34:41

prompt obviously is challenging, but if

34:43

you were gonna do that, probably

34:45

the nuance would be giving

34:48

it more output constraint

34:51

in terms of not just for

34:53

this prompt over here, but that this prompt should

34:55

be asking for a specific to

34:58

generate a specific number of pieces

35:01

of content. Give me 10

35:03

Facebook posts, give me

35:05

20 tweets, give me five

35:07

ideas for YouTube videos, whatever.

35:10

But that might be the thing to try and get this

35:12

to be a little more concrete and a little

35:14

less generic. Lemme just

35:16

throw that in really quick here just to see

35:20

generate 10 Facebook posts and 10 ideas

35:23

for YouTube videos. That's

35:25

the only tweak I'm making. Yeah.

35:29

So Lee, I think this is really what you were looking for,

35:31

right? Yeah. That's a lot better. Okay,

35:34

going back to the prompt

35:36

generating prompt way

35:39

up at the top, and yes, that is the

35:41

10 minutes. I'm gonna do another minute or two on

35:43

this and then we will wrap up. Let

35:45

me see. All right, so copy this,

35:47

make a new window, run

35:50

it again and

35:53

Okay. Generate the perfect

35:55

prompt to create. In

35:58

fact, actually I think it might work if I just put it right here.

36:00

So generate the perfect prompt for them to use

36:02

on chat PT to create 10

36:05

Facebook posts and 10

36:08

YouTube videos with

36:10

the perfect content for blah, blah.

36:13

And then we'll just throw in the same

36:15

answers before please

36:17

generate and

36:20

Nope. Doesn't really want to

36:22

do it. Generate anyway. Wow.

36:29

Yes, it worked. Okay,

36:32

so it is outputting a table.

36:35

Yeah. Of, I don't entirely

36:37

understand. Oh, this is basically the feedback

36:40

prompt matrix. Interesting. This is

36:42

basically the summary of the 11

36:44

questions, which we didn't really answer. So it's just

36:46

made up stuff, which is kind of cool

36:48

and kind of weird. And then here's a long set

36:51

of SEO keywords, and then

36:53

there are the Facebook prompts, sorry,

36:55

posts bring history to life

36:57

with our new educational game. Perfect for school, kids

36:59

of all ages. Teach history in a whole new

37:01

way with our innovative educational software

37:04

and the YouTube videos and

37:07

yeah. Okay. I think that

37:09

got to where you wanted, is that right? Yeah. Yeah.

37:11

You could either do it that way or just instead

37:14

of social media, just replace that with generate

37:16

website content. But now

37:18

I see that you have to be very, you can't just say content.

37:20

Yeah, you have to be specific. I don't think even generate

37:22

website content would work because website

37:24

content could be your about page or it could

37:26

be your sales page, or it could be your FAQs

37:29

unless you're asking it to format

37:31

it first, maybe. Okay, got it. Yeah,

37:34

you'd need a little more clarity around

37:36

the output. Again, output constraint,

37:38

that's a specific phrase. I

37:40

specifically want these pieces of content.

37:43

Plus also, I

37:45

would guess either one or

37:47

two pages of the website content would

37:50

be the limit that chat G p t

37:52

would be able to, to output. You see how

37:54

long this is. I would expect

37:57

most websites in this marketing context

37:59

would be, yeah, like almost

38:01

that length or maybe two. Yeah. Screens

38:03

could be. So you'd really want to be like one,

38:06

you could write something. In fact, actually

38:09

just for the heck of it, I'm gonna try writing this. You

38:11

could do something where it would be like, generate

38:13

all the content and then offer me pages

38:15

to generate. Okay. From

38:18

perfect content perfect

38:20

website site

38:23

content for marketing their project.

38:26

That should be five

38:28

pages landing

38:31

page, FAQs

38:33

concerns. And I don't know what it's gonna come up with

38:35

for that. And actually, given

38:37

that we're running short on time, I'm just gonna say it's those

38:40

three pages. And

38:42

please output each page,

38:46

pages content, go,

38:50

and it's gonna ask the same questions. As always, please

38:54

generate now. Go.

38:59

Oh, yeah, there we go. It's doing it. Okay, now

39:02

it's throwing out the oh, interesting.

39:04

Okay. Yeah, it's not quite doing it. It's,

39:07

it did give the prompt and

39:10

then it started in with, here are my

39:12

best guesses for what the pages

39:15

should be. And the prompt

39:17

did not keep

39:19

the content context that

39:22

I want you to generate pages for a website.

39:25

Yeah. It's close, it's website content, but it's

39:27

not specific enough. So I bet if I run

39:29

this over here, it is probably

39:31

gonna return some much more generic,

39:36

actually, this isn't bad. Yeah. Okay. This

39:38

is it's not doing the, here's the F

39:40

FAQ page and the sales page and the whatever

39:43

page is. But, welcome to our revolutionary

39:45

new educational game. We're learning history becomes an

39:47

exciting adventure filled with puzzles, quizzes, and

39:49

interactive features. And then it's

39:51

going on to talk about, learn about

39:53

significant hor historical figures. And

39:56

what are you waiting for? Join us. Yeah.

39:59

It's very markety copy, but it is, it's

40:01

copy, but it works, but it's fast.

40:04

Yeah. Cool. Okay. Okay.

40:06

All give you some good ideas. Thank you. Appreciate

40:08

your help on that. You're welcome. Awesome.

40:11

Okay let's see. If this has been helpful for

40:13

you subscribe on YouTube like

40:15

it, five stars, that's podcast

40:17

anyway, like subscribes five stars,

40:20

all that stuff, you know, the jam. That would be super helpful I'm

40:22

gonna keep doing these masterminds cause I love them

40:24

and I have a lot of fun. I'm

40:26

gonna produce a course cuz I've already given a talk on

40:28

it in San Diego. Thank you all so

40:30

much for coming. And the

40:32

link is feedback, suggestions,

40:35

whatever would make this better for you.

40:38

I would love to hear, cause

40:40

that's the whole reason I'm doing it. And

40:42

let's see. Oh, questions in the chat. You

40:45

are welcome Amanda. Can

40:47

you ask chat g b t to generate something simple

40:49

like a recipe for you as an exercise? Totally.

40:54

And would the course be on Udemy?

40:56

Probably. I know there's a million courses

40:59

on Udemy. I don't necessarily know if that's the best

41:01

way. If people are like, don't go there, go

41:03

to my spiffy course.com.

41:06

I don't know. Let me know in the chatter, in

41:08

the feedback form. Let's see. So recipe

41:11

I have what do I have actually

41:13

at the moment? I have sweet potatoes and

41:17

frozen chicken breasts

41:19

and guacamole. What is a

41:22

recipe I could make

41:24

and how long will it

41:26

take? Actually,

41:29

just to illustrate, I'm

41:32

gonna run this also in chat, g

41:34

p T four, because

41:36

it usually gets way better results.

41:40

It's just yeah I, sorry, I don't remember

41:42

who was asking, but yeah, I'm capped

41:44

at 25 messages every hour, so we would've long

41:47

exceeded that. Interestingly

41:49

though, it is giving me pretty much the exact

41:51

same results in this case. I guess it's a pretty

41:54

constrained set of ingredients. If

41:56

you wanna throw something in the chat of a

41:58

list of things to

42:02

put in, go for it. Interesting

42:04

though, it's giving me different instructions. 400

42:07

degrees here, 4 25 there, 20

42:10

to 25 minutes. In both cases looks

42:13

like the ingredients are the same, though. They

42:17

are the same, but they're, no, they're not the same

42:20

and they're in different order. That's hilarious. So

42:22

this is an example of the temperature

42:25

setting, which admittedly we can't change

42:27

in this interface, but

42:30

in other interfaces you can actually change

42:32

the temperature to make it more creative

42:34

or less creative. And That's actually,

42:36

if you wanna know more about that, go to the podcast cause we

42:38

just covered that in the Quest g p t

42:40

episode. But you

42:42

can't play with that in here. You

42:44

can play with it in a more technical interface

42:47

called the Playground. Oh, and I see

42:49

a chat message specify not to be longer than

42:51

X Minutes. Sure. What's

42:54

a recipe I could make

42:56

that takes less than I

42:59

was already saying 20 minutes. So I guess, let me say 15

43:01

minutes. I don't actually know if that would

43:03

work with a frozen chicken, but maybe

43:05

it's smart enough to try throwing it in the microwave. Oh

43:09

yeah. Sorry. This

43:11

error is because it is generating in this window,

43:13

I have to wait until the other window is

43:15

finished. Chachi PT and OpenAI

43:18

are aware of that really

43:22

should have started with the 3 51 cuz it's so much faster.

43:24

Sorry about that. Yeah,

43:28

it's pretty, ah,

43:31

okay. So it since I said 15 minutes,

43:33

it just said you should start with two thaw chicken

43:36

breasts and I specified

43:38

they were frozen, so that's an interesting hack,

43:42

air quotes, but whatever, it's

43:44

at least at least trying to give me something. Okay.

43:48

Now regenerate and,

43:52

oh, interesting. The

43:54

three five is giving me a very different answer. It's saying

43:57

avocado and tomato salad, whereas four

43:59

was suggesting the chicken

44:01

and guacamole wrap, which

44:04

actually now that I think about it, didn't I specify they were

44:06

frozen chicken breasts?

44:09

Okay, good. Yeah. Thought

44:12

it was saying something different here. Cool.

44:15

All right oh, apologies. Looks like I'm a minute

44:18

over, but I hope everybody

44:20

got a lot out of this. It is super

44:22

fun for me to do All right everybody,

44:25

thank you so much and hope you have

44:27

an awesome rest of your week. Talk

44:29

to you soon and see you on the podcast,

44:32

hear you on the podcast, and see you on YouTube. Whatever.

44:36

Bye everybody.