The future of generative AI, a great Chrome extension, and using AI to examine an exam

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, Dr. Craig Van Slyke.

The podcast as a companion to the AI goes to college

newsletter, you can sign up for the newsletter at AI goes to

college.com/newsletter.

In this episode, I talk about my view of the future of generative

AI. Talk about how I use Claude to check an exam, which turned

out pretty well actually. And I also talk about an interesting

new AI extension for Chrome, one that I think has a lot of

promise. So let's get started.

So what's the future of generative AI? It's not been

quite two years. So about a year and a half, really, since open

AI released chat GPT. And we've come a long way since then. But

I think it's time to start thinking about what the future

of generative AI is going to look like. I know a lot of

people are thinking about this, a lot of people are writing

about this. So I'm gonna give you my view, the first thing

that I think will happen is that AI will fade into the

background, it will go from being this thing that we're

talking a lot about, to something that just is it's just

there, the chat bots are all well and good. They're very

useful. They're very flexible, but they require too much

expertise to use, especially for complex topics, or complex

tasks.

Sasha agrees with me, I don't know if you heard her there in

the background. So I really think that what's going to

happen is we're going to see these tools, I call them task

based AI, that just do something they help you screen your

emails, they help you refine your writing, they keep you on

top of your task list, you know, whatever it might be.

The GPT store that opened up not too long ago is a step in that

direction. Custom GPT is can be really useful, especially when

they're well designed. If you want to just have some fun,

check out one called Cosmic dream, it creates some pretty

trippy images, it's a lot of fun to play around with, I created a

simple GPT to help with this newsletter, there really isn't

very complicated. It just gives Chet GPT some context so that I

don't have to enter that context. Every time I want to do

something related to the newsletter. Ai enabled tools

take this much, much further. For example, I use one called

caste magic to create all sorts of material for my podcasts,

Otter AI, which I think I've talked about before, is really

awesome at transcribing audio summarizing, zooms. Summary

feature is another example, Grammarly. The list goes on and

on and on. And it's growing every day. So here's a good

example. I think one that doesn't exist to my knowledge.

What would it be like for those of you who are faculty, if there

was a really good AI test generator, you record your

lectures, give it access to your notes, or textbook chapters, set

some parameters, and click on Go. So, before too long, you'll

have an exam with really solid questions ready for printing or

uploading to your Learning Management System, or whatever

you do. I think those kinds of tools could free higher ed

professionals, faculty and staff from a lot of drudgery. It's

something I called Moving humans up the stack, which just means

that AI can do the boring semi mindless work that so many of us

hate freeing us up to do more engaging interesting parts of

our job. Maybe you're different, but I hate developing and

grading exams. So alright, let's move on. I think the AI enabled

tools have almost endless possibilities. So that's step

number one in the future of AI, or the first thing I want to

mention. My next prediction is that we'll see a significant

increase in the use of AI agents, an agent and an AI tool.

They're really kind of hard to separate. An agent does

something on your behalf. Siri and Alexa are kind of agents.

But I'm talking about something different here. I'm talking

about things that perform specific tasks as sets of

workflows. And I realized I'm being a little bit confusing

here because it is very, very messy. So for example, an AI

agent can scan your email inbox and classify messages into

different topics. Or maybe an AI agent might scan student

admissions applications, and extract key data and

automatically enter that into an admission system. I'd love a

highly capable AI agent that could just grade homework or so

alignments.

So again, I know that the line between tools and agents is a

little bit fuzzy. So let me give you some key characteristics of

AI agents, AI agents mostly operate autonomously with

limited human intervention. I think that's a big difference

between what I was talking about with tools and agents. AI agents

are task oriented and typically limited in scope. So they have

one narrow thing that they do. AI agents are embedded in

workflows. So it's generally not like a chat bot, where you do a

bunch of things with it, it's just kind of there. And when

it's its turn to do something it doesn't. AI agents are often

embedded in application. So you may see it as part of a textbook

publishers app or as part of an admission systems app. That sort

of thing. Using AI agents doesn't require any real

expertise in AI. No prompt engineering. In fact, you may

not even know that you're using AI agents. And here's a big one

that I'll talk more about later, AI agents can be combined. So

these agents will be operating behind the scenes. As I've said

a couple of times, you may not even be aware they're running.

Although you may trigger the agents use or use its output

without really realizing it. When these are all refined, I

think these agents are going to be massively valuable. But

building and refining them is going to take a lot of effort

and considerable resources. That's one reason I see most of

these coming from large vendors. All right, so let's talk about

my third prediction agent to agent networks. This is where

the real power of AI agents will come. From networks of agents

working together just like organizations, networks of

people can accomplish really great things working together

more than they can separately, networks of AI agents will be

capable of some amazing things. Imagine if you had an AI agent

network that continuously scans your inbox sending messages off

to other agents for further processing, your email

processing agent recognizes an email from a student who needs

advising. So it hands off the message to your advising agent

who coordinates with the university's class scheduling

agent to find out what relevant classes are being offered in

future terms. Then the scheduling agent passes this

information back to the advising agent, who figures out what the

student should do and sends this information to your email

composition agent who writes a response and sends it to the

student, maybe after you review it. Except for the review. All

of this happens without you doing a thing. That can be

pretty sweet. Of course, there are risks, you're putting a lot

of trust in the agents. Man, there are places where this

could go pretty horribly wrong. But a smoothly working agent

network could be a wonderful thing.

Okay, there's more to this. But I think you've kind of got the

point. So my vision of the future of generative AI as a

shift from Ai chatbots to AI embedded tools, the rise of AI

agents and networks of AI agents taking on complex workflows. Of

course, I might be wrong about all of this AI is to say the

least dynamic area. Nobody really knows what the AI

landscape will be even a couple of years from now, maybe not

even a couple of months from now. It's an exciting time, it's

going to be an interesting future. And it has a lot of

potential. Maybe to free many of us higher ed professionals from

some dreaded, drudgery. least that's what I'm hoping to see.

But time will tell.

Normally, it takes me quite a bit of time to scan for

duplicate questions in questions that are hard to understand. I

was reasonably confident of the quality of the questions on this

exam, but I needed to double check.

I might pass this off to my GA but I really didn't want to do

it in this case, because the turnaround time was so quick and

that wasn't fair to him. So I decided to use Claude three Opus

to do the checking. First thing I did was create a PDF version

of the exam when uploaded to Claude. You know, there's some

privacy concerns with this, but I wasn't really worried about

Claude having my exam questions and there was no personal data

involved. So I want to hit

that I put in a pretty simple prompt. Please analyze the

attached exam for two areas, unclear questions, and duplicate

questions.

And the newsletter which is available at AI goes to

college.com I have a screenshot of clods response course I don't

have the actual questions in there, I've kind of blurred

that. It gave me a list of unclear questions, but Claude

didn't really know some of the context that the students would

have. So the questions really weren't, weren't bad. So I

didn't make any modifications. I could have given Claude more

context, but it really wasn't necessary in this case. So for

example, Claude thought a question about the differences

between requirements and goals and decision making was unclear.

But I'd spent a lot of time on this in class, so I'm sure the

students would understand the questions, at least the ones who

were paying attention. It didn't identify any duplicate

questions, which was good. So I decided since this had gone

pretty well, I was going to take it another step. This time I

asked Claude to check the exam against the study guide. I give

extensive study guides and promise my students that have a

concept isn't included in the study guide, it won't be on the

exam. Now the questions aren't on the study guide. But the

topics are. For example, the study guide might say compare

and contrast requirements and goals in the context of decision

making. But the exam includes multiple choice question on the

differences between requirements and goals. Now, checking the

exam against the study guide isn't hard, but it's kind of

tedious. So I asked Claude to do it for me. I put in another

another pretty simple prompt. Thanks, I'll wait. Thank you i

ai. Please compare the exam to the attached study guide. Are

there any questions on the exam that relate to topics not on the

study guide? Note that the study guide does not contain specific

questions for the exam. Rather, it covers topics that students

should know. And Claude went through did its thing and raised

a few potential mismatches. But after I reviewed these consider

these concerning questions, I decided that the questions were

fine clods concerns were reasonable. But really, human

judgment was required to make the final decision about whether

or not the questions recovered in the study guide. Since I was

already pretty confident about the exam, I felt entirely

comfortable using Claude to do the checking. But I if I were

you, I would run a couple of experiments before fully

trusting AI with this sort of a task. The big message here is

that Claude made a couple of tedious tasks much more

palatable. It also saved me about an hour, maybe an hour and

a half of time, which was nice at the end of a long day. So if

you have some tasks you find mundane and tedious, see if you

can find a way to get AI to help free you from that drudgery.

That seems to be the word of the day drudgery. But remember, when

all of a sudden done, you're still responsible for your work.

By the way, if you want some help in figuring out how AI can

reduce some pain points, feel free to contact me, Craig at AEI

goes to college.com. By the way, I love to talk about this stuff.

So I'm happy to come to your campus to put on a workshop give

a little talk, we can do something virtually doing this

on a number of campuses in the next few months. So if you're

interested again, send me an email Craig at AI goes to

college.com. Alright, we're gonna wrap up with an

interesting new tool called Harpa. That's H ARP a.ai. It's a

Chrome extension that integrates AI into your web browser. Harper

can use a bunch of different AI models, but the normal chat

based interface seems to use GPT 3.5. But you can also connect it

to Claude and Gemini. For me, harpist main use is to summarize

webpages, but it has some other functions as well, maybe too

many, it's a little overwhelming. Here are a few

summarizing web pages, generating content, including

email messages based on a webpage, monitoring websites for

changes, extracting YouTube transcripts, that could be handy

extracting data from websites. And there's a whole lot more.

All of this goes from a Chrome browser window, which is pretty

handy. So it's kind of tough to do here on a podcast. But if you

go to the newsletter, which once again is available at AI goes to

college.com/newsletter. I run through an example and it works

pretty well. Right now there's a lifetime deal on Harpa. I think

it's $240, which isn't bad. harpaz pricing structure is kind

of odd and a little bit confusing. Most people from what

I can tell would probably be fine with what they call their

ex membership, or subscription that's their lifetime deal. Or

their s one which is a monthly membership. You can go to

harpa.ai and find out the pricing for yourself and decide

what's what's good for you. All right, well that's it for this

time. I hope you got something out of this episode. If there's

anything you'd like for me to talk about, feel free to send me

some suggestions. Craig at AI goes to college.com And

remember, you can subscribe to the newsletter and I goes to

college.com/newsletter. All right, I'll talk to you next

time. Thank you. 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 AI goes to college.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