1
00:00:00,288 --> 00:00:04,111
AI it's not magic It's not magic.
2
00:00:04,111 --> 00:00:14,117
It responds based on how clearly you communicate right bad Some of you may have heard bad
input bad output right bad data in bad data out
3
00:00:14,117 --> 00:00:15,310
Here's a bad prompt.
4
00:00:15,310 --> 00:00:17,936
Write a marketing plan for my business.
5
00:00:17,937 --> 00:00:20,064
Oh, yeah, I mean, there's not enough information there.
6
00:00:20,064 --> 00:00:20,305
It's good.
7
00:00:20,305 --> 00:00:21,550
It's bound to fail.
8
00:00:21,550 --> 00:00:22,503
Yeah, that's a good.
9
00:00:22,503 --> 00:00:24,126
That's a good bad one.
10
00:00:24,127 --> 00:00:25,748
Super vague, right?
11
00:00:25,748 --> 00:00:27,008
Super vague.
12
00:00:27,169 --> 00:00:29,810
So here's how you take that and make it a good prompt, right?
13
00:00:30,309 --> 00:00:32,522
Welcome to the AI Advantage podcast.
14
00:00:32,522 --> 00:00:36,291
I'm your host Sheridan Wint, and I've got my friend Jon Foster here.
15
00:00:36,291 --> 00:00:43,117
And we're going to be talking about how to talk to AI and get better results, also known
as prompt engineering.
16
00:00:43,117 --> 00:00:45,360
You may have heard of prompt engineering before.
17
00:00:45,361 --> 00:00:53,884
little of the experience that yesterday when we asked Chad GPT and Gemini and I the two
different things and once we rephrased it I can definitely see how this is going to be
18
00:00:53,884 --> 00:00:54,735
helpful.
19
00:00:54,736 --> 00:00:57,837
Yeah, yeah, yeah.
20
00:00:57,837 --> 00:00:59,228
That was on our last episode.
21
00:00:59,228 --> 00:01:07,842
I think it's episode four in the series, where we did a side by side of Gemini and Chat
GPT to create an image of Florida man.
22
00:01:07,843 --> 00:01:11,414
And in honor of that image, Jon is dressed like Florida man today.
23
00:01:11,415 --> 00:01:12,688
Ha ha ha ha!
24
00:01:12,886 --> 00:01:13,991
Get on get.
25
00:01:13,993 --> 00:01:18,451
All right, so digging into why prompting matters
26
00:01:18,729 --> 00:01:22,552
AI it's not magic It's not magic.
27
00:01:22,552 --> 00:01:32,558
It responds based on how clearly you communicate right bad Some of you may have heard bad
input bad output right bad data in bad data out
28
00:01:32,981 --> 00:01:37,510
It's the same thing when you're prompting a model right any model
29
00:01:37,510 --> 00:01:41,288
Good prompts, better results, less editing, way more accuracy.
30
00:01:41,289 --> 00:01:44,757
You put any funny prompts in, Jon, that you got some crazy outputs on?
31
00:01:44,758 --> 00:01:48,085
Uh, mine, my favorite is when I misspell things.
32
00:01:48,085 --> 00:01:50,620
Cause I type too fast or my phone's just too small.
33
00:01:50,620 --> 00:01:51,733
I don't know what it is.
34
00:01:51,733 --> 00:01:56,021
And then something completely what I did not ask comes up and I'm like, Oh.
35
00:01:56,022 --> 00:01:56,902
Yeah.
36
00:01:57,985 --> 00:02:00,968
Usually the spelling, usually for me, I haven't seen like a spelling thing.
37
00:02:00,968 --> 00:02:02,490
Usually it's like, I know what you meant.
38
00:02:02,490 --> 00:02:04,953
And it just keeps on going.
39
00:02:05,294 --> 00:02:07,076
Must be misspelling some.
40
00:02:07,076 --> 00:02:11,662
There's some words though that you move one or two letters around and the whole meaning of
the word changes.
41
00:02:11,662 --> 00:02:12,422
m
42
00:02:12,652 --> 00:02:15,593
Yep, like, um, like, uh, like duck.
43
00:02:15,594 --> 00:02:16,571
like duck.
44
00:02:16,572 --> 00:02:18,170
How would you rearrange duck?
45
00:02:18,475 --> 00:02:25,808
I mean, you click one key to the right, you're in a little bit of trouble.
46
00:02:25,808 --> 00:02:26,678
I see, I see.
47
00:02:26,678 --> 00:02:28,048
Got you.
48
00:02:28,049 --> 00:02:29,009
Typo.
49
00:02:29,009 --> 00:02:30,109
Typo.
50
00:02:30,110 --> 00:02:30,560
All right.
51
00:02:30,560 --> 00:02:32,331
So let's talk about prompts, right?
52
00:02:32,331 --> 00:02:35,772
What actually is a prompt?
53
00:02:35,773 --> 00:02:36,893
Any ideas?
54
00:02:36,936 --> 00:02:38,376
a prompt.
55
00:02:38,777 --> 00:02:48,623
Well, I would assume it would be setting the AI up for, you know, a request of information
that you, whatever you put in.
56
00:02:48,623 --> 00:02:49,813
So input.
57
00:02:49,814 --> 00:02:52,785
is input, yeah, it's anything, any input that you give it.
58
00:02:52,785 --> 00:03:02,411
So a good way to structure your prompt are instructions and then context and then your
desired outcome, right?
59
00:03:02,411 --> 00:03:07,583
Instructions plus context plus desired outcome equals a good prompt.
60
00:03:07,584 --> 00:03:07,814
Right?
61
00:03:07,814 --> 00:03:12,367
So you want to, it tends to work best like any given model, right?
62
00:03:12,367 --> 00:03:19,980
AI tends to work best when it knows who it's writing for, what you want, how you want it
delivered, and what good looks like, right?
63
00:03:19,980 --> 00:03:21,251
What you expect.
64
00:03:21,252 --> 00:03:28,806
So we're going to dive in to a four-part framework, which we touched a little bit on, but
it's going to sound a little different, right?
65
00:03:28,806 --> 00:03:32,178
Number one, the role or the perspective, right?
66
00:03:32,471 --> 00:03:40,793
Act as a marketing strategist, or act as a customer support expert, or act as a mad
scientist from 1970.
67
00:03:40,793 --> 00:03:41,923
And then give it the context.
68
00:03:41,923 --> 00:03:42,634
What's the goal?
69
00:03:42,634 --> 00:03:43,494
Who's the audience?
70
00:03:43,494 --> 00:03:45,595
What problem are we solving?
71
00:03:45,595 --> 00:03:47,596
Give it a little bit of context.
72
00:03:47,598 --> 00:03:49,598
Well, we'll get into some examples here in a minute, right?
73
00:03:49,598 --> 00:03:51,538
So number one, we got the role and the perspective.
74
00:03:51,538 --> 00:03:53,198
Number two, we've got context.
75
00:03:53,198 --> 00:03:54,418
Number three, we've got task.
76
00:03:54,418 --> 00:03:56,718
What exactly do you want it to do?
77
00:03:56,718 --> 00:03:58,498
And number four, the format, right?
78
00:03:58,498 --> 00:03:59,558
Do you need bullet points?
79
00:03:59,558 --> 00:04:00,178
Do you need a script?
80
00:04:00,178 --> 00:04:01,298
Do you need an email, a table?
81
00:04:01,298 --> 00:04:03,418
Do you need steps, a blog article?
82
00:04:03,418 --> 00:04:04,878
What is it, right?
83
00:04:04,879 --> 00:04:12,799
So we're going to talk about some good prompts and some bad prompts and just do a little
comparison, right?
84
00:04:12,991 --> 00:04:13,798
Okay.
85
00:04:13,799 --> 00:04:15,504
Can you give me a bad prompt?
86
00:04:15,505 --> 00:04:17,213
Oh, I could probably do that all day.
87
00:04:17,213 --> 00:04:19,662
um
88
00:04:19,663 --> 00:04:20,816
No, I can't.
89
00:04:20,816 --> 00:04:22,198
Not on the spot.
90
00:04:22,240 --> 00:04:23,433
I thought I could.
91
00:04:23,433 --> 00:04:24,624
I failed you.
92
00:04:24,782 --> 00:04:26,116
right, all right, a bad prompt.
93
00:04:26,088 --> 00:04:27,281
Here's a bad prompt.
94
00:04:27,281 --> 00:04:29,907
Write a marketing plan for my business.
95
00:04:29,908 --> 00:04:32,035
Oh, yeah, I mean, there's not enough information there.
96
00:04:32,035 --> 00:04:32,276
It's good.
97
00:04:32,276 --> 00:04:33,521
It's bound to fail.
98
00:04:33,521 --> 00:04:34,474
Yeah, that's a good.
99
00:04:34,474 --> 00:04:36,097
That's a good bad one.
100
00:04:36,098 --> 00:04:37,719
Super vague, right?
101
00:04:37,719 --> 00:04:38,979
Super vague.
102
00:04:39,140 --> 00:04:41,781
So here's how you take that and make it a good prompt, right?
103
00:04:41,779 --> 00:04:46,152
Let's go back and look at our four part prompt framework, right?
104
00:04:46,152 --> 00:04:52,405
Role, task, and format.
105
00:04:52,446 --> 00:04:55,558
So you are a marketing strategist, right?
106
00:04:55,558 --> 00:04:56,648
That's the role.
107
00:04:56,714 --> 00:04:57,666
Mm-hmm.
108
00:04:57,667 --> 00:05:01,793
Helping a small real estate team struggling with lead consistency.
109
00:05:01,793 --> 00:05:03,515
That's some context.
110
00:05:04,357 --> 00:05:11,517
The task, create a 30 day marketing plan focused on email nurturing and social content.
111
00:05:11,518 --> 00:05:12,536
and then the format.
112
00:05:12,536 --> 00:05:14,961
Format it as a weekly checklist.
113
00:05:15,243 --> 00:05:19,709
So that's going to put out something way better than write a marketing plan for my
business.
114
00:05:20,011 --> 00:05:21,011
100%.
115
00:05:21,011 --> 00:05:25,966
So a bad prompt.
116
00:05:25,966 --> 00:05:29,489
We can walk through another example again together.
117
00:05:29,490 --> 00:05:31,218
where can I buy a shirt?
118
00:05:31,219 --> 00:05:33,091
Where can I buy your shirt?
119
00:05:33,193 --> 00:05:36,147
Definitely a shirt.
120
00:05:36,279 --> 00:05:37,843
Not just any shirt.
121
00:05:37,844 --> 00:05:38,955
Where can I buy a shirt?
122
00:05:38,955 --> 00:05:40,377
Yeah, definitely.
123
00:05:40,377 --> 00:05:42,379
You don't know what you're gonna get with that one, right?
124
00:05:42,455 --> 00:05:43,882
No, not at all.
125
00:05:44,269 --> 00:05:46,363
So role, let's go through it.
126
00:05:46,363 --> 00:05:46,915
What's a role?
127
00:05:46,915 --> 00:05:48,407
What's the person's role?
128
00:05:48,527 --> 00:05:49,755
or perspective.
129
00:05:49,755 --> 00:05:52,082
from the perspective of...
130
00:05:52,083 --> 00:05:54,635
What was the name for somebody that's in the fashion?
131
00:05:54,842 --> 00:05:59,648
so in the role of someone who was up to date on today's fashion.
132
00:05:59,649 --> 00:06:00,322
Okay.
133
00:06:00,322 --> 00:06:01,843
Would that be a good start?
134
00:06:01,843 --> 00:06:02,956
Would that be a good roll?
135
00:06:02,956 --> 00:06:04,838
Or would you improve that?
136
00:06:05,299 --> 00:06:08,735
Yeah, mean, you could say a model, as a fashion model.
137
00:06:08,736 --> 00:06:15,234
mean, that's the thing, like, my thing is like I overthink things and I'm like, the
fashion model wouldn't do it, they just put, they just wear the clothes.
138
00:06:15,235 --> 00:06:18,186
but you want to buy the fashionable clothes, right?
139
00:06:18,187 --> 00:06:20,973
Yeah, but a fashion designer, let's say that.
140
00:06:23,200 --> 00:06:24,778
As a fashion designer.
141
00:06:24,778 --> 00:06:26,260
and then context.
142
00:06:26,473 --> 00:06:31,933
And then pick out a shirt that would look good this season.
143
00:06:31,934 --> 00:06:32,854
Okay, let's season it.
144
00:06:32,854 --> 00:06:33,615
on
145
00:06:33,617 --> 00:06:37,363
the recent fashion week event that just happened
146
00:06:37,721 --> 00:06:38,063
Okay.
147
00:06:38,063 --> 00:06:42,498
I thought you meant like fall, summer, spring, winter, but fashion season.
148
00:06:42,498 --> 00:06:43,261
I got it.
149
00:06:43,261 --> 00:06:44,242
I got it.
150
00:06:46,595 --> 00:06:48,158
I thought you meant weather seasons.
151
00:06:48,158 --> 00:06:48,939
uh
152
00:06:57,447 --> 00:06:58,433
Here's what we got so far.
153
00:06:59,186 --> 00:07:04,126
Role number one, as a fashion designer, to the context, what problem are we solving?
154
00:07:04,266 --> 00:07:10,986
I need to find a new shirt to wear to an event.
155
00:07:12,526 --> 00:07:16,666
Task, find me a shirt and explain the event, right?
156
00:07:16,666 --> 00:07:18,506
Maybe is it like a black tie event?
157
00:07:18,506 --> 00:07:19,246
Is it a ball?
158
00:07:19,246 --> 00:07:22,486
Is it just a kickback at somebody's house?
159
00:07:22,486 --> 00:07:25,366
Is it like a at?
160
00:07:25,366 --> 00:07:28,448
So you want to give it that context, what kind of event is it?
161
00:07:28,448 --> 00:07:35,116
And then the task, find me a shirt that I can wear that's blue.
162
00:07:35,116 --> 00:07:36,676
Yeah, OK.
163
00:07:36,997 --> 00:07:48,478
So even if the problem has not as much information, you've got to kind of add more detail
into it to be able to generate the better results.
164
00:07:48,478 --> 00:07:49,076
I'm assuming.
165
00:07:49,076 --> 00:07:49,826
Yeah, yeah.
166
00:07:49,826 --> 00:07:51,667
So here's another way we could do it, right?
167
00:07:51,667 --> 00:08:02,774
uh As a fashion model, find me, I need a, as a fashion model, that's the role, I need a
blue dress for a cocktail party.
168
00:08:02,774 --> 00:08:04,154
That's the context.
169
00:08:04,154 --> 00:08:09,036
Task, find me a blue dress I can wear.
170
00:08:09,037 --> 00:08:12,799
And then the format, provide this as a list of
171
00:08:12,800 --> 00:08:17,140
URLs where I can go look at the dresses.
172
00:08:17,402 --> 00:08:18,299
Okay
173
00:08:18,300 --> 00:08:19,640
Now you can shorten that up a little bit too.
174
00:08:19,640 --> 00:08:26,760
You could say, as a fashion model, I need a blue dress for a cocktail party.
175
00:08:26,860 --> 00:08:31,740
Provide me with a list of dress options.
176
00:08:31,800 --> 00:08:31,940
Right?
177
00:08:31,940 --> 00:08:36,040
Now that last sentence, you put the task and the format in the same sentence, right?
178
00:08:36,040 --> 00:08:40,300
The task is give me a list of dresses and the format is a list, right?
179
00:08:40,300 --> 00:08:41,880
Of dresses though.
180
00:08:41,881 --> 00:08:44,141
So yeah, so it doesn't have to be crazy.
181
00:08:44,141 --> 00:08:48,033
You can kind of combine a lot of that stuff, but basically the more information you give
it.
182
00:08:48,477 --> 00:08:50,432
better result you're going to get.
183
00:08:50,433 --> 00:08:55,566
Yep, yep, yep, All right, prompting mistakes, common prompting mistakes.
184
00:08:55,566 --> 00:08:57,247
Being too vague, right?
185
00:08:57,247 --> 00:08:58,847
Give me a business plan.
186
00:08:58,947 --> 00:09:02,118
Well, we need to know what business it is and what you're planning for, right?
187
00:09:02,118 --> 00:09:02,798
What do you sell?
188
00:09:02,798 --> 00:09:03,849
All of that good stuff.
189
00:09:03,849 --> 00:09:04,609
What are your services?
190
00:09:04,609 --> 00:09:05,489
What are your offerings?
191
00:09:05,489 --> 00:09:06,330
What's your budget?
192
00:09:06,330 --> 00:09:09,311
Asking for too much at once, right?
193
00:09:09,371 --> 00:09:13,233
Asking for it to give you a plan.
194
00:09:13,605 --> 00:09:24,016
for how you're going to get from a high school graduate to landing on the moon with just
your own money, your own budget.
195
00:09:24,017 --> 00:09:24,938
Right?
196
00:09:26,259 --> 00:09:33,414
It might give you some sort of plan, but it's gonna be pretty tough to execute because it
will lack a lot of details, right?
197
00:09:33,415 --> 00:09:35,716
Another common mistake, leaving out the audience, right?
198
00:09:35,716 --> 00:09:43,862
Telling it to generate something, but not telling the model who the intended audience is.
199
00:09:43,862 --> 00:09:44,562
Who's gonna read it?
200
00:09:44,562 --> 00:09:45,592
Who's gonna look at this thing?
201
00:09:45,592 --> 00:09:49,534
Another common mistake, forgetting to specify the tone, right?
202
00:09:49,534 --> 00:09:51,946
You want to say is it going to be friendly, right?
203
00:09:51,946 --> 00:10:00,552
Let's say you're drafting an email, is it supposed to be professional, is it supposed to
be friendly, is it supposed to be authoritative, you know, what kind of tone do you want
204
00:10:00,552 --> 00:10:01,084
it to have?
205
00:10:01,084 --> 00:10:02,074
Examples, right?
206
00:10:02,074 --> 00:10:04,866
Not giving examples, that's another common mistake.
207
00:10:05,080 --> 00:10:12,166
it gets much better if you give it an example and say, here I'm looking for a JSON file
structured in this particular way, right?
208
00:10:12,166 --> 00:10:17,852
And if you give it the example, well, the example should have all the values, all the keys
in it that you're looking for, right?
209
00:10:17,852 --> 00:10:28,462
So then it can say, OK, well, let me just replace the values and keep all the keys and
your output will be structured exactly like probably you wanted it to be because you gave
210
00:10:28,462 --> 00:10:29,632
it an example.
211
00:10:29,634 --> 00:10:35,788
Another common mistake is not telling it, you know, not telling the model what you don't
want it to do, right?
212
00:10:35,788 --> 00:10:38,589
Don't format this as HTML.
213
00:10:38,710 --> 00:10:47,324
Don't, right, don't give me a list of texts, give me a list of hyperlinks, right?
214
00:10:47,324 --> 00:10:49,706
So you're telling it what not to do and what to do.
215
00:10:49,706 --> 00:10:53,248
Another common mistake is not iterating enough, right?
216
00:10:53,248 --> 00:11:00,988
first response that you get from a model, not always the best response, but you can modify
it, you can adjust it, it's a conversation, right?
217
00:11:00,988 --> 00:11:03,288
It's not like a one-off request.
218
00:11:03,579 --> 00:11:11,731
So when you're saying you have to ask it a multitude of times or just a couple times about
the same thing, redundancy?
219
00:11:11,732 --> 00:11:20,829
Not always, but if you don't like the response that you get and you want it to be better,
you can tell it exactly in what way you want it to be better.
220
00:11:20,829 --> 00:11:30,077
So let's say in our example earlier, it gives us a list of dresses and it says, you know,
this dress by this designer, this dress by that designer, right?
221
00:11:30,077 --> 00:11:32,688
Dress number three by designer number three.
222
00:11:32,729 --> 00:11:34,230
And it's just giving you list of text.
223
00:11:34,230 --> 00:11:35,271
Well, that's not what you wanted.
224
00:11:35,271 --> 00:11:40,795
What you wanted was a list of links, hyperlinks that you can click on and then it opens it
up.
225
00:11:40,985 --> 00:11:53,412
in uh Sephora or I don't know some fashion brand I have no idea so if it gives you not
what you wanted then you can just say hey format this you know I like to actually just put
226
00:11:53,412 --> 00:12:05,128
the word output with a colon instead of having to say format it like this I just put
output colon list of urls right very brief
227
00:12:05,128 --> 00:12:07,066
And what is what is that by doing that?
228
00:12:07,066 --> 00:12:07,447
What is that?
229
00:12:07,447 --> 00:12:09,515
What does that do specifically?
230
00:12:09,515 --> 00:12:19,333
Well, instead of giving you a list that says, you know, this blue dress by designer one,
this blue dress by designer two, instead it will literally just give you a list of
231
00:12:19,333 --> 00:12:20,935
hyperlinks you can click on.
232
00:12:20,936 --> 00:12:30,234
Okay, so but but the the colon adding the colon in there and stuff like that, is that is
that a key to all the non-it people?
233
00:12:30,235 --> 00:12:37,650
So I like to, let's say we're talking about context, we're talking about giving it more
information, better data in, better data out.
234
00:12:37,650 --> 00:12:47,814
Pro tip that I like to do is I'll just write a single sentence or question, but I give it
context within the question.
235
00:12:47,814 --> 00:12:53,137
Let me go, let me pull up an example and just go look at something recently that I did.
236
00:12:53,138 --> 00:12:56,162
Alright, so I wanted to...
237
00:12:56,163 --> 00:12:57,300
you find it.
238
00:12:57,301 --> 00:13:06,030
I was working on a growth plan for a client and I put in a sentence that says, output
this...
239
00:13:06,312 --> 00:13:09,394
with formatting instead of markdown, right?
240
00:13:09,394 --> 00:13:18,076
And so it's one sentence, output this as formatting instead of markdown, which it might
sound like a bad prompt, right?
241
00:13:18,076 --> 00:13:27,861
But in between those two words where I said, this, after this, I put parentheses, and
inside the parentheses, I put like a massive amount of text.
242
00:13:27,861 --> 00:13:32,402
It's like 10 or 12 paragraphs, and it's all written in
243
00:13:32,402 --> 00:13:43,508
Markdown format and all I wanted it to do was give me To actually give it to me as bold
right because it gave it to me in plain text so when you look at the text it's like Star
244
00:13:43,508 --> 00:13:48,329
star star week one star star star and that makes it bold right?
245
00:13:49,130 --> 00:13:52,191
And I didn't want all that in there, so I just wrote a sentence.
246
00:13:52,191 --> 00:13:58,733
said output this with the formatting instead of in markdown and I put all of that text in
parentheses
247
00:13:58,733 --> 00:14:10,492
in there so it knows what text I'm talking about and it knows that I don't want it to look
like that right I want it so parentheses I use all over the place I'll use a colon I might
248
00:14:10,492 --> 00:14:21,740
say if I'm building a website I might say make these make the following changes colon
right and then parentheses and I'll put all the changes in there and then close the
249
00:14:21,740 --> 00:14:26,063
parentheses and put a period and then I might also give it some more to do
250
00:14:26,262 --> 00:14:28,979
Although you could probably put all of the changes within the parentheses.
251
00:14:28,979 --> 00:14:30,784
um
252
00:14:30,784 --> 00:14:35,519
say markdown like for for the non-it people what does that mean
253
00:14:35,520 --> 00:14:37,237
Also me, because I don't know.
254
00:14:37,239 --> 00:14:40,349
So Markdown is just like a format, a syntax.
255
00:14:40,349 --> 00:14:42,123
It just...
256
00:14:42,124 --> 00:14:43,572
I don't know.
257
00:14:43,572 --> 00:14:46,643
Let me ask Chad JBT to explain it better than I can.
258
00:14:46,644 --> 00:14:50,387
I probably should give you a different example, a way more simple example.
259
00:14:50,388 --> 00:14:53,665
Show us an example of a good problem for solving this problem.
260
00:14:54,029 --> 00:14:54,766
So.
261
00:14:54,767 --> 00:14:55,731
Let's see.
262
00:14:55,732 --> 00:14:59,495
So let me give you an example using the colon and the parentheses, right?
263
00:14:59,495 --> 00:15:02,997
Back to our fashion designer, right?
264
00:15:02,997 --> 00:15:07,000
Let's say that they, let's say that they are our fashion model.
265
00:15:07,000 --> 00:15:19,088
Let's say that we made the prompt like this, help me find a dress that meets these
requirements and doesn't look like it's been bedazzled and wasn't in this list of dresses,
266
00:15:19,109 --> 00:15:19,589
right?
267
00:15:19,589 --> 00:15:21,574
And so that's like one sentence.
268
00:15:21,574 --> 00:15:23,497
It's like one run-on sentence.
269
00:15:23,497 --> 00:15:30,590
But if you use parentheses in there, in the right places, then you can give it much better
requirements.
270
00:15:30,590 --> 00:15:32,954
So let's start a new chat here.
271
00:15:32,954 --> 00:15:34,295
I'll share my screen.
272
00:15:34,671 --> 00:15:41,844
So like, putting like things like no sequins and other little details that go in the
dresses that you do not want.
273
00:15:41,844 --> 00:15:46,242
Like, you don't want shiny, sparkly, you want something more matte.
274
00:15:46,438 --> 00:15:47,339
Okay.
275
00:15:47,646 --> 00:15:49,033
Find me a-
276
00:15:49,034 --> 00:15:54,462
provide, let's do it this way, as a fashion model.
277
00:15:54,463 --> 00:16:04,208
provide me a list of dresses that meet these requirements, parentheses, right?
278
00:16:04,749 --> 00:16:09,201
That I can wear to a cocktail party.
279
00:16:09,202 --> 00:16:14,951
Make sure it doesn't have any of these features.
280
00:16:15,213 --> 00:16:16,278
Parentheses.
281
00:16:16,539 --> 00:16:33,038
and wasn't isn't on this list of dresses from last year right so in the requirements here
i might say
282
00:16:33,039 --> 00:16:33,971
Blue.
283
00:16:33,972 --> 00:16:47,244
sleek not fluffy right no has lace i don't know i don't know i'm not not really into
dresses so i'm just making this up as i go but
284
00:16:47,244 --> 00:16:49,119
Just mini skirts.
285
00:16:49,420 --> 00:16:50,642
Yeah, mini skirts.
286
00:16:50,642 --> 00:16:54,146
ah Covers the knees, right?
287
00:16:54,146 --> 00:16:55,287
Let's keep it classy.
288
00:16:55,287 --> 00:16:57,829
Covers the knees.
289
00:16:58,491 --> 00:16:59,583
That I can wear to a cocktail
290
00:16:59,583 --> 00:17:00,475
yeah.
291
00:17:01,059 --> 00:17:06,068
Oh, yeah, we'll put is at least knee length.
292
00:17:06,069 --> 00:17:06,732
uh...
293
00:17:06,732 --> 00:17:11,945
make sure doesn't have any these features sequence i'm probably gonna spell that wrong but
whatever ah
294
00:17:11,945 --> 00:17:12,837
I have no idea.
295
00:17:12,837 --> 00:17:15,113
Actually, I think you spelled it right, but I'm not sure.
296
00:17:15,113 --> 00:17:16,873
bedazzled.
297
00:17:16,874 --> 00:17:17,694
glitter.
298
00:17:17,694 --> 00:17:24,474
Alright, doesn't have any of that and is not on this list of dresses from last year.
299
00:17:24,474 --> 00:17:25,724
We'll say...
300
00:17:25,725 --> 00:17:28,158
I don't know anything about dresses.
301
00:17:28,158 --> 00:17:33,845
uh Not from this list of brands.
302
00:17:33,845 --> 00:17:35,026
We'll do that.
303
00:17:35,467 --> 00:17:38,710
So no Nike dresses, no Adidas.
304
00:17:38,892 --> 00:17:43,198
Jimmy Choo, Versace, you know.
305
00:17:43,368 --> 00:17:45,749
yeah, sure, no Versace.
306
00:17:47,330 --> 00:17:49,901
So anyway, so we just put all of that into parentheses.
307
00:17:49,901 --> 00:17:59,555
So now this prompt, looks pretty long, but it's really just a sentence, know, find me the
dress that meets these requirements that I can wear to a cocktail party.
308
00:17:59,555 --> 00:18:02,696
Two sentences, make sure it doesn't have any of these features, right?
309
00:18:02,817 --> 00:18:04,538
And isn't on this list of brands.
310
00:18:04,538 --> 00:18:06,418
So that's very specific.
311
00:18:06,418 --> 00:18:07,959
Now it's gonna give me a list.
312
00:18:07,959 --> 00:18:08,519
right?
313
00:18:08,519 --> 00:18:11,695
And it's probably, I don't know if they'll have URLs or not.
314
00:18:11,696 --> 00:18:12,916
I'm curious how this is going to go.
315
00:18:12,916 --> 00:18:14,816
Looking and address recommendation.
316
00:18:14,817 --> 00:18:17,944
focusing on reputable retailers.
317
00:18:17,944 --> 00:18:19,026
It's telling us how it's thinking.
318
00:18:19,026 --> 00:18:19,487
Look at this.
319
00:18:19,487 --> 00:18:24,147
ah All these sites that it's um searching.
320
00:18:24,148 --> 00:18:26,574
double check that they don't have sequins or glitter.
321
00:18:26,817 --> 00:18:27,670
It's very cool, man.
322
00:18:27,670 --> 00:18:30,598
You can see the thought process it's going through.
323
00:18:31,080 --> 00:18:33,277
Yeah, that is extremely neat.
324
00:18:33,278 --> 00:18:34,093
Hmm.
325
00:18:34,094 --> 00:18:43,333
It's said it came across some dresses that it's not sure if it has it but that brand
typically does have some sparkly stuff going on so it's just gonna stay away from that
326
00:18:43,333 --> 00:18:44,214
brand.
327
00:18:44,344 --> 00:18:45,301
Okay.
328
00:18:45,302 --> 00:18:46,625
Searching some more.
329
00:18:46,626 --> 00:18:48,199
Saks Fifth Avenue.
330
00:18:48,466 --> 00:18:49,348
Okay.
331
00:18:49,349 --> 00:18:52,289
Here we go, it says each product page is going to be cited.
332
00:18:52,290 --> 00:19:01,490
All right, it's giving a list of dresses and a description of why that dress meets the
requirements, and there's a URL.
333
00:19:01,491 --> 00:19:02,284
Dope.
334
00:19:02,772 --> 00:19:03,969
So we have five dresses here.
335
00:19:03,969 --> 00:19:05,805
Now I can say output.
336
00:19:06,189 --> 00:19:08,736
list of URLs only.
337
00:19:08,737 --> 00:19:11,051
is maybe I want to email this to somebody, right?
338
00:19:11,051 --> 00:19:12,229
Okay, yeah.
339
00:19:12,230 --> 00:19:14,262
So now we're iterating on the prompt.
340
00:19:14,262 --> 00:19:16,983
We're telling it exactly what output we want.
341
00:19:16,984 --> 00:19:17,693
thinking.
342
00:19:17,693 --> 00:19:19,565
All right, there's the URLs.
343
00:19:19,566 --> 00:19:26,994
Write this as write this list as an email to a
344
00:19:26,995 --> 00:19:35,538
Fashion model friend So we're just iterating on what it came out with and now here we go
now It basically did everything for us.
345
00:19:35,538 --> 00:19:37,698
It drafted the email.
346
00:19:37,818 --> 00:19:40,759
It says hey friend's name, right?
347
00:19:40,759 --> 00:19:50,242
We didn't tell it our friend's name I've been scouting some cocktail ready dresses and
found a bunch that I think you'll like Blue sleek lace knee-length or longer completely
348
00:19:50,242 --> 00:19:53,713
free of sequins glitter and bedazzling perfect for a chic evening
349
00:19:53,714 --> 00:19:57,532
And like that, we now know where to go get dresses.
350
00:19:58,015 --> 00:20:00,539
I'm sure everybody was super interested in that.
351
00:20:00,539 --> 00:20:03,182
uh
352
00:20:03,183 --> 00:20:04,803
It is 2025.
353
00:20:05,985 --> 00:20:07,607
Ladies, here you go.
354
00:20:07,607 --> 00:20:13,232
The top one on here is self, number one is self portrait by blue fine lace midi dress.
355
00:20:13,232 --> 00:20:15,673
no, maybe that brand is self portrait.
356
00:20:16,074 --> 00:20:17,917
Anyways, dress.
357
00:20:17,917 --> 00:20:20,639
at Nordstrom, so there's one by my house.
358
00:20:21,058 --> 00:20:21,990
Yeah.
359
00:20:22,215 --> 00:20:23,123
All let me stop sharing.
360
00:20:23,483 --> 00:20:25,004
All right, let me stop sharing.
361
00:20:25,004 --> 00:20:30,529
So anyway, that's some examples of how you can make your prompts better, how you can avoid
some of the mistakes.
362
00:20:30,529 --> 00:20:32,450
Let's do some, there's some templates.
363
00:20:32,450 --> 00:20:34,171
Some people actually sell these templates.
364
00:20:34,171 --> 00:20:36,113
I don't think that's necessary for you to do that.
365
00:20:36,113 --> 00:20:41,657
You can just ask AI to help you create it into a better prompt, right?
366
00:20:41,657 --> 00:20:45,720
But if you wanna go buy templates, feel free, go buy them.
367
00:20:45,720 --> 00:20:47,422
Maybe we'll put some on the website, who knows?
368
00:20:47,422 --> 00:20:50,424
I think it's a little, I think it's a, I
369
00:20:50,425 --> 00:20:51,971
seems to me it's like...
370
00:20:52,519 --> 00:20:53,309
huh.
371
00:20:53,511 --> 00:20:56,231
What happens if we ask AI?
372
00:20:56,232 --> 00:21:03,452
say, like, yeah, like, like, how can I, how can I best format my questions for you to help
me?
373
00:21:03,740 --> 00:21:06,522
So there's a cool setting in a lot of these models.
374
00:21:06,522 --> 00:21:11,216
A chat GPT, for example, you can go into the settings and you can say preferences, right?
375
00:21:11,216 --> 00:21:14,839
And you can tell it, you know, talk to me like a five-year-old, right?
376
00:21:14,839 --> 00:21:19,994
You can also say, if I don't give you enough information, ask a clarifying question,
right?
377
00:21:19,994 --> 00:21:22,947
Or ask clarifying questions if you don't have enough information.
378
00:21:22,947 --> 00:21:27,368
Another thing I like to do is I'll use one model to...
379
00:21:27,368 --> 00:21:29,759
draft a prompt and put it into a different model.
380
00:21:29,759 --> 00:21:35,181
So let's say I'm like building a website and I'm using AI to like build a whole web app or
something.
381
00:21:35,181 --> 00:21:37,062
I'll format all of it.
382
00:21:37,062 --> 00:21:47,067
Like I'll get it all ready to go in and say like Claude or ChatGBT and say, help me create
a prompt for this tool that I'm using to build a website.
383
00:21:47,067 --> 00:21:50,549
I want the website to have to meet these requirements, right?
384
00:21:50,549 --> 00:21:52,229
I want it to match my brand.
385
00:21:52,229 --> 00:21:56,291
I want it to align with my tone and you've got to tell it, you these are my brand colors.
386
00:21:56,363 --> 00:21:58,044
and this is the tone I use.
387
00:21:58,044 --> 00:22:06,569
Here's some, I'll go to a website and grab all the text on the website and say, here's my
existing website and everything that is on there now, right?
388
00:22:06,569 --> 00:22:08,231
And, you know, build out a site map.
389
00:22:08,231 --> 00:22:13,014
I want it to have these pages, contact us and, you know, features and offerings.
390
00:22:13,014 --> 00:22:19,304
And so you just put all that and then chat GPT or whatever model you're using will spit
out a prompt.
391
00:22:19,304 --> 00:22:24,520
And you can take that and put it in the other one and it'll build out the whole website,
the whole app for you.
392
00:22:24,521 --> 00:22:25,132
That's nuts.
393
00:22:25,132 --> 00:22:26,163
That's crazy.
394
00:22:26,163 --> 00:22:28,887
Is it able to verify its own creation?
395
00:22:28,887 --> 00:22:37,989
Like, is it able to like, you know how when you create a website and everything like that,
like the contact us button isn't mapped correctly, you know, and it doesn't, you know, is
396
00:22:37,989 --> 00:22:39,821
it able to check on itself?
397
00:22:39,822 --> 00:22:41,454
Some of the models can, yeah.
398
00:22:41,454 --> 00:22:44,237
Some of the tools can.
399
00:22:44,237 --> 00:22:46,919
One of the ones I use, it'll test it, right?
400
00:22:46,919 --> 00:22:58,131
It'll do like, when you do a software development cycle, got dev, test, QA maybe, uh or
UAT, and then production, right?
401
00:22:58,131 --> 00:23:04,677
Everybody's, all the organizations tend to have their own way that they do those, but this
tool that I use, will.
402
00:23:04,678 --> 00:23:11,021
it will develop it and then it will test it make sure that it works and then i'll say okay
it's ready to go and publish it
403
00:23:11,022 --> 00:23:18,882
That is that is amazing because I was I was just thinking in my head and everything like
one of the biggest things and one of the hardest things that I experienced when learning
404
00:23:18,882 --> 00:23:23,082
this and stuff like that was like user friendliness.
405
00:23:23,082 --> 00:23:30,742
You know, you could probably like ask AI and be like from, you know, the perspective of a
web developer.
406
00:23:30,743 --> 00:23:36,023
Um, make input, whatever the website or whatever.
407
00:23:36,043 --> 00:23:39,523
What would you suggest to make this more user friendly?
408
00:23:39,705 --> 00:23:40,446
Yeah.
409
00:23:40,447 --> 00:23:47,121
And then, you know, maybe a little bit more context sprinkled in, but you know, it
probably tell you exactly how to do it.
410
00:23:47,121 --> 00:23:48,302
That's crazy.
411
00:23:48,676 --> 00:23:50,218
It would, it would.
412
00:23:50,677 --> 00:23:58,827
All right, with that, I think that's a good point for us to hear a quick message from our
friends and sponsors, and we'll be right back after this.
413
00:23:58,957 --> 00:24:04,842
Right now, AI is deciding who your next customer trusts, and it might not be you.
414
00:24:04,842 --> 00:24:15,790
You might not realize it yet, but ChatGPT, Gemini, Perplexity, all of these models are
already deciding who they choose to recommend in industries like yours.
415
00:24:15,790 --> 00:24:17,741
And they don't always list options either.
416
00:24:17,741 --> 00:24:19,913
Sometimes they give one answer.
417
00:24:19,913 --> 00:24:25,847
And if that answer isn't you, then AI just referred your ideal customer to your
competitor.
418
00:24:25,847 --> 00:24:29,260
That's why we created the free AI Visibility Review.
419
00:24:29,260 --> 00:24:39,829
It's a short live session where we will sit down with you and show you what AI thinks of
you and your brand and your company, who it thinks is an authority, and what you can do to
420
00:24:39,829 --> 00:24:42,791
become the authority that it starts to recommend.
421
00:24:42,791 --> 00:24:44,853
Don't just guess where you stand.
422
00:24:44,853 --> 00:24:45,854
Find out.
423
00:24:45,854 --> 00:24:55,327
Schedule your free AI Visibility Review today at advantagelabs.ai and see if AI knows that
your business exists.
424
00:24:55,327 --> 00:25:00,643
Imagine what it would do for your business if chat GPT was actually referring people to
you.
425
00:25:00,664 --> 00:25:02,286
Advantage Labs .
426
00:25:02,286 --> 00:25:05,629
AI, your AI advantage.
427
00:25:06,552 --> 00:25:07,442
And we're back.
428
00:25:07,442 --> 00:25:09,761
This is the AI Advantage podcast.
429
00:25:09,761 --> 00:25:16,837
I'm your host, Sheridan, and we were just talking about prompt engineering, specifically
how do you talk to AI and get better results.
430
00:25:16,837 --> 00:25:25,861
The next thing that we were gonna go through were some ideas for how to do different
things with your prompts, right?
431
00:25:25,862 --> 00:25:27,056
Okay, let's get into it.
432
00:25:27,056 --> 00:25:29,876
Let's get into it.
433
00:25:29,936 --> 00:25:31,276
First up, brainstorming, right?
434
00:25:31,276 --> 00:25:34,036
Give me 10 ideas for blank, right?
435
00:25:34,036 --> 00:25:35,676
You know, it's a good way to start a prompt.
436
00:25:35,676 --> 00:25:40,956
Give me 10 ideas for this or let them give me 10, give me 100, give me 50, whatever,
right?
437
00:25:40,956 --> 00:25:42,456
Give it a number.
438
00:25:42,457 --> 00:25:44,777
Another thing that you can do is rewrite, right?
439
00:25:44,777 --> 00:25:49,437
Rewrite this to be clearer, shorter, more casual, that kind of thing.
440
00:25:49,438 --> 00:25:55,218
Obviously you gotta put the context in there of what you want it to rewrite.
441
00:25:55,274 --> 00:25:56,904
But we're not going to get into the whole prompt.
442
00:25:56,904 --> 00:26:00,817
We're just going to give you real quick some ideas on how you can start off a prompt.
443
00:26:00,817 --> 00:26:01,658
Summarization.
444
00:26:01,658 --> 00:26:03,519
Summarize this as a bullet point list.
445
00:26:03,519 --> 00:26:06,019
Turn this into a step-by-step workflow.
446
00:26:06,019 --> 00:26:11,462
If you want to create a standard operating procedure, create a process, turn this into a
step-by-step workflow.
447
00:26:11,462 --> 00:26:12,463
Comparisons.
448
00:26:12,463 --> 00:26:13,884
Compare option A and B.
449
00:26:13,884 --> 00:26:14,774
I do this a lot.
450
00:26:14,774 --> 00:26:17,285
I'll say compare and contrast.
451
00:26:17,359 --> 00:26:22,550
the these two softwares or these two products I'm thinking about buying, right?
452
00:26:22,550 --> 00:26:28,194
And give me like, give me a well, that's all I'll say I say compare and contrast these.
453
00:26:28,634 --> 00:26:29,454
Yeah.
454
00:26:29,454 --> 00:26:38,838
Well, yeah, it'll it'll do that on its own, but you can also specify like, you know, I'm
concerned about the price or I'm I really want to make sure I have this feature, right?
455
00:26:38,838 --> 00:26:42,081
So you can get that context in there, but I use it a lot.
456
00:26:42,081 --> 00:26:43,410
for comparing things.
457
00:26:43,410 --> 00:26:45,742
All right, now onto the power of iteration.
458
00:26:45,742 --> 00:26:50,544
We talked about that a little bit earlier with uh the example that we did with the dress,
right?
459
00:26:50,544 --> 00:26:52,085
Iterating on that prompt.
460
00:26:52,085 --> 00:27:00,830
So just a few things when you get some output and it's not exactly what you want, just
tell it how it can be exactly what you want.
461
00:27:00,830 --> 00:27:01,090
Right?
462
00:27:01,090 --> 00:27:04,065
Make it friendlier, make it shorter, make it more concise.
463
00:27:04,065 --> 00:27:08,191
Give me three variations to pick from or give it an example.
464
00:27:09,005 --> 00:27:14,764
A lot of times the second prompt is better than the first or the second output is better
than the first as well.
465
00:27:14,765 --> 00:27:18,546
and we learned that on the dresses part now for sure i agree to that
466
00:27:19,041 --> 00:27:20,224
Yeah, yeah, yeah.
467
00:27:20,224 --> 00:27:21,567
And we learned it with Florida Man.
468
00:27:21,567 --> 00:27:24,908
Now let's say that you are getting an output that you don't like, right?
469
00:27:24,908 --> 00:27:26,468
And you did everything you can think of.
470
00:27:26,468 --> 00:27:29,419
You used the four part prompt framework, right?
471
00:27:29,419 --> 00:27:34,791
You did the role, the role or perspective, the context, the task and the format.
472
00:27:34,791 --> 00:27:41,672
You did all of that and you iterated on it and it's still coming out bad and you don't
like the output, right?
473
00:27:41,672 --> 00:27:47,275
A simple thing you can do is say, what information do you need from me to finish this,
right?
474
00:27:47,275 --> 00:27:48,295
What do you need from me?
475
00:27:48,295 --> 00:27:51,088
What details do you need from me to make this better?
476
00:27:51,088 --> 00:27:54,167
uh You can just ask it and see what it says.
477
00:27:54,167 --> 00:27:57,667
Maybe it needs some details that have not been provided yet.
478
00:27:57,668 --> 00:28:00,268
Yeah, I like that feature for sure.
479
00:28:00,268 --> 00:28:08,908
Cause especially like, you mean, you, know what you want something and you need it done,
but you don't know exactly how to, how to, how to phrase it.
480
00:28:08,908 --> 00:28:09,368
You know what I mean?
481
00:28:09,368 --> 00:28:12,888
And that kind of gives you like a guideline to where you can like level it out.
482
00:28:12,888 --> 00:28:13,228
Yeah.
483
00:28:13,228 --> 00:28:14,332
I like that.
484
00:28:14,333 --> 00:28:15,723
Yeah, yes, sir.
485
00:28:15,864 --> 00:28:17,866
All right, I think that about does it for today.
486
00:28:17,866 --> 00:28:20,248
A few takeaways from today's episode.
487
00:28:20,248 --> 00:28:25,611
A good prompt is clarity plus context plus constraints.
488
00:28:25,611 --> 00:28:26,533
That's a good prompt.
489
00:28:26,533 --> 00:28:31,529
We've got the four prompt framework here that anyone can reference.
490
00:28:31,529 --> 00:28:32,069
any time.
491
00:28:32,069 --> 00:28:34,281
We're going to have a blog article about that that you can check out.
492
00:28:34,281 --> 00:28:35,904
That'll be a link in the description.
493
00:28:35,904 --> 00:28:38,646
Something to think about, AI is your partner.
494
00:28:38,646 --> 00:28:41,389
Think of it like a junior team member that can help you out.
495
00:28:41,389 --> 00:28:44,662
And I'd like to challenge you to go try a few prompts today.
496
00:28:44,662 --> 00:28:45,222
It's free.
497
00:28:45,222 --> 00:28:47,715
You can try a lot of these tools for free.
498
00:28:47,715 --> 00:28:51,620
Go try them out, see what you can do, and let us know how it goes.
499
00:28:51,620 --> 00:28:52,340
All right.
500
00:28:52,340 --> 00:28:53,461
Well, take care.
501
00:28:53,461 --> 00:28:55,362
This is Sheridan and signing off.
502
00:28:55,363 --> 00:28:57,274
See you on the next episode of
503
00:28:57,274 --> 00:28:59,101
the AI Advantage podcast.
00:00:00,288 --> 00:00:04,111
AI it's not magic It's not magic.
2
00:00:04,111 --> 00:00:14,117
It responds based on how clearly you communicate right bad Some of you may have heard bad
input bad output right bad data in bad data out
3
00:00:14,117 --> 00:00:15,310
Here's a bad prompt.
4
00:00:15,310 --> 00:00:17,936
Write a marketing plan for my business.
5
00:00:17,937 --> 00:00:20,064
Oh, yeah, I mean, there's not enough information there.
6
00:00:20,064 --> 00:00:20,305
It's good.
7
00:00:20,305 --> 00:00:21,550
It's bound to fail.
8
00:00:21,550 --> 00:00:22,503
Yeah, that's a good.
9
00:00:22,503 --> 00:00:24,126
That's a good bad one.
10
00:00:24,127 --> 00:00:25,748
Super vague, right?
11
00:00:25,748 --> 00:00:27,008
Super vague.
12
00:00:27,169 --> 00:00:29,810
So here's how you take that and make it a good prompt, right?
13
00:00:30,309 --> 00:00:32,522
Welcome to the AI Advantage podcast.
14
00:00:32,522 --> 00:00:36,291
I'm your host Sheridan Wint, and I've got my friend Jon Foster here.
15
00:00:36,291 --> 00:00:43,117
And we're going to be talking about how to talk to AI and get better results, also known
as prompt engineering.
16
00:00:43,117 --> 00:00:45,360
You may have heard of prompt engineering before.
17
00:00:45,361 --> 00:00:53,884
little of the experience that yesterday when we asked Chad GPT and Gemini and I the two
different things and once we rephrased it I can definitely see how this is going to be
18
00:00:53,884 --> 00:00:54,735
helpful.
19
00:00:54,736 --> 00:00:57,837
Yeah, yeah, yeah.
20
00:00:57,837 --> 00:00:59,228
That was on our last episode.
21
00:00:59,228 --> 00:01:07,842
I think it's episode four in the series, where we did a side by side of Gemini and Chat
GPT to create an image of Florida man.
22
00:01:07,843 --> 00:01:11,414
And in honor of that image, Jon is dressed like Florida man today.
23
00:01:11,415 --> 00:01:12,688
Ha ha ha ha!
24
00:01:12,886 --> 00:01:13,991
Get on get.
25
00:01:13,993 --> 00:01:18,451
All right, so digging into why prompting matters
26
00:01:18,729 --> 00:01:22,552
AI it's not magic It's not magic.
27
00:01:22,552 --> 00:01:32,558
It responds based on how clearly you communicate right bad Some of you may have heard bad
input bad output right bad data in bad data out
28
00:01:32,981 --> 00:01:37,510
It's the same thing when you're prompting a model right any model
29
00:01:37,510 --> 00:01:41,288
Good prompts, better results, less editing, way more accuracy.
30
00:01:41,289 --> 00:01:44,757
You put any funny prompts in, Jon, that you got some crazy outputs on?
31
00:01:44,758 --> 00:01:48,085
Uh, mine, my favorite is when I misspell things.
32
00:01:48,085 --> 00:01:50,620
Cause I type too fast or my phone's just too small.
33
00:01:50,620 --> 00:01:51,733
I don't know what it is.
34
00:01:51,733 --> 00:01:56,021
And then something completely what I did not ask comes up and I'm like, Oh.
35
00:01:56,022 --> 00:01:56,902
Yeah.
36
00:01:57,985 --> 00:02:00,968
Usually the spelling, usually for me, I haven't seen like a spelling thing.
37
00:02:00,968 --> 00:02:02,490
Usually it's like, I know what you meant.
38
00:02:02,490 --> 00:02:04,953
And it just keeps on going.
39
00:02:05,294 --> 00:02:07,076
Must be misspelling some.
40
00:02:07,076 --> 00:02:11,662
There's some words though that you move one or two letters around and the whole meaning of
the word changes.
41
00:02:11,662 --> 00:02:12,422
m
42
00:02:12,652 --> 00:02:15,593
Yep, like, um, like, uh, like duck.
43
00:02:15,594 --> 00:02:16,571
like duck.
44
00:02:16,572 --> 00:02:18,170
How would you rearrange duck?
45
00:02:18,475 --> 00:02:25,808
I mean, you click one key to the right, you're in a little bit of trouble.
46
00:02:25,808 --> 00:02:26,678
I see, I see.
47
00:02:26,678 --> 00:02:28,048
Got you.
48
00:02:28,049 --> 00:02:29,009
Typo.
49
00:02:29,009 --> 00:02:30,109
Typo.
50
00:02:30,110 --> 00:02:30,560
All right.
51
00:02:30,560 --> 00:02:32,331
So let's talk about prompts, right?
52
00:02:32,331 --> 00:02:35,772
What actually is a prompt?
53
00:02:35,773 --> 00:02:36,893
Any ideas?
54
00:02:36,936 --> 00:02:38,376
a prompt.
55
00:02:38,777 --> 00:02:48,623
Well, I would assume it would be setting the AI up for, you know, a request of information
that you, whatever you put in.
56
00:02:48,623 --> 00:02:49,813
So input.
57
00:02:49,814 --> 00:02:52,785
is input, yeah, it's anything, any input that you give it.
58
00:02:52,785 --> 00:03:02,411
So a good way to structure your prompt are instructions and then context and then your
desired outcome, right?
59
00:03:02,411 --> 00:03:07,583
Instructions plus context plus desired outcome equals a good prompt.
60
00:03:07,584 --> 00:03:07,814
Right?
61
00:03:07,814 --> 00:03:12,367
So you want to, it tends to work best like any given model, right?
62
00:03:12,367 --> 00:03:19,980
AI tends to work best when it knows who it's writing for, what you want, how you want it
delivered, and what good looks like, right?
63
00:03:19,980 --> 00:03:21,251
What you expect.
64
00:03:21,252 --> 00:03:28,806
So we're going to dive in to a four-part framework, which we touched a little bit on, but
it's going to sound a little different, right?
65
00:03:28,806 --> 00:03:32,178
Number one, the role or the perspective, right?
66
00:03:32,471 --> 00:03:40,793
Act as a marketing strategist, or act as a customer support expert, or act as a mad
scientist from 1970.
67
00:03:40,793 --> 00:03:41,923
And then give it the context.
68
00:03:41,923 --> 00:03:42,634
What's the goal?
69
00:03:42,634 --> 00:03:43,494
Who's the audience?
70
00:03:43,494 --> 00:03:45,595
What problem are we solving?
71
00:03:45,595 --> 00:03:47,596
Give it a little bit of context.
72
00:03:47,598 --> 00:03:49,598
Well, we'll get into some examples here in a minute, right?
73
00:03:49,598 --> 00:03:51,538
So number one, we got the role and the perspective.
74
00:03:51,538 --> 00:03:53,198
Number two, we've got context.
75
00:03:53,198 --> 00:03:54,418
Number three, we've got task.
76
00:03:54,418 --> 00:03:56,718
What exactly do you want it to do?
77
00:03:56,718 --> 00:03:58,498
And number four, the format, right?
78
00:03:58,498 --> 00:03:59,558
Do you need bullet points?
79
00:03:59,558 --> 00:04:00,178
Do you need a script?
80
00:04:00,178 --> 00:04:01,298
Do you need an email, a table?
81
00:04:01,298 --> 00:04:03,418
Do you need steps, a blog article?
82
00:04:03,418 --> 00:04:04,878
What is it, right?
83
00:04:04,879 --> 00:04:12,799
So we're going to talk about some good prompts and some bad prompts and just do a little
comparison, right?
84
00:04:12,991 --> 00:04:13,798
Okay.
85
00:04:13,799 --> 00:04:15,504
Can you give me a bad prompt?
86
00:04:15,505 --> 00:04:17,213
Oh, I could probably do that all day.
87
00:04:17,213 --> 00:04:19,662
um
88
00:04:19,663 --> 00:04:20,816
No, I can't.
89
00:04:20,816 --> 00:04:22,198
Not on the spot.
90
00:04:22,240 --> 00:04:23,433
I thought I could.
91
00:04:23,433 --> 00:04:24,624
I failed you.
92
00:04:24,782 --> 00:04:26,116
right, all right, a bad prompt.
93
00:04:26,088 --> 00:04:27,281
Here's a bad prompt.
94
00:04:27,281 --> 00:04:29,907
Write a marketing plan for my business.
95
00:04:29,908 --> 00:04:32,035
Oh, yeah, I mean, there's not enough information there.
96
00:04:32,035 --> 00:04:32,276
It's good.
97
00:04:32,276 --> 00:04:33,521
It's bound to fail.
98
00:04:33,521 --> 00:04:34,474
Yeah, that's a good.
99
00:04:34,474 --> 00:04:36,097
That's a good bad one.
100
00:04:36,098 --> 00:04:37,719
Super vague, right?
101
00:04:37,719 --> 00:04:38,979
Super vague.
102
00:04:39,140 --> 00:04:41,781
So here's how you take that and make it a good prompt, right?
103
00:04:41,779 --> 00:04:46,152
Let's go back and look at our four part prompt framework, right?
104
00:04:46,152 --> 00:04:52,405
Role, task, and format.
105
00:04:52,446 --> 00:04:55,558
So you are a marketing strategist, right?
106
00:04:55,558 --> 00:04:56,648
That's the role.
107
00:04:56,714 --> 00:04:57,666
Mm-hmm.
108
00:04:57,667 --> 00:05:01,793
Helping a small real estate team struggling with lead consistency.
109
00:05:01,793 --> 00:05:03,515
That's some context.
110
00:05:04,357 --> 00:05:11,517
The task, create a 30 day marketing plan focused on email nurturing and social content.
111
00:05:11,518 --> 00:05:12,536
and then the format.
112
00:05:12,536 --> 00:05:14,961
Format it as a weekly checklist.
113
00:05:15,243 --> 00:05:19,709
So that's going to put out something way better than write a marketing plan for my
business.
114
00:05:20,011 --> 00:05:21,011
100%.
115
00:05:21,011 --> 00:05:25,966
So a bad prompt.
116
00:05:25,966 --> 00:05:29,489
We can walk through another example again together.
117
00:05:29,490 --> 00:05:31,218
where can I buy a shirt?
118
00:05:31,219 --> 00:05:33,091
Where can I buy your shirt?
119
00:05:33,193 --> 00:05:36,147
Definitely a shirt.
120
00:05:36,279 --> 00:05:37,843
Not just any shirt.
121
00:05:37,844 --> 00:05:38,955
Where can I buy a shirt?
122
00:05:38,955 --> 00:05:40,377
Yeah, definitely.
123
00:05:40,377 --> 00:05:42,379
You don't know what you're gonna get with that one, right?
124
00:05:42,455 --> 00:05:43,882
No, not at all.
125
00:05:44,269 --> 00:05:46,363
So role, let's go through it.
126
00:05:46,363 --> 00:05:46,915
What's a role?
127
00:05:46,915 --> 00:05:48,407
What's the person's role?
128
00:05:48,527 --> 00:05:49,755
or perspective.
129
00:05:49,755 --> 00:05:52,082
from the perspective of...
130
00:05:52,083 --> 00:05:54,635
What was the name for somebody that's in the fashion?
131
00:05:54,842 --> 00:05:59,648
so in the role of someone who was up to date on today's fashion.
132
00:05:59,649 --> 00:06:00,322
Okay.
133
00:06:00,322 --> 00:06:01,843
Would that be a good start?
134
00:06:01,843 --> 00:06:02,956
Would that be a good roll?
135
00:06:02,956 --> 00:06:04,838
Or would you improve that?
136
00:06:05,299 --> 00:06:08,735
Yeah, mean, you could say a model, as a fashion model.
137
00:06:08,736 --> 00:06:15,234
mean, that's the thing, like, my thing is like I overthink things and I'm like, the
fashion model wouldn't do it, they just put, they just wear the clothes.
138
00:06:15,235 --> 00:06:18,186
but you want to buy the fashionable clothes, right?
139
00:06:18,187 --> 00:06:20,973
Yeah, but a fashion designer, let's say that.
140
00:06:23,200 --> 00:06:24,778
As a fashion designer.
141
00:06:24,778 --> 00:06:26,260
and then context.
142
00:06:26,473 --> 00:06:31,933
And then pick out a shirt that would look good this season.
143
00:06:31,934 --> 00:06:32,854
Okay, let's season it.
144
00:06:32,854 --> 00:06:33,615
on
145
00:06:33,617 --> 00:06:37,363
the recent fashion week event that just happened
146
00:06:37,721 --> 00:06:38,063
Okay.
147
00:06:38,063 --> 00:06:42,498
I thought you meant like fall, summer, spring, winter, but fashion season.
148
00:06:42,498 --> 00:06:43,261
I got it.
149
00:06:43,261 --> 00:06:44,242
I got it.
150
00:06:46,595 --> 00:06:48,158
I thought you meant weather seasons.
151
00:06:48,158 --> 00:06:48,939
uh
152
00:06:57,447 --> 00:06:58,433
Here's what we got so far.
153
00:06:59,186 --> 00:07:04,126
Role number one, as a fashion designer, to the context, what problem are we solving?
154
00:07:04,266 --> 00:07:10,986
I need to find a new shirt to wear to an event.
155
00:07:12,526 --> 00:07:16,666
Task, find me a shirt and explain the event, right?
156
00:07:16,666 --> 00:07:18,506
Maybe is it like a black tie event?
157
00:07:18,506 --> 00:07:19,246
Is it a ball?
158
00:07:19,246 --> 00:07:22,486
Is it just a kickback at somebody's house?
159
00:07:22,486 --> 00:07:25,366
Is it like a at?
160
00:07:25,366 --> 00:07:28,448
So you want to give it that context, what kind of event is it?
161
00:07:28,448 --> 00:07:35,116
And then the task, find me a shirt that I can wear that's blue.
162
00:07:35,116 --> 00:07:36,676
Yeah, OK.
163
00:07:36,997 --> 00:07:48,478
So even if the problem has not as much information, you've got to kind of add more detail
into it to be able to generate the better results.
164
00:07:48,478 --> 00:07:49,076
I'm assuming.
165
00:07:49,076 --> 00:07:49,826
Yeah, yeah.
166
00:07:49,826 --> 00:07:51,667
So here's another way we could do it, right?
167
00:07:51,667 --> 00:08:02,774
uh As a fashion model, find me, I need a, as a fashion model, that's the role, I need a
blue dress for a cocktail party.
168
00:08:02,774 --> 00:08:04,154
That's the context.
169
00:08:04,154 --> 00:08:09,036
Task, find me a blue dress I can wear.
170
00:08:09,037 --> 00:08:12,799
And then the format, provide this as a list of
171
00:08:12,800 --> 00:08:17,140
URLs where I can go look at the dresses.
172
00:08:17,402 --> 00:08:18,299
Okay
173
00:08:18,300 --> 00:08:19,640
Now you can shorten that up a little bit too.
174
00:08:19,640 --> 00:08:26,760
You could say, as a fashion model, I need a blue dress for a cocktail party.
175
00:08:26,860 --> 00:08:31,740
Provide me with a list of dress options.
176
00:08:31,800 --> 00:08:31,940
Right?
177
00:08:31,940 --> 00:08:36,040
Now that last sentence, you put the task and the format in the same sentence, right?
178
00:08:36,040 --> 00:08:40,300
The task is give me a list of dresses and the format is a list, right?
179
00:08:40,300 --> 00:08:41,880
Of dresses though.
180
00:08:41,881 --> 00:08:44,141
So yeah, so it doesn't have to be crazy.
181
00:08:44,141 --> 00:08:48,033
You can kind of combine a lot of that stuff, but basically the more information you give
it.
182
00:08:48,477 --> 00:08:50,432
better result you're going to get.
183
00:08:50,433 --> 00:08:55,566
Yep, yep, yep, All right, prompting mistakes, common prompting mistakes.
184
00:08:55,566 --> 00:08:57,247
Being too vague, right?
185
00:08:57,247 --> 00:08:58,847
Give me a business plan.
186
00:08:58,947 --> 00:09:02,118
Well, we need to know what business it is and what you're planning for, right?
187
00:09:02,118 --> 00:09:02,798
What do you sell?
188
00:09:02,798 --> 00:09:03,849
All of that good stuff.
189
00:09:03,849 --> 00:09:04,609
What are your services?
190
00:09:04,609 --> 00:09:05,489
What are your offerings?
191
00:09:05,489 --> 00:09:06,330
What's your budget?
192
00:09:06,330 --> 00:09:09,311
Asking for too much at once, right?
193
00:09:09,371 --> 00:09:13,233
Asking for it to give you a plan.
194
00:09:13,605 --> 00:09:24,016
for how you're going to get from a high school graduate to landing on the moon with just
your own money, your own budget.
195
00:09:24,017 --> 00:09:24,938
Right?
196
00:09:26,259 --> 00:09:33,414
It might give you some sort of plan, but it's gonna be pretty tough to execute because it
will lack a lot of details, right?
197
00:09:33,415 --> 00:09:35,716
Another common mistake, leaving out the audience, right?
198
00:09:35,716 --> 00:09:43,862
Telling it to generate something, but not telling the model who the intended audience is.
199
00:09:43,862 --> 00:09:44,562
Who's gonna read it?
200
00:09:44,562 --> 00:09:45,592
Who's gonna look at this thing?
201
00:09:45,592 --> 00:09:49,534
Another common mistake, forgetting to specify the tone, right?
202
00:09:49,534 --> 00:09:51,946
You want to say is it going to be friendly, right?
203
00:09:51,946 --> 00:10:00,552
Let's say you're drafting an email, is it supposed to be professional, is it supposed to
be friendly, is it supposed to be authoritative, you know, what kind of tone do you want
204
00:10:00,552 --> 00:10:01,084
it to have?
205
00:10:01,084 --> 00:10:02,074
Examples, right?
206
00:10:02,074 --> 00:10:04,866
Not giving examples, that's another common mistake.
207
00:10:05,080 --> 00:10:12,166
it gets much better if you give it an example and say, here I'm looking for a JSON file
structured in this particular way, right?
208
00:10:12,166 --> 00:10:17,852
And if you give it the example, well, the example should have all the values, all the keys
in it that you're looking for, right?
209
00:10:17,852 --> 00:10:28,462
So then it can say, OK, well, let me just replace the values and keep all the keys and
your output will be structured exactly like probably you wanted it to be because you gave
210
00:10:28,462 --> 00:10:29,632
it an example.
211
00:10:29,634 --> 00:10:35,788
Another common mistake is not telling it, you know, not telling the model what you don't
want it to do, right?
212
00:10:35,788 --> 00:10:38,589
Don't format this as HTML.
213
00:10:38,710 --> 00:10:47,324
Don't, right, don't give me a list of texts, give me a list of hyperlinks, right?
214
00:10:47,324 --> 00:10:49,706
So you're telling it what not to do and what to do.
215
00:10:49,706 --> 00:10:53,248
Another common mistake is not iterating enough, right?
216
00:10:53,248 --> 00:11:00,988
first response that you get from a model, not always the best response, but you can modify
it, you can adjust it, it's a conversation, right?
217
00:11:00,988 --> 00:11:03,288
It's not like a one-off request.
218
00:11:03,579 --> 00:11:11,731
So when you're saying you have to ask it a multitude of times or just a couple times about
the same thing, redundancy?
219
00:11:11,732 --> 00:11:20,829
Not always, but if you don't like the response that you get and you want it to be better,
you can tell it exactly in what way you want it to be better.
220
00:11:20,829 --> 00:11:30,077
So let's say in our example earlier, it gives us a list of dresses and it says, you know,
this dress by this designer, this dress by that designer, right?
221
00:11:30,077 --> 00:11:32,688
Dress number three by designer number three.
222
00:11:32,729 --> 00:11:34,230
And it's just giving you list of text.
223
00:11:34,230 --> 00:11:35,271
Well, that's not what you wanted.
224
00:11:35,271 --> 00:11:40,795
What you wanted was a list of links, hyperlinks that you can click on and then it opens it
up.
225
00:11:40,985 --> 00:11:53,412
in uh Sephora or I don't know some fashion brand I have no idea so if it gives you not
what you wanted then you can just say hey format this you know I like to actually just put
226
00:11:53,412 --> 00:12:05,128
the word output with a colon instead of having to say format it like this I just put
output colon list of urls right very brief
227
00:12:05,128 --> 00:12:07,066
And what is what is that by doing that?
228
00:12:07,066 --> 00:12:07,447
What is that?
229
00:12:07,447 --> 00:12:09,515
What does that do specifically?
230
00:12:09,515 --> 00:12:19,333
Well, instead of giving you a list that says, you know, this blue dress by designer one,
this blue dress by designer two, instead it will literally just give you a list of
231
00:12:19,333 --> 00:12:20,935
hyperlinks you can click on.
232
00:12:20,936 --> 00:12:30,234
Okay, so but but the the colon adding the colon in there and stuff like that, is that is
that a key to all the non-it people?
233
00:12:30,235 --> 00:12:37,650
So I like to, let's say we're talking about context, we're talking about giving it more
information, better data in, better data out.
234
00:12:37,650 --> 00:12:47,814
Pro tip that I like to do is I'll just write a single sentence or question, but I give it
context within the question.
235
00:12:47,814 --> 00:12:53,137
Let me go, let me pull up an example and just go look at something recently that I did.
236
00:12:53,138 --> 00:12:56,162
Alright, so I wanted to...
237
00:12:56,163 --> 00:12:57,300
you find it.
238
00:12:57,301 --> 00:13:06,030
I was working on a growth plan for a client and I put in a sentence that says, output
this...
239
00:13:06,312 --> 00:13:09,394
with formatting instead of markdown, right?
240
00:13:09,394 --> 00:13:18,076
And so it's one sentence, output this as formatting instead of markdown, which it might
sound like a bad prompt, right?
241
00:13:18,076 --> 00:13:27,861
But in between those two words where I said, this, after this, I put parentheses, and
inside the parentheses, I put like a massive amount of text.
242
00:13:27,861 --> 00:13:32,402
It's like 10 or 12 paragraphs, and it's all written in
243
00:13:32,402 --> 00:13:43,508
Markdown format and all I wanted it to do was give me To actually give it to me as bold
right because it gave it to me in plain text so when you look at the text it's like Star
244
00:13:43,508 --> 00:13:48,329
star star week one star star star and that makes it bold right?
245
00:13:49,130 --> 00:13:52,191
And I didn't want all that in there, so I just wrote a sentence.
246
00:13:52,191 --> 00:13:58,733
said output this with the formatting instead of in markdown and I put all of that text in
parentheses
247
00:13:58,733 --> 00:14:10,492
in there so it knows what text I'm talking about and it knows that I don't want it to look
like that right I want it so parentheses I use all over the place I'll use a colon I might
248
00:14:10,492 --> 00:14:21,740
say if I'm building a website I might say make these make the following changes colon
right and then parentheses and I'll put all the changes in there and then close the
249
00:14:21,740 --> 00:14:26,063
parentheses and put a period and then I might also give it some more to do
250
00:14:26,262 --> 00:14:28,979
Although you could probably put all of the changes within the parentheses.
251
00:14:28,979 --> 00:14:30,784
um
252
00:14:30,784 --> 00:14:35,519
say markdown like for for the non-it people what does that mean
253
00:14:35,520 --> 00:14:37,237
Also me, because I don't know.
254
00:14:37,239 --> 00:14:40,349
So Markdown is just like a format, a syntax.
255
00:14:40,349 --> 00:14:42,123
It just...
256
00:14:42,124 --> 00:14:43,572
I don't know.
257
00:14:43,572 --> 00:14:46,643
Let me ask Chad JBT to explain it better than I can.
258
00:14:46,644 --> 00:14:50,387
I probably should give you a different example, a way more simple example.
259
00:14:50,388 --> 00:14:53,665
Show us an example of a good problem for solving this problem.
260
00:14:54,029 --> 00:14:54,766
So.
261
00:14:54,767 --> 00:14:55,731
Let's see.
262
00:14:55,732 --> 00:14:59,495
So let me give you an example using the colon and the parentheses, right?
263
00:14:59,495 --> 00:15:02,997
Back to our fashion designer, right?
264
00:15:02,997 --> 00:15:07,000
Let's say that they, let's say that they are our fashion model.
265
00:15:07,000 --> 00:15:19,088
Let's say that we made the prompt like this, help me find a dress that meets these
requirements and doesn't look like it's been bedazzled and wasn't in this list of dresses,
266
00:15:19,109 --> 00:15:19,589
right?
267
00:15:19,589 --> 00:15:21,574
And so that's like one sentence.
268
00:15:21,574 --> 00:15:23,497
It's like one run-on sentence.
269
00:15:23,497 --> 00:15:30,590
But if you use parentheses in there, in the right places, then you can give it much better
requirements.
270
00:15:30,590 --> 00:15:32,954
So let's start a new chat here.
271
00:15:32,954 --> 00:15:34,295
I'll share my screen.
272
00:15:34,671 --> 00:15:41,844
So like, putting like things like no sequins and other little details that go in the
dresses that you do not want.
273
00:15:41,844 --> 00:15:46,242
Like, you don't want shiny, sparkly, you want something more matte.
274
00:15:46,438 --> 00:15:47,339
Okay.
275
00:15:47,646 --> 00:15:49,033
Find me a-
276
00:15:49,034 --> 00:15:54,462
provide, let's do it this way, as a fashion model.
277
00:15:54,463 --> 00:16:04,208
provide me a list of dresses that meet these requirements, parentheses, right?
278
00:16:04,749 --> 00:16:09,201
That I can wear to a cocktail party.
279
00:16:09,202 --> 00:16:14,951
Make sure it doesn't have any of these features.
280
00:16:15,213 --> 00:16:16,278
Parentheses.
281
00:16:16,539 --> 00:16:33,038
and wasn't isn't on this list of dresses from last year right so in the requirements here
i might say
282
00:16:33,039 --> 00:16:33,971
Blue.
283
00:16:33,972 --> 00:16:47,244
sleek not fluffy right no has lace i don't know i don't know i'm not not really into
dresses so i'm just making this up as i go but
284
00:16:47,244 --> 00:16:49,119
Just mini skirts.
285
00:16:49,420 --> 00:16:50,642
Yeah, mini skirts.
286
00:16:50,642 --> 00:16:54,146
ah Covers the knees, right?
287
00:16:54,146 --> 00:16:55,287
Let's keep it classy.
288
00:16:55,287 --> 00:16:57,829
Covers the knees.
289
00:16:58,491 --> 00:16:59,583
That I can wear to a cocktail
290
00:16:59,583 --> 00:17:00,475
yeah.
291
00:17:01,059 --> 00:17:06,068
Oh, yeah, we'll put is at least knee length.
292
00:17:06,069 --> 00:17:06,732
uh...
293
00:17:06,732 --> 00:17:11,945
make sure doesn't have any these features sequence i'm probably gonna spell that wrong but
whatever ah
294
00:17:11,945 --> 00:17:12,837
I have no idea.
295
00:17:12,837 --> 00:17:15,113
Actually, I think you spelled it right, but I'm not sure.
296
00:17:15,113 --> 00:17:16,873
bedazzled.
297
00:17:16,874 --> 00:17:17,694
glitter.
298
00:17:17,694 --> 00:17:24,474
Alright, doesn't have any of that and is not on this list of dresses from last year.
299
00:17:24,474 --> 00:17:25,724
We'll say...
300
00:17:25,725 --> 00:17:28,158
I don't know anything about dresses.
301
00:17:28,158 --> 00:17:33,845
uh Not from this list of brands.
302
00:17:33,845 --> 00:17:35,026
We'll do that.
303
00:17:35,467 --> 00:17:38,710
So no Nike dresses, no Adidas.
304
00:17:38,892 --> 00:17:43,198
Jimmy Choo, Versace, you know.
305
00:17:43,368 --> 00:17:45,749
yeah, sure, no Versace.
306
00:17:47,330 --> 00:17:49,901
So anyway, so we just put all of that into parentheses.
307
00:17:49,901 --> 00:17:59,555
So now this prompt, looks pretty long, but it's really just a sentence, know, find me the
dress that meets these requirements that I can wear to a cocktail party.
308
00:17:59,555 --> 00:18:02,696
Two sentences, make sure it doesn't have any of these features, right?
309
00:18:02,817 --> 00:18:04,538
And isn't on this list of brands.
310
00:18:04,538 --> 00:18:06,418
So that's very specific.
311
00:18:06,418 --> 00:18:07,959
Now it's gonna give me a list.
312
00:18:07,959 --> 00:18:08,519
right?
313
00:18:08,519 --> 00:18:11,695
And it's probably, I don't know if they'll have URLs or not.
314
00:18:11,696 --> 00:18:12,916
I'm curious how this is going to go.
315
00:18:12,916 --> 00:18:14,816
Looking and address recommendation.
316
00:18:14,817 --> 00:18:17,944
focusing on reputable retailers.
317
00:18:17,944 --> 00:18:19,026
It's telling us how it's thinking.
318
00:18:19,026 --> 00:18:19,487
Look at this.
319
00:18:19,487 --> 00:18:24,147
ah All these sites that it's um searching.
320
00:18:24,148 --> 00:18:26,574
double check that they don't have sequins or glitter.
321
00:18:26,817 --> 00:18:27,670
It's very cool, man.
322
00:18:27,670 --> 00:18:30,598
You can see the thought process it's going through.
323
00:18:31,080 --> 00:18:33,277
Yeah, that is extremely neat.
324
00:18:33,278 --> 00:18:34,093
Hmm.
325
00:18:34,094 --> 00:18:43,333
It's said it came across some dresses that it's not sure if it has it but that brand
typically does have some sparkly stuff going on so it's just gonna stay away from that
326
00:18:43,333 --> 00:18:44,214
brand.
327
00:18:44,344 --> 00:18:45,301
Okay.
328
00:18:45,302 --> 00:18:46,625
Searching some more.
329
00:18:46,626 --> 00:18:48,199
Saks Fifth Avenue.
330
00:18:48,466 --> 00:18:49,348
Okay.
331
00:18:49,349 --> 00:18:52,289
Here we go, it says each product page is going to be cited.
332
00:18:52,290 --> 00:19:01,490
All right, it's giving a list of dresses and a description of why that dress meets the
requirements, and there's a URL.
333
00:19:01,491 --> 00:19:02,284
Dope.
334
00:19:02,772 --> 00:19:03,969
So we have five dresses here.
335
00:19:03,969 --> 00:19:05,805
Now I can say output.
336
00:19:06,189 --> 00:19:08,736
list of URLs only.
337
00:19:08,737 --> 00:19:11,051
is maybe I want to email this to somebody, right?
338
00:19:11,051 --> 00:19:12,229
Okay, yeah.
339
00:19:12,230 --> 00:19:14,262
So now we're iterating on the prompt.
340
00:19:14,262 --> 00:19:16,983
We're telling it exactly what output we want.
341
00:19:16,984 --> 00:19:17,693
thinking.
342
00:19:17,693 --> 00:19:19,565
All right, there's the URLs.
343
00:19:19,566 --> 00:19:26,994
Write this as write this list as an email to a
344
00:19:26,995 --> 00:19:35,538
Fashion model friend So we're just iterating on what it came out with and now here we go
now It basically did everything for us.
345
00:19:35,538 --> 00:19:37,698
It drafted the email.
346
00:19:37,818 --> 00:19:40,759
It says hey friend's name, right?
347
00:19:40,759 --> 00:19:50,242
We didn't tell it our friend's name I've been scouting some cocktail ready dresses and
found a bunch that I think you'll like Blue sleek lace knee-length or longer completely
348
00:19:50,242 --> 00:19:53,713
free of sequins glitter and bedazzling perfect for a chic evening
349
00:19:53,714 --> 00:19:57,532
And like that, we now know where to go get dresses.
350
00:19:58,015 --> 00:20:00,539
I'm sure everybody was super interested in that.
351
00:20:00,539 --> 00:20:03,182
uh
352
00:20:03,183 --> 00:20:04,803
It is 2025.
353
00:20:05,985 --> 00:20:07,607
Ladies, here you go.
354
00:20:07,607 --> 00:20:13,232
The top one on here is self, number one is self portrait by blue fine lace midi dress.
355
00:20:13,232 --> 00:20:15,673
no, maybe that brand is self portrait.
356
00:20:16,074 --> 00:20:17,917
Anyways, dress.
357
00:20:17,917 --> 00:20:20,639
at Nordstrom, so there's one by my house.
358
00:20:21,058 --> 00:20:21,990
Yeah.
359
00:20:22,215 --> 00:20:23,123
All let me stop sharing.
360
00:20:23,483 --> 00:20:25,004
All right, let me stop sharing.
361
00:20:25,004 --> 00:20:30,529
So anyway, that's some examples of how you can make your prompts better, how you can avoid
some of the mistakes.
362
00:20:30,529 --> 00:20:32,450
Let's do some, there's some templates.
363
00:20:32,450 --> 00:20:34,171
Some people actually sell these templates.
364
00:20:34,171 --> 00:20:36,113
I don't think that's necessary for you to do that.
365
00:20:36,113 --> 00:20:41,657
You can just ask AI to help you create it into a better prompt, right?
366
00:20:41,657 --> 00:20:45,720
But if you wanna go buy templates, feel free, go buy them.
367
00:20:45,720 --> 00:20:47,422
Maybe we'll put some on the website, who knows?
368
00:20:47,422 --> 00:20:50,424
I think it's a little, I think it's a, I
369
00:20:50,425 --> 00:20:51,971
seems to me it's like...
370
00:20:52,519 --> 00:20:53,309
huh.
371
00:20:53,511 --> 00:20:56,231
What happens if we ask AI?
372
00:20:56,232 --> 00:21:03,452
say, like, yeah, like, like, how can I, how can I best format my questions for you to help
me?
373
00:21:03,740 --> 00:21:06,522
So there's a cool setting in a lot of these models.
374
00:21:06,522 --> 00:21:11,216
A chat GPT, for example, you can go into the settings and you can say preferences, right?
375
00:21:11,216 --> 00:21:14,839
And you can tell it, you know, talk to me like a five-year-old, right?
376
00:21:14,839 --> 00:21:19,994
You can also say, if I don't give you enough information, ask a clarifying question,
right?
377
00:21:19,994 --> 00:21:22,947
Or ask clarifying questions if you don't have enough information.
378
00:21:22,947 --> 00:21:27,368
Another thing I like to do is I'll use one model to...
379
00:21:27,368 --> 00:21:29,759
draft a prompt and put it into a different model.
380
00:21:29,759 --> 00:21:35,181
So let's say I'm like building a website and I'm using AI to like build a whole web app or
something.
381
00:21:35,181 --> 00:21:37,062
I'll format all of it.
382
00:21:37,062 --> 00:21:47,067
Like I'll get it all ready to go in and say like Claude or ChatGBT and say, help me create
a prompt for this tool that I'm using to build a website.
383
00:21:47,067 --> 00:21:50,549
I want the website to have to meet these requirements, right?
384
00:21:50,549 --> 00:21:52,229
I want it to match my brand.
385
00:21:52,229 --> 00:21:56,291
I want it to align with my tone and you've got to tell it, you these are my brand colors.
386
00:21:56,363 --> 00:21:58,044
and this is the tone I use.
387
00:21:58,044 --> 00:22:06,569
Here's some, I'll go to a website and grab all the text on the website and say, here's my
existing website and everything that is on there now, right?
388
00:22:06,569 --> 00:22:08,231
And, you know, build out a site map.
389
00:22:08,231 --> 00:22:13,014
I want it to have these pages, contact us and, you know, features and offerings.
390
00:22:13,014 --> 00:22:19,304
And so you just put all that and then chat GPT or whatever model you're using will spit
out a prompt.
391
00:22:19,304 --> 00:22:24,520
And you can take that and put it in the other one and it'll build out the whole website,
the whole app for you.
392
00:22:24,521 --> 00:22:25,132
That's nuts.
393
00:22:25,132 --> 00:22:26,163
That's crazy.
394
00:22:26,163 --> 00:22:28,887
Is it able to verify its own creation?
395
00:22:28,887 --> 00:22:37,989
Like, is it able to like, you know how when you create a website and everything like that,
like the contact us button isn't mapped correctly, you know, and it doesn't, you know, is
396
00:22:37,989 --> 00:22:39,821
it able to check on itself?
397
00:22:39,822 --> 00:22:41,454
Some of the models can, yeah.
398
00:22:41,454 --> 00:22:44,237
Some of the tools can.
399
00:22:44,237 --> 00:22:46,919
One of the ones I use, it'll test it, right?
400
00:22:46,919 --> 00:22:58,131
It'll do like, when you do a software development cycle, got dev, test, QA maybe, uh or
UAT, and then production, right?
401
00:22:58,131 --> 00:23:04,677
Everybody's, all the organizations tend to have their own way that they do those, but this
tool that I use, will.
402
00:23:04,678 --> 00:23:11,021
it will develop it and then it will test it make sure that it works and then i'll say okay
it's ready to go and publish it
403
00:23:11,022 --> 00:23:18,882
That is that is amazing because I was I was just thinking in my head and everything like
one of the biggest things and one of the hardest things that I experienced when learning
404
00:23:18,882 --> 00:23:23,082
this and stuff like that was like user friendliness.
405
00:23:23,082 --> 00:23:30,742
You know, you could probably like ask AI and be like from, you know, the perspective of a
web developer.
406
00:23:30,743 --> 00:23:36,023
Um, make input, whatever the website or whatever.
407
00:23:36,043 --> 00:23:39,523
What would you suggest to make this more user friendly?
408
00:23:39,705 --> 00:23:40,446
Yeah.
409
00:23:40,447 --> 00:23:47,121
And then, you know, maybe a little bit more context sprinkled in, but you know, it
probably tell you exactly how to do it.
410
00:23:47,121 --> 00:23:48,302
That's crazy.
411
00:23:48,676 --> 00:23:50,218
It would, it would.
412
00:23:50,677 --> 00:23:58,827
All right, with that, I think that's a good point for us to hear a quick message from our
friends and sponsors, and we'll be right back after this.
413
00:23:58,957 --> 00:24:04,842
Right now, AI is deciding who your next customer trusts, and it might not be you.
414
00:24:04,842 --> 00:24:15,790
You might not realize it yet, but ChatGPT, Gemini, Perplexity, all of these models are
already deciding who they choose to recommend in industries like yours.
415
00:24:15,790 --> 00:24:17,741
And they don't always list options either.
416
00:24:17,741 --> 00:24:19,913
Sometimes they give one answer.
417
00:24:19,913 --> 00:24:25,847
And if that answer isn't you, then AI just referred your ideal customer to your
competitor.
418
00:24:25,847 --> 00:24:29,260
That's why we created the free AI Visibility Review.
419
00:24:29,260 --> 00:24:39,829
It's a short live session where we will sit down with you and show you what AI thinks of
you and your brand and your company, who it thinks is an authority, and what you can do to
420
00:24:39,829 --> 00:24:42,791
become the authority that it starts to recommend.
421
00:24:42,791 --> 00:24:44,853
Don't just guess where you stand.
422
00:24:44,853 --> 00:24:45,854
Find out.
423
00:24:45,854 --> 00:24:55,327
Schedule your free AI Visibility Review today at advantagelabs.ai and see if AI knows that
your business exists.
424
00:24:55,327 --> 00:25:00,643
Imagine what it would do for your business if chat GPT was actually referring people to
you.
425
00:25:00,664 --> 00:25:02,286
Advantage Labs .
426
00:25:02,286 --> 00:25:05,629
AI, your AI advantage.
427
00:25:06,552 --> 00:25:07,442
And we're back.
428
00:25:07,442 --> 00:25:09,761
This is the AI Advantage podcast.
429
00:25:09,761 --> 00:25:16,837
I'm your host, Sheridan, and we were just talking about prompt engineering, specifically
how do you talk to AI and get better results.
430
00:25:16,837 --> 00:25:25,861
The next thing that we were gonna go through were some ideas for how to do different
things with your prompts, right?
431
00:25:25,862 --> 00:25:27,056
Okay, let's get into it.
432
00:25:27,056 --> 00:25:29,876
Let's get into it.
433
00:25:29,936 --> 00:25:31,276
First up, brainstorming, right?
434
00:25:31,276 --> 00:25:34,036
Give me 10 ideas for blank, right?
435
00:25:34,036 --> 00:25:35,676
You know, it's a good way to start a prompt.
436
00:25:35,676 --> 00:25:40,956
Give me 10 ideas for this or let them give me 10, give me 100, give me 50, whatever,
right?
437
00:25:40,956 --> 00:25:42,456
Give it a number.
438
00:25:42,457 --> 00:25:44,777
Another thing that you can do is rewrite, right?
439
00:25:44,777 --> 00:25:49,437
Rewrite this to be clearer, shorter, more casual, that kind of thing.
440
00:25:49,438 --> 00:25:55,218
Obviously you gotta put the context in there of what you want it to rewrite.
441
00:25:55,274 --> 00:25:56,904
But we're not going to get into the whole prompt.
442
00:25:56,904 --> 00:26:00,817
We're just going to give you real quick some ideas on how you can start off a prompt.
443
00:26:00,817 --> 00:26:01,658
Summarization.
444
00:26:01,658 --> 00:26:03,519
Summarize this as a bullet point list.
445
00:26:03,519 --> 00:26:06,019
Turn this into a step-by-step workflow.
446
00:26:06,019 --> 00:26:11,462
If you want to create a standard operating procedure, create a process, turn this into a
step-by-step workflow.
447
00:26:11,462 --> 00:26:12,463
Comparisons.
448
00:26:12,463 --> 00:26:13,884
Compare option A and B.
449
00:26:13,884 --> 00:26:14,774
I do this a lot.
450
00:26:14,774 --> 00:26:17,285
I'll say compare and contrast.
451
00:26:17,359 --> 00:26:22,550
the these two softwares or these two products I'm thinking about buying, right?
452
00:26:22,550 --> 00:26:28,194
And give me like, give me a well, that's all I'll say I say compare and contrast these.
453
00:26:28,634 --> 00:26:29,454
Yeah.
454
00:26:29,454 --> 00:26:38,838
Well, yeah, it'll it'll do that on its own, but you can also specify like, you know, I'm
concerned about the price or I'm I really want to make sure I have this feature, right?
455
00:26:38,838 --> 00:26:42,081
So you can get that context in there, but I use it a lot.
456
00:26:42,081 --> 00:26:43,410
for comparing things.
457
00:26:43,410 --> 00:26:45,742
All right, now onto the power of iteration.
458
00:26:45,742 --> 00:26:50,544
We talked about that a little bit earlier with uh the example that we did with the dress,
right?
459
00:26:50,544 --> 00:26:52,085
Iterating on that prompt.
460
00:26:52,085 --> 00:27:00,830
So just a few things when you get some output and it's not exactly what you want, just
tell it how it can be exactly what you want.
461
00:27:00,830 --> 00:27:01,090
Right?
462
00:27:01,090 --> 00:27:04,065
Make it friendlier, make it shorter, make it more concise.
463
00:27:04,065 --> 00:27:08,191
Give me three variations to pick from or give it an example.
464
00:27:09,005 --> 00:27:14,764
A lot of times the second prompt is better than the first or the second output is better
than the first as well.
465
00:27:14,765 --> 00:27:18,546
and we learned that on the dresses part now for sure i agree to that
466
00:27:19,041 --> 00:27:20,224
Yeah, yeah, yeah.
467
00:27:20,224 --> 00:27:21,567
And we learned it with Florida Man.
468
00:27:21,567 --> 00:27:24,908
Now let's say that you are getting an output that you don't like, right?
469
00:27:24,908 --> 00:27:26,468
And you did everything you can think of.
470
00:27:26,468 --> 00:27:29,419
You used the four part prompt framework, right?
471
00:27:29,419 --> 00:27:34,791
You did the role, the role or perspective, the context, the task and the format.
472
00:27:34,791 --> 00:27:41,672
You did all of that and you iterated on it and it's still coming out bad and you don't
like the output, right?
473
00:27:41,672 --> 00:27:47,275
A simple thing you can do is say, what information do you need from me to finish this,
right?
474
00:27:47,275 --> 00:27:48,295
What do you need from me?
475
00:27:48,295 --> 00:27:51,088
What details do you need from me to make this better?
476
00:27:51,088 --> 00:27:54,167
uh You can just ask it and see what it says.
477
00:27:54,167 --> 00:27:57,667
Maybe it needs some details that have not been provided yet.
478
00:27:57,668 --> 00:28:00,268
Yeah, I like that feature for sure.
479
00:28:00,268 --> 00:28:08,908
Cause especially like, you mean, you, know what you want something and you need it done,
but you don't know exactly how to, how to, how to phrase it.
480
00:28:08,908 --> 00:28:09,368
You know what I mean?
481
00:28:09,368 --> 00:28:12,888
And that kind of gives you like a guideline to where you can like level it out.
482
00:28:12,888 --> 00:28:13,228
Yeah.
483
00:28:13,228 --> 00:28:14,332
I like that.
484
00:28:14,333 --> 00:28:15,723
Yeah, yes, sir.
485
00:28:15,864 --> 00:28:17,866
All right, I think that about does it for today.
486
00:28:17,866 --> 00:28:20,248
A few takeaways from today's episode.
487
00:28:20,248 --> 00:28:25,611
A good prompt is clarity plus context plus constraints.
488
00:28:25,611 --> 00:28:26,533
That's a good prompt.
489
00:28:26,533 --> 00:28:31,529
We've got the four prompt framework here that anyone can reference.
490
00:28:31,529 --> 00:28:32,069
any time.
491
00:28:32,069 --> 00:28:34,281
We're going to have a blog article about that that you can check out.
492
00:28:34,281 --> 00:28:35,904
That'll be a link in the description.
493
00:28:35,904 --> 00:28:38,646
Something to think about, AI is your partner.
494
00:28:38,646 --> 00:28:41,389
Think of it like a junior team member that can help you out.
495
00:28:41,389 --> 00:28:44,662
And I'd like to challenge you to go try a few prompts today.
496
00:28:44,662 --> 00:28:45,222
It's free.
497
00:28:45,222 --> 00:28:47,715
You can try a lot of these tools for free.
498
00:28:47,715 --> 00:28:51,620
Go try them out, see what you can do, and let us know how it goes.
499
00:28:51,620 --> 00:28:52,340
All right.
500
00:28:52,340 --> 00:28:53,461
Well, take care.
501
00:28:53,461 --> 00:28:55,362
This is Sheridan and signing off.
502
00:28:55,363 --> 00:28:57,274
See you on the next episode of
503
00:28:57,274 --> 00:28:59,101
the AI Advantage podcast.
