Welcome to another episode of AI Goes to College, the podcast that helps higher ed professionals figure out just what in the world is going on with generative AI.

As always, I am joined by my friend, colleague and co host, Dr. Robert E. Crossler from Washington State University.

Rob, how are things?

It's great, Craig.

It's summertime.

It feels like school will be here before we know it, but been able to find some opportunities to enjoy some downtime, watch some great World cup matches and have long days of warm weather.

Yeah, US is out, I guess.

Yeah, that was unfortunate.

But that doesn't change the fact that there's a lot of really good football being played for those people not in the United States.

Wait, that's not until like August.

Just kidding.

Okay, so enough of that, let's get right to it.

Rob, in the pre show you mentioned that you were using Gro.

Yeah, so I like to play with the various different tools, Craig, see what they're capable of.

And I had a need in this case to write a family friendly lease as my one of my children is moving back home.

They've got a good job that sets them up to save some money to hopefully buy their first house sooner rather than later.

But we wanted to have some agreements in place of what that would look like.

And with a very simple two sentence prompt, it created a Word document for me that was almost spot on with what I'd want.

And it looked all legalese to me.

And I was able to take that Word document, downloaded it to my computer, made a few edits and shared it with my wife who agreed with the terms and shared it with my son who thought they were reasonable.

And.

And here we.

So kind of cool.

Inquiring listeners want to know, did the lease include somewhere the language?

As long as you're living under my roof, you'll follow my rules.

Did say something about following house rules.

So not.

Not quite those words, but.

But yes.

Nice, nice, nice.

So GROK did as good a job as you would have gotten out of Copilot or something else.

Yeah, I felt like it did Copilot.

A lot of my uses there I do professionally for work and so I wanted to make sure I was in a separate space doing something in my personal needs and thought it would be a great way to compare some of those capabilities.

I think we're getting to the point where the top models are going to converge on a level of competence that's going to be okay for most tasks.

Yeah, and really when you push things out to the edge is where you're going to start to see the differences.

Yeah, that's what I'm really seeing.

I'm surprised more and more by how good every model is for 80% of what I'm trying to do and makes me question if it's capable of even doing those last 20% or if that's where human value and human expertise in the process comes in.

But yeah, tools are getting to be pretty good, pretty consistent across the board.

Yeah, they are.

They are.

And I guess that's not surprising.

The top model of yesterday is now the mediocre model of today.

That's where I think if we tie this to what we're teaching our students to do is the model, the tool doesn't matter so much as the skill sets around what's being used.

And we'll dive into that a little bit later today, I think.

But if you're hung up on which tool to pick for classroom exercises, I don't think it matters.

I think go with the one you're comfortable with and lean into it.

I largely agree, although I will tease an AI Goes to College newsletter article that's coming out tomorrow that talks about where that may not be true around the whole Fable 5 thing.

Fable 5, for those of you who may not know, was the big scary mythos model from Anthropic that the federal government put import controls on.

That was the public facing model that we're not going to have access to through regular subscription fees as soon.

I think that's going to create some accessibility gaps.

But go to aigostocollege.com subscribe to the Newsletter and you too can read that article.

And I'm sure we'll talk about this in a later episode, but we've got enough to talk about today.

So, Rob, I did something really cool over the last week.

You're not going to ask me what was it?

Craig.

Oh, I'm on the edge of my seat.

Craig.

I am so waiting for this.

That's right.

Some colleagues and I decided at the last minute to submit a paper to a pretty good conference.

It's a project that we've been working on for a while.

We've collected a little bit of interview data and we were pretty happy with the progress.

But we thought, oh, if we go to this particular conference, we're going to get really good feedback.

This is typically attended by some of the top people in the field.

When I've gone, the level of feedback has been extraordinary.

So we really wanted to get this idea in front of some smart people.

But there wasn't a lot of time and both of my co authors were busy with other things.

It's summertime, people are on vacation.

So I said, well, I'm just going to write this thing.

Rob, I don't know if you've ever tried to write an article completely by yourself, but it's not an easy thing to do.

Have you ever tried it?

Yeah, I tried it lots of times and I honestly get stuck.

Because having a partner to process through things with is so invaluable.

No, that's right.

Trying to be out there on your own.

Especially if it's an article that's trying to break some new ground, which is what this particular article is trying to do.

So what I did is I had, I used Codex and I could have done the same thing with Claude.

I'm not sure about Gemini, but I could have done the same thing with Claude Code or codec cowork.

But I gave it all the context.

We had a bunch of articles that we were going to reference, we had data interview transcripts, we had the interview protocol, some early little papers that we'd written.

So we had a bunch of stuff and I gave it all of that context.

And then over the course of, I don't know, maybe three days, two or three hours a day, we wrote a pretty decent 25 page, double spaced article that I think does a good job.

Now the thing that we're doing did not require any really rigorous data analysis yet, but the way I did it is I would write things and bounce it off of Codex.

So kind of did the back and forth that way.

Instead of asking Codex to write something and then I would edit it, I'd write something and ask for feedback.

Just as if we were working on a paper together and I wrote a section and sent it to you and said, hey, Rob, what do you think?

So it was really another set of eyes that were pretty well informed, not always right, but made the paper stronger than if I would have just written it on my own.

So that's AI as the learned colleague, but I also had AI as the gf.

You know, go, go get these papers, put them in APA format.

Right.

Or look, I need to find another paper that says this, go find me some.

It would find them, I'd look through them and say, okay, this is the paper that's got what we need.

Exactly the kinds of things we would do with a ga. You send them out, you don't take what they do as being gospel, but they save you a lot of time.

Yeah, that's not Surprising to me.

Craig, I've used copilot with the M365 version as the exact same thing.

It has two settings where you can have it basically respond to your prompt.

So I can highlight a paragraph and ask it, is this clear?

Am I clearly communicating it?

And it will come back and say, this doesn't quite make sense, and give you feedback like you'd have a colleague do.

And you can also put it in editing mode and give it permission to edit and update things where I can spew out ideas and not care about my writing and highlight it and say, take all of these thoughts and make this paragraph make sense, and it will go and do some of that precise work for you.

So the fact that you're doing various similarities to that using Codex, it's not surprising that you're finding good success there.

It worked really well.

And I also used it as an editor and a copy editor.

So where are the weaknesses in this argument?

I gave it one section.

I said, I think the paper's in pretty good shape overall, but this section feels thin to me.

What's missing?

Right.

I couldn't quite put my finger on it.

And it kind of said, well, you really haven't touched on this part.

And they're, oh, okay, that's what I needed.

So I wrote that part and it worked out well.

Now we'll see if it gets accepted.

I hope it will, but you never know about these things.

But I want to really emphasize that AI did not write anything.

Now it will go and write stuff on its own without you asking it to.

But none of that made its way into the paper.

It was really just having that colleague there.

So, Craig, one of the things I struggle with when it comes to research work and AI use is it seems like different journals, different conferences, different places have different rules as far as what's allowed and what's not allowed and how do you navigate that?

And one of my fears is, let's say I'm targeting a conference where they're fairly liberal on what they allow AI used for.

But one of the journals I may potentially submit it to is much more conservative or restrictive in what they would allow.

Have you put any thought to how you navigate how?

It seems like there's different rules in different places and different outlets.

No, not really.

I don't know that there's a good answer to that, except to disclose when it's required.

This conference didn't ask for any kind of a disclosure, and I. I don't know.

I think we're going to See a day before too long where this becomes less of an issue because it's just going to become part of how we do research and it's.

We're going to need to come to some sort of an agreement.

It really parallels what we're seeing at universities where one class has one set rules, another class has another set of rules.

But I might go so far as to not submit to that journal if it.

If it's overly restrictive.

Yeah, it's interesting.

I recently submitted a paper somewhere with an author who is an editor in chief at a journal, and one of the steps that she took in the editing process was their publisher basically said words like consequentially.

And anything that's like these superlatives that you put at beginning of sentences were telltale signs of AI use and needed to be scrubbed out of the paper in order for it to satisfy.

At least what their journals were telling them they had to do has created some very interesting conversations amongst friends that say, why is good writing frowned upon as AI misuse, if you will?

So I think, yes, I look forward to a day when we can all agree that much like good math is good math, whether you do it by hand or with spss, or good writing is good writing, whether you can do that through your own fingertips and your own thoughts, or if you use copy editors to help you get there.

Yeah, I mean, I think some of that stuff is complete nonsense, but the line's going to be blurred.

This idea of AI did this and I did that, it's just going to completely go away at some point.

We're just not going to know who did what.

So anyway, we'll see how it goes.

I'm confident that I did this in an ethical manner and certainly did it in a more efficient and probably more effective way than I could have on my own.

It was definitely in that AI.

I don't even want to call it augmented.

I'm not sure that's even the right word to use.

I got feed back from AI and sent it off to do some things that, you know, then I read the articles and found what I needed.

And so I don't see anything wrong with it personally, but I'm sure there are those who would disagree.

Well, I think this goes to the idea of the changing workflows and what that's going to look like.

And the pathway there is not going to be linear and smooth.

It's going to take some time to figure out what the human value is in the work that's being done and how that goes.

Yeah, so we'll see.

But anyway, we'll find out, I suppose.

One thing I wanted to share with you, Craig, before we jump into our topics is I've talked in previous episodes about creating a chatbot and copilot to help review syllabi to make sure they're up to snuff for what is expected come fall term.

And one of the cool features that we added just this week is the ability for it to create the HTML code that you can take and paste into Canvas.

So you can use the syllabus feature inside of Canvas and it works really great.

Take what's been a Word document or a PDF document and it will actually put it into the formatting nature that can just be pasted straight into Canvas.

It's pretty slick.

Nice, nice.

Yeah, that.

That single source of truth is nice.

I hate updating a syllabus and having to reload it to the learning management system and make sure you have a date on there so they know it's a new syllabus and all of that just goes away.

So yeah, yeah.

And so again, it's the idea that the workflow changes and the person's time is spent on the tasks that matter, not figuring out how to get formats to work or, you know, is this little nuanced thing been updated.

It can really evaluate that and do that well for us.

Nice, nice.

So with that, Craig, you shared with me a Higher Education Policy Institute document that has some really interesting results from some data that's being collected, some research that's being done.

Can you share that with us?

Yeah, and there will be a link to the report which is freely available.

I'll put that in the show notes.

It was out of the uk, there was a thousand plus students that they surveyed and there were some surprising results from the survey.

So I just want to run through a few of these and see what you think.

So generating text with tools like ChatGPT that use case actually declined from 64% of the students in 2025 to 56% of the students in 2026.

I don't know if that's really enough to be meaningful.

And I'm not convinced that the usage dropped versus I'm just not going to say I do that, but I thought that was interesting.

Well, I find that really interesting in the sense that any AI use increased from 66% to 92%, but the amount of people that were generating text with it decreased.

That begs the question a little bit of what is it being used for?

And we're getting to a place where almost everybody is using it.

Is it companion mode?

Is it just the back and forth of having something going on?

What new and interesting ways are students using this for?

Because I can really see this translating into how they would implement this in the workforce.

Yeah, it's an interesting set of results here.

My first thought was, well, they used to call everything generating text and now they're being a little more nuanced.

But that's really not what the data shows because a lot of the usages, matter of fact, most of them dropped off, except for they added some new questions to your point of just a second ago.

So some new questions for 2026, they ask if you used AI for friendship company, as in companionship advice or tackling loneliness for counseling or therapy, where this service is provided solely by AI.

And then another question where that was provided partially by AI.

And so maybe in earlier surveys, people that just wanted companionship viewed it as generating text, which it isn't.

Now they have a more granular view of it.

But there's no way to know that from the data that we have available to us.

But I want to come back to that idea of companionship friendship, because I think that's interesting.

So buried in this whole report was a case study from three different, I'm sorry, four different universities with medical students working with virtual patient cases and an AI tuned for medical care.

And the least skilled students were the ones who were the most confident in their usage of the AI.

Do you think that's because they had to rely on it more in order to stay competitive with their more highly skilled classmates?

What's your hypothesis there?

I don't know.

It could be that.

I think it could also be just thinking that the computer is right.

Algorithmic bias, it's coming out of the AI, so it must be right when.

Sure.

So maybe a little bit less critical thinking of the least skilled students that they're not going to challenge it as much and they feel that, well, it said it.

Who am I to question that's something.

That's worth keeping an eye on because I think we probably see anecdotally the same things in our own classrooms where the sharper students are the ones who are going to be both more confident in challenging AI and more skeptical that AI is just going to spit out an answer.

Yeah.

One of the things I think that's really interesting as we think about who's using it and how many people are using it is the barriers to using it.

Those risk based deterrents are Starting to fall that whether or not it's thought of as cheating and so forth, things that would stop students is beginning to come down, which suggests students are going to feel more confident that they can and should be using these tools when it comes to doing various different academic assignments.

Yeah, and the data from this survey bears that out.

So just.

I'm going to throw a couple of statistics out there.

So the exact question is, which of the below, if any, are reasons which make you less likely to use AI tools for your studies.

Being accused of cheating dropped to 42% from 53%.

Getting false results and hallucinations dropped from 51% to 35%.

Another one that was interesting.

My institution discourages or bans the use of AI dropped from 31% to 21%.

And again, I'm not quite sure what to make of this.

Maybe the rules are getting a little bit clearer for students so they're less worried about being accused of cheating.

And even though AI can still hallucinate, the number of hallucinations I'm getting is getting close to zero.

Part of that is knowing how to avoid them.

But part of that is a model of getting better at self verification.

I think another thing that we're seeing in that data, Craig, is experience.

That students, and everyone in general, right at first, when they don't have a lot of experience on it, they may have these fears, these things in the way that they've read about the what should you be watching out for sort of things.

And as people have tried it and used it, they've realized that they weren't as present as they thought they would be, that they have been able to rely on what they found.

The bias hasn't been there.

They've been able to create better reports or various different things for the work that they're doing.

I think with experience comes a confidence in using these things that tears away those fears.

Yeah, that makes sense.

And I would add to that that they may be experienced enough to be a little bit more critical now to actually check to make sure that facts are accurate or that articles are actually articles that exist.

So, but.

But I don't know.

It's probably encouraging because it seems to line up with what we're seeing more globally with better refined policies, more consistency, Although so many people are still using AI detectors, including journals.

So stop.

If you're doing that, stop.

They don't work.

They're bad.

Did I say they were bad?

Yeah, you're preaching to the choir here, Craig.

That's good.

That's good.

So I'm just proud of myself for not calling them garbage as I usually do.

This was a really long report, so we don't need to go through all of it.

But I do want to point out the loneliness effect.

So they asked, how does generative AI impact your level of loneliness?

And I found this to be fascinating.

So 59% said AI does not have an impact on my level of loneliness.

Okay, good.

The other 41% is pretty evenly split between it makes me feel less lonely and it makes me feel more lonely, really, 20 to 21%.

So about as even as you can get.

I don't know.

I'm really torn.

And we've talked about this before.

I can see AI for companionship being a very good thing if you feel like you don't have other options.

But it is not human companionship.

So now I know.

Rob, this is really a pretty big concern of yours given your position and your personal interests, I think in general around this topic.

So what did that say to you?

What are you thinking?

I think about this a lot.

One is I'm glad if this is filling a need for people that is able to meet them where they're at at the time that they're at.

So I do see some real pluses in this.

I'm concerned with long term impacts.

Right.

What does that do?

Does that drive you further into isolation as you're using a technology to speak into that loneliness?

Do we know what those long term impacts are?

And I'm hoping that people are doing studies on this so we can figure it out sooner rather than later.

One of the things that we've seen with social media, there was a whole lot of positive stuff that happened with social media, but there was also a lot of negative stuff for teens, for youth in various places.

And we didn't really see legislation or lawsuits that were successful and whatnot until a decade or more after social media had come out and it already had made a lasting impact on a generation of people.

So I am hopeful that we can learn from that and we can very quickly try to capture some long term thinking on what will this impact have and figure out ways to harness it so we really can use this technology to help people while at the same time putting guardrails in place that can protect us from what could potentially be very dire negative consequences.

Yeah, couldn't have said it better.

I want to go back to the use cases.

So if you'll remember, there were three questions that were kind of related to this.

Using AI for friendship, company advice, or tackling loneliness.

15% Of the respondents indicated that was something they used AI for, and then a combined 12% use it for therapy, either all on its own or in conjunction with some other form of therapy.

But I think we've got the same kind of tension there as we do with loneliness.

A lot of people don't have access to therapy.

And anybody who's ever tried to get a therapist or has had a loved one who's tried to get a therapist, it is not easy to get a good therapist.

Yeah, yeah.

There's definitely a lot of upside to this.

Just truly is an eyes wide open approach to it as we're leaning into more accessibility, more availability.

Going back to the idea of biased and so forth is you could really easily be taken down a rabbit trail of biased information if you're not using the right tools in the right ways.

And how do we ensure those sorts of things aren't happening?

Well, the tendency for AI to be sycophantic.

Sycophantic, I can never say that word.

But to kind of suck up, to put it more bluntly, is a really bad thing for therapy.

One of the things a good therapist will do is tell you the truth.

Truth, even when it's not what you want to hear.

And it's like, you know, it's not your fault, you know, that everybody's mean to you.

Maybe it is your fault.

Maybe you need to change some things about yourself or whatever it might be.

And so I don't know.

We're in interesting times, Rob.

We're in interesting times.

So I want to point out one last thing before we move on to the next topic.

Even though almost everybody in this survey said that they used AI, only about half of them had AI experience before they got to college.

That's not surprising because our colleagues in primary and secondary education have more than enough to do and don't have the resources and the some of the access to expertise that we do to help them figure out how to help their students understand the use of AI.

I think that's going to be a problem that we're going to feel in the next few years if it doesn't change.

And then there's a gender gap.

Men were about 12 points more likely to have school experience, although that seems to be narrowing a little bit versus 20, 25.

So we're still kind of in early days.

Anybody in higher ed should download this report, skim through it, put it in AI, put it in notebooklm get a sense of what's going on.

The report is pretty well done.

The data tables are all available.

The raw data are not, but all the data tables are available as well, if you want to do some of your own thing with it, but.

Hi, Rob.

Any last thoughts on the Hepi 2026 gen AI survey?

A takeaway I had as I looked through this is it didn't seem a whole lot different than data.

I remember seeing when the Internet came out as it got to the ubiquitous nature, it has gotten there way faster than the Internet did.

But the gender gaps that we're seeing, the use gaps between those who are confident and not confident, I see a lot of repeated trends.

So as you think about what this means, to even look back and see what, what thinking 20, 25 years ago, as the Internet was becoming pervasive and available to lots of people at various different rates and different use cases, we probably can learn from the past, and.

I'm with you on that, but I want to give a little caution.

This is different because the Internet was really focused on information.

We had greater access to information.

We have fewer gatekeepers, which was a good thing and a bad thing.

But here I think we move beyond that into thinking where it's really influencing our thinking in a different way.

Not just the input into our thinking, but the way in which we think.

But I agree with you.

We should pay attention to what happened back then and see what we can learn from it and maybe how we can handle a little bit better.

All right.

Oh, next topic, Rob.

So ChatGPT has these things called scheduled tasks now that came out of something they called Chat GPT Pulse, which only pro members had access to.

I think, at least for most of the time Pulse was around.

And basically what Pulse would do is it would go out, look at your conversations, and then give you a news feed based on that topic.

And so if you searched for.

You mentioned World cup, if you're talking about World cup all the time, it might give you a daily World cup report.

People didn't use it much.

It didn't get a lot of press.

They were kind of useful and I forgot about them, frankly.

And then they migrated these over to something called scheduled tasks.

And you should see those.

If you go on the web version, I don't know that they're on the app yet, but if you go to the Web version of ChatGPT, you'll see them scheduled.

And basically you can ask it for a daily or weekly or monthly newsfeed, hourly newsfeed, whatever you want to get these scheduled news feeds.

And you could do other things with it too.

What makes the scheduled tasks better is you can have it and go check your Gmail every morning.

What are my most critical messages?

It can check your calendar.

Whatever it's got a connector to, it can go in and check.

So really move beyond just this news feed.

So I'm going to read OpenAI's description of scheduled tasks.

Scheduled tasks are a ChatGPT feature for proactive work such as reminders, recurring tasks, daily briefings and monitoring, similar to routines in Claude code or automations and codecs.

That's my language.

All of the ChatGPT models can be used except for Pro models, which is not to be confused with a pro subscription, which is a different thing.

It's very confusing.

If you have the really cheap paid plan, you can have three active tasks plus, which is the $20 a month plan.

You get five.

Business and edu users get 10.

Pro and enterprise users get up to 15.

Rob, I sent you a compendium of several task news feeds around AI in the workforce.

So what did you think?

No, I thought it was good.

It didn't surprise me a lot.

It's the things that have been showing up on my radar.

But what I liked about what you sent was it summarized it nicely and so it's very short, easy to consume, and it provided references to where those news articles were.

So it gave me the ability to go in and say, ooh, that's a statistics I've been looking for.

Or I don't believe that as much as I want to and I can go and see for myself if that really is what that article says.

So I I could truly see this going back to an era where I had customized RSS feeds.

If you're familiar with the old days of RSS feeds, there are places where I pointed towards what would be news articles that would feed me what I wanted to read.

It ultimately went away.

A lot of the places that supported RSS feeds quit supporting it and doing those sorts of things.

Except for podcasts, right?

To me this looks like a very strong replacement for the ability to get that catered news from a variety of sources in a way that helps me to be informed about the things I want to be informed about.

I am concerned how this is going to affect the news business when we look at the revenue models for how different news companies work.

So there's going to be some interesting tensions as this plays out, but I can see this being super useful and removing the necessity for me to go to certain web pages to consume news.

Yeah, yeah, I can See that I really like routines and Claude code and automations and codecs better at this point.

But the scheduled tasks, it's pretty new.

It's a pretty new feature, so I'm sure it'll get better.

It was very repetitive.

So I think I sent you three days worth of these and at least a couple of the articles were repeated in multiple days.

And so I would probably go into Codex and do an automation before I'd use these, but they're really easy to do.

I mean you just go in, click on schedule task and it kind of asks you what do you want to look into?

And it works pretty well.

But.

Yeah, what I look forward to these getting to and it may be that Quad's doing this better than ChatGPT is the use of memory so it remembers what it's done.

To me, that's the big piece is that balance between repeating information and missing out on giving you information and figuring out how does it keep track of what you've already seen and consumed.

I share that concern and I think that's one reason that the routines and automations have a leg up here, because you point those to a folder where it's going to store whatever reports it creates and you can tell it don't repeat things.

Another cool feature that you and I played around with is all right, at the end of every week, give me some sort of a compendium of everything that happened this week.

So but there's nothing wrong with the scheduled task and I think as I play with them more to do things beyond just a news feed, it's going to get better.

And here's what's going to happen, Craig, is I think this was good enough to go to market with.

And what we're going to see is as use cases like this are pointed out, it's only going to get better.

So three weeks, six weeks from now, we may look back at this and say, wow, OpenAI saw what was going on and made it better.

Yeah, that's the way these things have gone all along.

But I do want to take an opportunity to briefly talk about a few of the things that we saw in the Pulse reports that were pretty interesting to me.

The most interesting one was this idea of entry level work being seniorized.

And I'm going to read this so I get it right.

So this was out of PwC's 2026 Global AI Jobs Barometer based on more than 1 billion job ads globally and 2.4 million US entry level roles.

PwC reports that AI exposed entry level jobs are now seven times more likely to require traditional senior skills such as judgment, leadership, creativity and face to face interaction.

It scares me that face to face interaction is considered a senior level skill now.

But okay, I think that tells us two things.

We're starting to see some of this kind of apprenticeship pipeline collapse where they're just not going to have these entry level training type jobs, which that's been shifting for a long time now.

But the other thing is that now I want to read that list again.

Judgment, leadership, creativity and face to face interactions.

Those are the humanistic skills that we've been trying to emphasize more fully for decades in higher ed and have largely had lip service from employers around.

We want these skills, but we're going to hire for technical skills.

So maybe that's shifting, I don't know.

It is shifting, Craig.

I really do think it is not in this report but things I've been seeing are talking about how while technical skilled majors are seeing decreases in employment rates and unemployment coming out of college, we're seeing more of the liberal arts degrees which have those sorts of skills baked into them, being targeted by employers and hiring people.

So I do think the pendulum is shifting, if you will, from requiring all these technical skills to saying you know what, you can have the technical skills but if you don't bring with it those more liberal arts humanities based skills to the equation, you're not going to have that value.

Add that we expect to see in people that we're hiring.

I'm curious, have you seen actual data about that?

Because I've been hearing about liberal arts grads getting hired more and more for years and the only data I've seen has been from top schools.

Yeah, I think it was reports out of places like Harvard and things like that.

A quick headline.

Liberal arts degrees have long paid the worst salaries, but Microsoft chief scientist says in the age of AI they'll be really important for Gen Z.

So do see organizations are at least saying this in the press in ways that they haven't been saying it before.

So yeah, we'll see, we'll see.

I'm gonna to keep myself as skeptical there, but at the same I think we're in the early phases of a shift in how the job pipeline is going to work.

There's an archive paper that came out a couple of months ago that claims that labor statistics are undercounting AI disruption because the real damage is to tacit knowledge transmission through early career work, which is something we've been saying for a while now.

That's Something that I think is a real concern is there's the skills of what you can create, what that value production is.

But how do you learn the norms of an institution and how they operate and how you get things done?

Which I don't think college can ever truly teach that on the micro level, perhaps at the macro level, developing students that have an attention to and an ability to see what's going on and know that's an important thing to learn.

But that will always be important anywhere you start.

I can say that's true of universities.

Having worked at my third institution now, every university I've worked at has had different tacit knowledge, things that you needed to know to get things done.

And some places it takes longer to figure that out than others.

And I don't know, you can teach awareness of that, but you can't really teach how to clue into a culture and tap into that tacit knowledge.

You just have to, because it's tacit knowledge, you have to gain it through experience.

So I know we'll see.

And I think we may talk about this archive paper on another episode because it's pretty interesting.

One last thing out of the Pulse reports that I thought was pretty interesting.

Reskilling is shifting from training to continuous work redesign.

So in the past when we've talked about reskilling, it's largely been about how do you bring new skills to old jobs, right.

How do you upskill?

But now the very nature of work, especially knowledge work, is changing and it's going to continue to change because these models are going to continue to evolve in their capabilities.

And so the way even somebody that's really good with it today uses it, it's not going to be the same way they would use it two years from now.

And so I don't know exactly what this means, but the data point that stuck out is that employers expect 39% of workers core skills to change by 2030.

By 2030, that's not that far off.

And if the global workforce were 100 people, 59 would need training by 2030.

Yeah.

What I think is interesting about this, Craig, and I've been observing this happening in my own spheres of work, is there aren't necessarily people that exist that know how to give you that training.

Right.

There is not a lot of the use case based expertise.

You might be able to teach you how to use some of the capabilities.

But figuring out how to use these tools to change your workflows and your processes is very much an individual journey, which seems a lot different than how we've onboarded new technologies in the past that there was, okay, here's a new system.

We're going to show you exactly how to use it and what you can do with it and how it's going to change your workflows to here's technology.

Figure out how you can use that in your day to day life and then maybe we share it.

So there are some real tensions, I think, about what that process of discovery and growth looks like as these work processes change.

I want to go back to what you said earlier to support that even further.

Those are the skills like critical thinking, discernment, decision making, that sort of thing that we should have been teaching all along but have been squeezed out by the need to teach technical skills.

And when I say technical skills, I mean more job focused skills rather than the broader skills.

So I am a huge believer in what liberal education tries to do.

I think sometimes it doesn't get done very well.

But we really do need to be doing what the Jesuits have talked about for over 500 years.

And that's educating the whole person.

And I think that's going to be more and more critical.

But we don't know how to do it very well.

Yep, yep.

It's going to be a journey, Craig.

So we talked a bit today, Craig, about technologies and how they change and do different things and how people are wrestling with that.

Notebook LM has made some changes as well.

What are you seeing with that?

It has and I, I love NotebookLM.

I was an early adopter.

I've used it a lot, but I kind of got away from it.

I got obsessed with codecs and cowork and, and just poor little Notebook LM got pushed to the side a little bit.

But my wife has started attending an Episcopal church and they have something called the Book of Common Prayer.

It's a thousand page book, the guides of the religion.

It forms the core of the religion.

But it's a lot to try to figure out.

And so she asked me if I could do some things with AOD to pull out excerpts that are commonly used in services and that sort of thing.

They do a lot of column response where the worship leader will say something and then the congregation says something back.

And it's hard to read if you're trying to find it in a book.

But you know what?

I'm going to pop this into Notebook lm.

And so there's a free publicly available PDF.

This is all legal and on the up and up that the church puts out, I put that into Notebook LM and I'll put a link to this in the show notes.

And she was blown away because it's really cool.

We forget how cool it is.

The ability to do flashcards, quizzes, ask questions, explain this thing to me.

So it's really, really nice.

But a couple of things stood out that I had not seen before.

One is short videos.

Rob, did you have a chance to look at the couple of videos I sent you?

Yeah, they were really good.

They've come a long way since the last video I saw out of NotebookLM.

Yeah.

And they were what, a minute, minute 10, something like that, little shorts and they're in the vertical, you know, the portrait orientation that's popular on social media.

But I think this could be really nice for higher ed, where I think one of them I sent you was about hallucinations.

You could give students that one thing.

It's in a format that they're used to watching and in a minute they can understand a little bit of why these hallucinations occur.

It's not in depth knowledge, but could be useful.

Yeah.

When I looked at it, I put my brain in the mode of like security awareness training type videos that I've had to watch.

You could customize some things and do some things really slick and cool and do it at little to no cost as an organization.

So from a training perspective and even going into the classroom, we could pivot how we engage with students in some ways to really meet them where they're at and with content that we control, not necessarily going out and finding someone who's already made that video that we want to bring into the classroom and.

Grounded in the content that we provide.

Because remember, NotebookLM builds things and answers questions according to the data sources you give.

Doesn't go out and search the web or anything like that.

So it's just based on what you give it it.

I just picked the built in prompts that it suggested, but you can put your own prompts in to really tailor those instructors.

You ought to check this out to see if there's some way you can use it to better engage your students.

Those of you who are in student services, this could be great.

How do I talk to my advisor?

Or what's the best place to go for sushi on campus?

I don't know.

But whatever it might be, you can ground that in your own policies and your own processes.

So I strongly encourage listeners to check that out because it took, I don't know, took me like 30 seconds to create the video.

So it's not like it's a big labor intensive thing.

Also, the slides are much, much better than they used to be.

Much better.

You can now revise slide by slide.

So if you want to edit a slide, it'll pop up a slide with a little chat box at the bottom and what do you want changed?

And so that could be very useful.

You can also see the prompt that was used to create the slides, which could also be useful.

Yeah, that's one of the things I'm loving about.

Whether it's this notebook, LM tool or other places.

I have always hated making PowerPoint slides.

Not because I don't know how, but because it was tedious and just took a long time.

And that's one space where we have seen huge efficiencies in creating something that may be necessary as part of a communication medium, but doesn't need to consume half of it day to get it done.

No, not at all.

Yeah, especially if you're clever and you use AI to help you refine your kind of your slide deck outline and then turn it over to whatever tool you're using.

But ChatGPT, Claude Design, some of these tools are getting really good at doing slides.

So I agree with you.

That is a wonderful thing.

I've got one more quick thing.

We are almost ready to launch the AI Resilient Learning Activities Database.

I would love for listeners to go to aigostocollege.com and subscribe to the newsletter and make sure you get the word when we actually launch this.

It's getting pretty close.

I'm working on a couple of tiny technical details, but what we're going to do is we want to build a repository where people can share assignments that have worked for them, not just in is or in business, but in whatever discipline you happen to be in.

And then we can learn from each other because it's so fragmented right now.

And this will be 100% free.

It's funded through a Just Business grant, donors from the College of Business at Louisiana Tech.

So once again, to quote Animal House, it don't cost nothing.

No, I'm looking forward to this, Craig.

I think this is going to be a great tool as we enter into fall.

That will help break down the silos or the isolation, if you will.

Utilizing technologies to share what's working as we all navigate how to ensure our students learn what they should be learning in the classroom.

In a world where AI is not going away.

It is not.

All right.

I think we have gone long enough.

Long episode.

But we talked about some interesting stuff.

Rob Any last thoughts?

Nope.

Craig, I think we touched on it all.

All right, that's it.

Thank you very much.

Talk to you next time.