The problem with prompt engineering, GPT-4o, and AI hysteria

I am delighted to announce that my friend and colleague Robert E.

Crossler, better known as Rob, is going to be joining as a co host on AI Goes

to College, starting with this episode.

Rob is a professor of information systems at the Carson College of Business at

Washington State University, where he's chair of the Department of Management

Information Systems and Entrepreneurship.

He's, uh, been published all over the place.

He's a distinguished member of the Association for Information Systems,

Cum Laude, which is kind of a big deal.

Rob has held a bunch of different leadership roles in our field,

and he's a very knowledgeable guy.

Not only as an expert teacher and world renowned researcher, and as a

department chair, he's also very familiar with all of the administrative tasks

that take place around a university.

So Rob's going to be a great addition and best friend.

thing for you is that you're not going to have to listen to me as much, I suppose.

All right.

Welcome, Rob.

And let's get to it.

Welcome to AI Goes to College, the podcast that helps higher education

professionals navigate the changes brought on by generative AI.

I'm your host, Rob.

Dr.

Craig Van Slyke.

The podcast is a companion to the AI Goes to College newsletter.

You can sign up for the newsletter at AIGoesToCollege.

com slash newsletter.

Hello everybody.

Welcome to another episode of AI Goes to College.

Today we've got three topics for you, starting off with the big set of

announcements from OpenAI that were countered by Google the next day.

We'll talk about Google at another time.

Today, we want to talk about OpenAI's set of announcements.

The big piece was the announcement of GPT 4 Omni, which is 4 0, but not

for number zero, it's for lowercase letter O, which is just going to be

confusing for all of us for a while.

Um, the idea behind, uh, GPT 4 0.

is that it's faster and it is much faster.

It will eventually have a larger context window, which means

it can remember more stuff.

It's supposed to require fewer resources to run, which isn't a big deal for any

individual user, but it's a big deal from an environmental standpoint and to,

for OpenAI and their cost of running it.

There's a lot you can do.

You can upload multiple files and query those files.

You can chat about photos, which is really kind of interesting.

In the next week or so, I'm going to put up a cute little example

that of course involves animals.

And you can do lots of other things.

The other big piece, um, that got a lot of people excited, or

one other big piece, was the, the enhancements to the voice capabilities.

And the idea is that it's going to be much more human like.

I'm not sure that that's really rolled out entirely.

Yeah.

But that, that will make it more like a conversation where you can interrupt

the chat bot and that sort of thing.

Rob, I don't know if you saw this.

I just saw it in the news that they took one of the favorite voices

down because it was too close to Scarlett Johansson to her voice.

So that was Sky.

So they just kind of put Sky on pause.

And so I'm sure the interwebs will be in an uproar about that.

Yeah, it's funny how, how that works, right?

The unintended consequences of these new things that we implement.

I'm not sure I would recognize her voice if I heard it, but, uh, okay,

there's also going to be a Mac OS app, which frustratingly I've

been able to download and install, but do not have access to yet.

And then they're making most of these, at least some of these

features, including GPT 4.

0 available to everybody, not just to paid users.

Another little feature that I think might be handy from time to time is you can

change models in mid conversation now, which it could come in handy sometimes.

I'm not so sure that'll be a huge step forward, but it's kind of nice.

So, Rob, there is a lot going on.

What do you think is the most significant for those of us in higher ed?

I'm most excited about the fact that it's free.

One of my concerns with the CHAT GPT 4 versus CHAT GPT 3.

5 has been creating an issue of the haves versus the have nots.

And, and if you can afford the better version, you had a leg up on others.

So in, in, as I look at this, the thing that I got most excited about

was the fact that if I want to use this in the classroom, I don't have to

struggle with asking everybody to pay 20 a month, uh, for a subscription fee.

Or, you know, how do we figure out a way to bake it in as a class fee?

So it's covered by financial aid stuff, or how do we find a donor

who's willing to support it?

By making this available, it really does take away that struggle of Am I creating

a burden for a student where college is already a burden for them to attend?

Yeah, I think this is a huge step in the right direction, so I agree with you 100%.

Although, to be clear, there are still going to be some differences

between the paid and the free versions.

The big thing with the big part of the announcement for the free

version is that access to 4 Omni.

Which is huge, which is huge.

So are you going to keep I actually turned it off to see

what life was like without it.

So I want to see what changed and what is the free version like?

So I'm going to use it for a while and, and just see, I look at a lot of these

things that come with generative AI technologies and these changes as an

opportunity to experiment and to learn.

And, and with a view towards.

what's possible.

And then I'll let that inform, you know, where I'm going to be next August

when students are back on campus and we're starting to engage again, because

I will not be surprised if there's more changes before we get to August.

Oh, yeah.

Yeah, absolutely.

This is kind of an aside.

I've always been of the opinion that if you only had 20 to spend and you're

not very technically sophisticated, a subscription to Poe, poe.

com is a way to go because it gives you access to all of the open AI models

that you can get at with ChatGPT, the Gemini models, LLAMA, and last time

I counted, it was almost 30 models.

That you had access to for 20 a month.

It's got some downsides to it, but it, it's really not a bad thing to check out.

If you are limited in what you can spend on these sorts of tools.

But one of the things that, that makes me temper my enthusiasm

just a little bit is that the free version is going to have like 20%.

Of the usage cap of the paid version, but I, I, I'm not so sure that Omni is

going to be necessary for most of what our students are going to want to do.

So we'll just have to, like you said, we're going to have to sit back and see.

I'm looking forward to playing with it to see what does it

take for me to hit that cap?

And, and what are the restrictions of, of hitting it?

And, and, and how does all that work?

Because I think understanding.

You know, in pragmatic real life sort of use, exactly how that

happens is going to be telling.

In my free time this summer and how I'm going to explore and play with this,

uh, one of the things I plan on doing.

And I should have mentioned Poe also gives access to the Claude

models, which are quite good.

Yeah, my experience has been that the hitting the usage cap

seems to be pretty inconsistent.

So I've used it intensively and not hit the cap and then, you know,

two days later, I'm using it and a lot lighter way and I hit the cap.

So I would imagine it has something to do with the resource load that they have.

But as Rob said, stay tuned, because.

You know, whatever's going on now, it'll be different in a few months.

Which brings us to the second topic.

Rob, how, how in the world do you keep up with all of the developments?

That's a great question.

And I can honestly say, I'm not certain I'm perfect at doing that, right?

There is so much that comes down that I try to, to read about

it on a daily basis, what is AI technology, uh, when the main players

Uh, feeds, I'll listen to them, but in a lot of ways, what, what you see

when these things are released is it's, it's so much, you're all the grand

great things that can do, and I'm not sure what is beyond the hype, right?

If you think about a salesperson, their job is to sell you on how

great the next new thing is.

So that's why I want to take the time to touch them, to play with

them, to see what's going on.

To know if the hype is real or if it's just trying to convince people to pick.

Whichever company was releasing as, as the winner of the new technology.

And before we keep going, it's, it's worthwhile for me to give a plug to

AI goes to college, the podcast and the newsletter, both of which are

available at AI goes to college.

com.

If you want to subscribe to the newsletter, there's a form or you can go.

There's a link to my, to the sub stack for AI goes to college at AI goes to college.

com slash newsletter.

Because one of the reasons that we do what we're doing today is

to try to help you all stay up with the most important of these.

So it wouldn't surprise me if between the two of us, we probably look

at 15 or more newsletters a week.

In addition to surfing and YouTube channels and all of that sort of thing.

So we can help a little bit with this, but, but you touched on a couple of

things that I think we ought to explore.

One is you, you used hype, but I want to go a step further and call it hysteria.

So there, there's a hype on kind of the vendor side and the, the AI proponent

side, all these great things it's going to do, most of which won't be true near

term, but there's also this hysteria.

That the world is going to end.

I don't mean that literally like existentially, but the world of

higher ed is going to be massively different and, you know, people are

going to lose their jobs and we're not going to need faculty anymore.

And they're going to have us teach.

And those of us who remain, we're going to teach 1000.

Students, uh, you know, class and blah, blah, blah.

So what do you think about all of that?

I would agree with the idea of hysteria.

I'd also agree with the idea that it gets weary, right?

It wears you out, fatigue, if you will.

And as I've just paid attention to it, I think of Chicken Little.

And you kind of mentioned, you know, the world's going to end,

but the sky is falling, right?

Every time you turn around, the sky is falling, which leads to

a level of fatigue, especially if you're in this space.

And you're trying to figure out, well, what does this mean to me in my classroom?

Depending on the voices you're listening to, you might actually come to that

conclusion that the sky is falling.

But I think the reality is different.

I think we have to pay attention and we have to adapt.

But much like any other new tool that comes out, There are risks

and there are benefits, right?

There are some great benefits to doing some things with this

generative AI technology, but there are risks of doing it wrong.

There's risks of throwing the, the everything that we know and that

we're good at out the window in order to adjust this new technology.

The big takeaway I have is I think about how I do my courses.

Is there's a new tool that's on the market and there's a new tool that all

of my students should be equipped and know how to use, which is no different.

When I taught a Microsoft Excel class back in the day, it was a tool that

students needed to know how to use in order to be successful at their job.

Very quickly, we've gotten a new tool that is going to be crucially

important for our students to use.

And it's the.

Emerge on the market in some incredible ways.

But if we step back and we think about it, it's, how do I empower, how do

I enable, how do I equip students to utilize a new tool to give themselves that

competitive advantage in the marketplace?

Then in a lot of ways, it's not a whole lot different than when we

introduced other new tools in the past.

Yeah.

I, I think one of the differences though is that this tool is

so, um, massively flexible.

I think that's a fundamental difference and you have, you and I have talked

about this before, not to, to go into the research weeds, but you

know, a tool is what you make of it.

We have a thing in the research world we call affordances.

So a rock is a weapon if a predator is coming after you.

But if you've got something in the pasture that you need to nail in, it

becomes a hammer, or it can help control erosion, or it can be a decoration.

And so generative AI is kind of like that.

It can do really serious things.

It can do silly things.

It can help you with an almost infinite variety of tasks.

And I think that's part of what leads to some of the hysteria and

some of the fatigue is just trying to get your head around all of that.

Forget about The almost daily developments of new tools and changes to the underlying

models and the latest way to cheat with AI and all of those kinds of things.

But just trying to figure out what it is in terms of what it does is enough

to make your kind of head explode.

Yeah, I absolutely agree.

And that's where I think being okay with giving yourself the time to

figure out how does these changes.

Effect what I want to do and to ultimately walk through it purposefully and not

reactively helps you to get to a place where you can be comfortable using it.

And I've talked to a lot of different faculty members who

are in different places with how they're utilizing generative AI.

Some people who are trying to push themselves to be at the cutting

edge of exactly what it can do and how to completely revamp what

they're doing in the classroom.

And other people who are looking for little ways to begin engaging,

to still be teaching those same concepts, but to do so in a purposeful

way that brings in some of these technologies and some of these tools.

And that's what I would encourage everyone to do is at least a little

bit, start to think about that.

How, how can I do something, but without feeling like you have to

completely change the way that you've been doing everything through.

The last 20 plus years, perhaps,

well, and to move it away from just being focused on faculty and teaching.

I think for those people in the audience that are in more administrative

jobs, it's the same kind of thing.

You don't have to change everything you're doing today.

I'm a big fan of the low hanging fruit approach, you know, learn

enough to where you can see a few things that might be really useful.

And then kind of put those into your routine and then as you learn how to

do those effectively with AI, you'll see some other opportunities and just

let it expand almost organically.

Don't try to do everything at once.

You'll just make yourself crazy.

Especially with the rate of change.

So, so to kind of say what you said in a little bit different way.

Take a deep breath, relax.

It's probably all going to be okay.

Notice the probably.

It's all going to be okay.

And you don't need to do all of this at once.

Because you will make yourself crazy.

Trying to stay up with all of it and trying to change too

many things at any given time.

The other thing I would add to that, Craig, is take the time to listen

to the, the stakeholders that are involved with what you're doing, whether

you're the faculty member and you're making changes and there's impact

from the students, the students have a different perspective than you do it.

And there's some value there, but for an administrative role, if you're doing

some things to try to make things more efficient from the perspective of.

People in student services that are advising students

that are answering questions.

Take the time to listen to those advisors and to those people and to incorporate

that into what your approach is.

Because I don't think there is a magical solution that we just know what it is.

There's going to be trial and error.

There's going to be things that work the way we expected it would.

And there will be a lot of things that didn't work.

But being able to listen and to hear that and to adjust, I think is the only way.

to truly be successful as you begin utilizing some of these new technologies.

Absolutely.

Absolutely.

All right.

So I want to kind of close this segment by pointing out that, and Rob, tell

me if, if this is not making sense, because it's, it's kind of forming

in my brain, one of the points about making this a chat based interface.

is to make it friendlier.

So you don't have to know Python.

You don't have to know how to code.

You don't have to learn an application programming interface, API.

You just go in and do what a lot of us do every day and you just chat.

And I really like that.

I think that's one of the reasons that ChatGPT took off so quickly

is because you didn't really have to know a lot to be able to use it.

Now, using it really well, using it ethically, you know, it gets a little bit

more complicated, but for its basic use.

It's not that hard.

That brings me to our third topic.

I think one of the most unfortunate aspects of this whole generative AI world

that's come out in the last year and a half or so is somehow people started

using this term prompt engineering.

It's really a thing.

And in some circumstances, it can be the right thing, but for the vast majority of

people that are going to use these tools, it's the wrong way to think about it.

We're starting to advocate for something called prompt design.

And Robert, what I want to do is Is kind of run through what I mean by

that and see what your reaction is.

So prompt engineering, engineering is a scary word.

I was an engineering major for three years before I decided I didn't really

like engineering and it was literally the, look to your left, look to your

right, you know, two of you won't be here at the end of the, term.

Low grades, high fail out rates, high burnout rates.

Engineering is hard.

So imagine you're talking to somebody and say, Oh, you know, you need to

learn prompt engineering to use AI.

It's going to be like, well, screw that.

I'm not an engineer and I don't want to be one.

No disrespect to engineers, but engineers are highly trained people.

If you're a professional engineer, you've got strict licensing exam requirements

and ongoing education requirements.

It's a really technical discipline.

It's also really precise.

You plug formulas or you plug numbers into a formula and you get a precise answer.

When I was in the engineering world, you know, we take things down to two or

three decimal places fairly routinely.

So all of that is not indicative of what we do with generative AI.

You know, we go in.

You don't need to know a lot.

You start chatting.

You try to get kind of close.

You refine it as you go along.

It's really iterative and kind of emergent.

It's not precise.

You know, you put the same prompt in three times in a row, you're going to get

three at least slightly different answers.

So to me, all of that is much better captured with the idea of design.

Design is a little bit messy.

Like I said, it's iterative.

You go down these paths that don't really work very well, but that

helps you learn some things to where your ultimate design is really good.

It's non linear, it's imprecise, and so, what do you think, prompt engineering?

Utilize, um, this prompting with tentative AI is more art than it is science, and

that the exploration, the curiosity, the, uh, as you change the wording of your

prompt is going to give you different outputs, and it's through that art of

exploration that you're going to come up with the prompts that Result that

in your solution and that each time you utilize the tools, you're going to go

through a different process that will ultimately help you to take the output

you're getting and make it useful.

So you could have saved everybody two and a half, three minutes.

If you would have just interrupted me and said that as soon as I got started,

cause you said it in a much better way.

You're inspiring words, Craig, to get me there.

There you go.

Yeah, yeah, yeah, yeah.

I guess the big message is don't let prompt engineering scare you.

And I really am going to start trying to use prompt design in my classes.

And when I'm giving talks and that kind of thing, it's going to be

forget about prompt engineering.

Although there is, there is space for that.

So if you're doing something that needs to be really.

Repeatable, very precise.

You're going to deploy it broadly.

Let's say you're doing a customer service chat bot.

You need to engineer that, but that's for the experts.

You know, that, that's not for those of us who are using this more or less

casually, even if it's seriously, we're not trying to dig into it to that level.

So, uh, that, that's why I'm going to really start.

And what I would encourage to Craig is as people are designing these prompts

and getting different outputs to begin.

Creating a prompt library where they capture kind of the essence of what

prompts they use, the sort of output they got, they received, and then how

happy they were with those outputs.

So that way, as we're doing things in the future, they don't

have to reinvent the wheel.

And, and it'll help them to go back and be like, in this particular

environment, where I was doing this work, or I was doing this thing for fun.

I want to do something similar.

They can refresh themselves, remind themselves of what sort of prompts they

use, they got them there, which should help them to be more efficient to get kind

of back into that same zone of the sort of output they're receiving when they do

something, um, different, but related.

Yeah, or, or even if you're going to do that, or maybe even more so, if you're

going to do the same thing again, like I'll give you a quick example, or I'm

giving my final exam in the morning, well, I want to make sure that there are

no duplicate questions on the final exam.

And then I want to make sure it matches my study guide.

I have a deal with my students.

If it's not on the study guide, it's not on the test.

They're different form for the questions.

And there's a lot more on the study guide than I could put on

a test, but that's a perfect use for something like generative AI.

Literally put both documents in, do this with a test, match it up against a

study guide, flag any unclear questions.

Well, I'm going to do that every test now.

So instead of what, what I hear you saying, instead of reinventing that wheel

every time, once I get the process, and I think this is an important part of it.

Once you get the process down, put that in a notepad file and an Apple notes

file and a Google doc, whatever you use.

Okay.

And then just copy and paste.

So I think that's great advice, but, but also just to make sure nobody missed it.

Rob's also saying when you do something similar, so, you know, a lot of our

tasks are variations of the same theme, but they're not exactly the same thing.

So yeah, you can, you can adapt.

As you go along, of course, and they'll change the models.

Well, that's when they change the models that gives you the opportunity

to say, do I like this new thing that came out better or worse?

Because you can go back and kind of be the same exact thing and say, you know what?

My reaction to what I'm getting now is this was a step backwards, or this

was a step in the right direction.

I really like it.

I'm looking forward to using these new things.

Great, great.

All right.

Well, I think we're close to a wrap.

Anything you want to add?

All I would say is the theme from this episode that I think will

probably be a theme from a lot of our episodes is keep exploring, keep

playing, keep experimenting and having fun with these new technologies.

No one's judging you.

And unless you share your results, people aren't going to know the

creativity you used to get to the place where you found something useful.

So feel free to play with these things and see how they're helpful to you.

And that's really the best way to learn a new technology is to play.

All right, great.

Well, as we wrap up.

Anything you want us to dig into, let us know.

You can email me at craig at AI goes to college.

com.

You can use the contact form at AI goes to college.

com.

And we are going to be rolling out and have rolled out some videos, some what

I'm calling lo fi kind of one take videos on things we're doing with AI at

our YouTube channel, which you got it.

It's called AI goes to college.

There'll be links for all of that in the show notes.

And thank you very much.

We are out.

Talk to you next time.

Thanks for listening to AI Goes to College.

If you found this episode useful, you'll love the AI Goes to College newsletter.

Each edition brings you useful tips, news, and insights that you can use

to help you figure out what in the world is going on with generative

AI and how it's affecting higher ed.

Just go to AIGoesToCollege.

com to sign up.

I won't try to sell you anything, and I won't spam you or share your

information with anybody else.

As an incentive for subscribing, I'll send you the Getting

Started with Generative AI Guide.

Even if you're an expert with AI, you'll find the guide useful for helping

your less knowledgeable colleagues.