How Prompt Engineering Could Become Obsolete in the Near Future

Kasim: 00:00:11

I have an unpopular opinion, and I'm interested to

2

: 00:00:13

see how well this video ages.

3

: 00:00:14

Because maybe I just get nailed.

4

: 00:00:16

I think that prompt engineering

is not the wave of the future

5

: 00:00:21

that everybody believes it to be.

6

: 00:00:22

For the sake of the discussion,

we'll first define terms.

7

: 00:00:24

I took a semi definition

from GetConnected.

8

: 00:00:27

The goal of prompt engineering is to

control the output of the language model,

9

: 00:00:30

or AI tool, by providing it with specific

context, constraints, rules, and so on.

10

: 00:00:35

the problem.

11

: 00:00:35

From my perspective, a gold

rush on prompt engineering.

12

: 00:00:37

There's courses and training

and books and videos.

13

: 00:00:40

It's unbelievable to me how quickly

all that crap came out, by the way.

14

: 00:00:42

It's actually a marvel.

15

: 00:00:43

Like, I'm impressed at us as a community.

16

: 00:00:47

Info marketers who are like,

just, I mean, talk about, wow.

17

: 00:00:50

Here's the issue.

18

: 00:00:51

Prompt engineering is predicated

on the idea that AI tools

19

: 00:00:53

are simple and fallible.

20

: 00:00:54

And that's true for now.

21

: 00:00:56

The problem is, is, is AI tools

are also, if this isn't something

22

: 00:01:00

that we've ever encountered before.

23

: 00:01:01

And so the idea, the concept of product

engineering is, based on a paradigm that.

24

: 00:01:06

It looks at AI like software,

and AI is not software.

25

: 00:01:11

Software doesn't allow for dynamic,

iterative, and in many cases, growth.

26

: 00:01:19

That's the real word that I want there.

27

: 00:01:20

Compounding growth.

28

: 00:01:22

These AI tools are learning, and

they're learning at record rates.

29

: 00:01:26

It's unbelievable to see the

improvements, truly unbelievable.

30

: 00:01:30

So and these improvements

are all being crowdsourced.

31

: 00:01:32

that means that it's not just

learning from you, it's learning

32

: 00:01:34

from everybody, and then taking

what it's learned from everybody...

33

: 00:01:37

And then delivering for you.

34

: 00:01:38

every identifiable mistake any

AI ever makes with anyone only

35

: 00:01:43

needs to be made once, ostensibly.

36

: 00:01:44

Now obviously there's going to be

some repetition and might need to

37

: 00:01:47

reach critical mass before it even

acknowledges anything as a mistake.

38

: 00:01:50

But what that means is, as AI gets

smarter, the need for specific

39

: 00:01:54

context and constraints and

rules becomes less necessary.

40

: 00:01:57

Now, that doesn't mean that

prompt engineering becomes...

41

: 00:02:00

less necessary, it means that

the skill becomes less difficult.

42

: 00:02:07

It's like driving.

43

: 00:02:09

Everybody needs to know

how to drive, right?

44

: 00:02:11

Depending on where you live.

45

: 00:02:12

But everybody in a, drivable

city needs to know how to drive.

46

: 00:02:14

But as cars get easier to

drive, that skill becomes less

47

: 00:02:17

important and less valuable.

48

: 00:02:19

And as AI becomes smarter, that

skill becomes less valuable.

49

: 00:02:23

And given at the speed at which it's

already moving, which blows my mind

50

: 00:02:27

by the way, I think prompt engineering

is headed for near term obsolescence.

51

: 00:02:30

That's the problem.

52

: 00:02:31

It's not that we don't need...

53

: 00:02:32

Prompts it's that we don't need

engineering the engineering component

54

: 00:02:36

of the prompt engineering Construct

is based off of the idea that the AI

55

: 00:02:39

isn't going to be able to intuit what

it is that you want When in fact, that's

56

: 00:02:43

more or less the reason the AI exists

57

: 00:02:46

We're not gonna need the AI is going

to start prompting us It's gonna

58

: 00:02:51

get so good and in some instances

to some of the conversations.

59

: 00:02:54

I have a chat GPT are not entirely just

yet, but lean in the direction of...

60

: 00:02:59

Equal prompting, in order for chat

GPT to deliver on what it is that I'm

61

: 00:03:03

really asking, it requires clarity

and it's actually asking for that

62

: 00:03:05

clarity and as it gets better at

understanding, from prompt to completion,

63

: 00:03:11

this was the sequence of events.

64

: 00:03:13

It's like Google's path to purchase.

65

: 00:03:15

It knows like, okay, this is

what you're really asking.

66

: 00:03:17

And it will begin to engineer

the prompts on your behalf.

67

: 00:03:20

I'd like to make a soft landing

and say that I still think that

68

: 00:03:23

humanity's last bastion is creativity.

69

: 00:03:25

We still need to offer the machine intent.

70

: 00:03:28

And even for machines that can

be creative, some of imagery that

71

: 00:03:31

Midjourney can come up with is

unbelievable, but you have to ask for it.

72

: 00:03:34

You have to ask for it.

73

: 00:03:35

I want a water bottle that looks

like the queen's face being poured

74

: 00:03:37

out of the hand of a small nun.

75

: 00:03:39

In the sky on a hot air balloon

at sunset over the ocean and then

76

: 00:03:44

the AI is like, boom, here you go.

77

: 00:03:45

almost instantly.

78

: 00:03:46

It's amazing.

79

: 00:03:46

But you have to come up with that.

80

: 00:03:48

You have to come up with that.

81

: 00:03:49

and what's interesting is that

image that I'm looking for right

82

: 00:03:52

now today probably would take some

massaging and some prompt engineering.

83

: 00:03:55

you might need to say like

the sunset needs to be at this

84

: 00:03:57

aperture and we need to put the

imagery style or type or whatever.

85

: 00:04:01

but again, over time, I think that the

prompt engineering is going to get less

86

: 00:04:04

and less necessary as the AI mechanisms

get more and more sophisticated.

87

: 00:04:07

So, and again, I'm not saying

that you don't need it.

88

: 00:04:09

It's like Google search.

89

: 00:04:10

I'm really good at searching on Google.

90

: 00:04:13

I know that sounds like a weird

thing to say, but there's like,

91

: 00:04:15

nuances to search creativity.

92

: 00:04:17

And what words you use and how you use it

and how familiar you are with the product.

93

: 00:04:21

But I'd never make a living searching

on Google better than others.

94

: 00:04:24

Right?

95

: 00:04:24

Like, maybe I'd get a

little bit better at my job.

96

: 00:04:26

But it's not this role that

people are going to be hiring for.

97

: 00:04:29

I feel the same way

about prompt engineering.

98

: 00:04:30

I'd be really careful about

hanging your hat on that peg.

99

: 00:04:32

I'd be interested in

knowing what you think.

100

: 00:04:34

Do you think this is

going to be a real job?

101

: 00:04:36

and it's unfair for you to say, let me

just, scaffold the argument a little bit.

102

: 00:04:38

You can't say, well, yes,

for software engineers.

103

: 00:04:41

That's not what we're talking about.

104

: 00:04:42

That's not prompt engineering anymore.

105

: 00:04:43

That's software engineering

using AI, right?

106

: 00:04:46

the minute you say, well, yeah, it's

going to be great for graphic design,

107

: 00:04:49

like a graphic designer is going

to have to know prompt engineering.

108

: 00:04:50

No, a graphic designer is going

to have to know graphic design.

109

: 00:04:52

That's different.

110

: 00:04:53

Knowing, this theme, this

style, this color, this mix,

111

: 00:04:56

this opacity, like, that's not

prompting, that's graphic design.

112

: 00:05:00

So, prompt engineering specifically,

and all the nomenclature and the

113

: 00:05:04

semantics and the AI driven jargon,

I don't think that becomes, I think

114

: 00:05:07

that's, doomed to fail as a role.

115

: 00:05:09

The same way, I mean, if you look at,

some of the more basic and rudimentary

116

: 00:05:12

forms of coding now, coding back in

the day, man, you needed to know like

117

: 00:05:17

syntax and now it's, basic language

and now you just need to know logic.

118

: 00:05:20

So fight me in the comments.

119

: 00:05:22

I'll talk to you tomorrow.