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.

I'm joined once again by my friend, colleague, and co host, Dr. Robert E. Crossler from Washington State University.

Rob, it's hot here.

What about up there?

We are starting to warm up.

We crested 80 degrees yesterday, Craig.

We're actually below 80 at the moment, but that will not last.

Okay, let's get right to it, because we've got a lot to talk about today.

Today.

Rob, you want to start us off?

Yeah, Craig.

It's been an interesting news cycle lately.

Anthropic dropped Fable 5, which was related to Mythos, which wasn't allowed to be released, and they scaled that back, releasing Fable 5.

And it's been available.

Not available, difficult to be available, expensive, capable, all sorts of interesting things around that.

So I'm curious if you've had a chance to play with that and what your thoughts are.

I did have a chance to play with it.

So for those of you who might not know, Mythos was this model that Anthropic came out with that was so good at exposing, primarily at exposing vulnerabilities, that they did not release it to the general public.

It was released to a very small number of highly secure organizations not long after the hype over meet them Mythos, they came out with Fable 5, which was just a guardrailed version that's driven by the models in the Mythos family.

So there was a lot of hype about this.

Apparently, for coding in particular, it was phenomenally good.

I mean, not just good, but, like, blowing people away.

Good.

I used it, and it was good.

They were making it freely available at the time through June 22, to, I think, at least paid users.

I don't know if free users.

Just for context, Craig, we're having this conversation on June 16th.

So freely available to almost a week from yesterday,.

Until the federal government placed import controls on it and Anthropic pulled the model.

It was interesting.

I got up really early in the morning.

I got up, was using Fable, and then I wasn't using Fable.

And it's like, what's going on?

I thought, it's Claude down.

And then I went and tried a different model, and that worked.

So I went to their status page and said it was unavailable, and then I got a message that said, it's temporarily unavailable in the clawed app, but apparently there's this import control restriction that was placed on it for reasons that are still very murky.

And so Claude just pulled it because they didn't have a way figured out to restrict it to users of only certain countries just yet.

So we'll put some links in the show notes about this, but I was pretty impressed with my usage and I think it's really good for highly complex multi step reasoning based operations like going through a code base, a bunch of computer code and saying here's a security vulnerability, here's an inconsistency in the code that's really hard to do because it involves a lot of steps, a lot of back and forth and it was quite good at that.

And so then they pulled it.

No word on when it will be back.

I've been watching Reddit and some other spaces and it's everything from tomorrow to never from all the experts that don't really know anything and so we don't know when or if it will ever come back.

Yeah, and some history with Anthropic just for people who aren't following this, it'll be interesting the politics of AI because I think it's going to drive a lot of what happens as capabilities increase, whether it's export controls or we only want something developed in our country, used within our country.

The Pentagon was aligned with anthropic for a while.

Ultimately Anthropic didn't like how the federal government was going to use their product.

And so some relationships changed and OpenAI became the AI flavor for the US government.

So there are a lot of tensions between these different companies and governmental agencies that just color this whole conversation.

Yeah, the conspiracy sphere is having a lot of fun with this one.

This is where I think about conversations in the classroom with students.

It's not about using AI in these conversations, but from an organizational perspective.

As you are looking for these different tools to use, how do you decide which ones you're going to license and utilize?

What are the pro and con analyses?

Where are you going to get your best return on investment?

There's a lot of really interesting conversations around decision making in this emerging technology space where prices are changing regularly, capabilities are changing regularly, there's geopolitical arguments happening regularly.

It creates a really interesting space to have some real time.

What would you do in these situation type conversations?

I don't want to go too far down this path because it's only tangential to higher ed, but models are being built off of models.

And so if this is a supermodel, not that kind of supermodel, if it's a super capable model, that really moves the bar on what that Means other entities, often state actors, can take these models, decompose them, do all kinds of technical stuff to create their own models.

And so there's some real.

I don't know whether what the government's doing is legitimate or reasonable, who knows?

But you can paint a picture where it's pretty reasonable to not allow this out into the public.

But let's for the moment say this all gets figured out and we get Fable 5 back.

My big concern is we are going to create an even bigger gap between the haves and have nots.

Let me just give you some numbers.

Claude's models are, in terms of their tiers are Haiku, Sonnet, Opus and Fable.

In terms of their capabilities and more importantly, their expense.

I usually use Opus 4.8, something that if I was paying for it through credits and not through a subscription, that would cost me about $25 to run.

An Opus 4.8 would cost me about $50 in Fable 5.

So it's about twice as expensive, Is that what I'm hearing?

It's roughly twice as expensive.

And you can go get this data from Anthropic Haiku would be about five bucks.

So 10% of the cost of Fable 5.

So we're looking at a significant increase from their lower end, very accessible models to these super capable models that are hitting the market.

And what made this even more interesting is until June 22nd, at the time, you didn't have to pay extra to run Fable 5.

It would drain your quota faster, but you didn't have to pay extra for it.

You just couldn't use it for some period of time until your quota reset.

But after June 22, they were going to start charging usage credits, which roughly translate into what I just talked about with the costs.

And so here's where it gets problematic.

After June 22, let's say I wanted to write a lit review.

They're doing annotated bibliography, which is, I do a lot of annotated bibliographies with these tools.

If I ran it under Opus 4.8, it cost me nothing.

It's included in my.

I have $100 subscription, but it would work under a $20 subscription.

I'd have to pay 50 bucks to do that with Fable 5.

Well, you know, if you're at a big, heavy, Ivy League, high end research school, maybe that's not a problem.

Or if you're a rich kid, maybe that's not a problem.

If you're a kid from a poor background, less well resourced background, or you're at A less well resourced institution, all of a sudden you can't do that, which means you have lower capabilities versus these other institutions.

And that gap is just going to widen and widen and widen.

I think that's a serious concern that I have not heard anybody talking about yet.

But I think we need to.

Right.

And to put that into the perspective of the classroom.

Right.

Because the research argument, I think we're going to see some really interesting things around research and the number of manuscripts that are being submitted, especially as AI tools allow for people to quickly do things in ways that they've not been able to before.

But if I'm in the classroom and one student has access to the free model of Claude and another student invests in Fable 5, the disparity between what they're creating to turn in with these tools is going to be huge.

And so it really forces you to look at assessments in a way to ensure that someone's not going to be penalized in their learning outcomes and learning assessments just because of capabilities of AI models.

And I'm not certain what those solutions look like, but it should cause us to pause and say, how does the existence of these tools impact the learning outcomes that we're looking for with our students?

I agree with everything you just said, but it's even worse.

It's the ability to learn.

We saw this when it became the norm to get GMAT and GRE and SAT and ACT prep courses back in the day.

You might get a book and that was it.

And most people could afford to buy the book or you get a book from a friend who took the test before.

But then it became these tutoring centers that would do the prep.

And we started to see this separation in test scores not because of innate ability, but because of resource availability.

It's going to be like that on steroids, I'm afraid.

So we need to be ready for this because it may not be Fable 5 that comes back, but it's going to be GPT 5.6 or, you know,.

Whatever, Craig, I'll push that in a little different direction.

I wonder if as people begin using these more highly capable models versus those who aren't, if the people who aren't using these highly capable models might actually learn more because they're not going to be able to cognitively offload as much of the things that they're doing and they'll need to be more involved in it.

So I wonder if learning will actually be increased with the more simplistic models than the more advanced models.

And it might actually be a more useful tool for a learning perspective.

It could be.

We just don't know.

But that's a reasonable projection.

So there's another higher ed angle that's not being talked about.

God, that sounded like AI wrote that.

Here's the story no one's talking about.

So I was thinking a lot about this with my own use, and then how I'm going to help my students think about it as well.

I think of Fable 5 as being the really expensive employee.

So you don't need somebody that's making a couple hundred thousand dollars a year to do routine work.

And so you don't use Fable 5 to do something that haiku can do.

So one of the things we're going to have to help students understand is how to match model capabilities to the task.

And that's something that really isn't being taught in any widespread way that I know of, but it's one of the things that can help us mitigate not only the resource effects, but also the learning effects.

You know, if you're really stuck and need an expert tutor to help you get unstuck, Maybe it's worth Fable 5 if you just want to check your understanding of some basic concept.

Maybe haiku is enough to tell you, nope, you don't quite have this right, or you got it.

So I think it's a direction we're going to have to go sooner rather than later, as if we needed something else to do, but there you go.

Well, and I think this all points towards Craig having opportunities, as we're teaching classes, to engage with the news of the day.

And what does that really mean?

Because if I sat back and designed a class today to address these things we're talking about on this episode, by the time I get to late August, early September, when my class is being delivered, what I designed today is probably going to be different than what I would design in September.

So being able to bring in just kind of live case studies, if you will, or live news assessments, and what does it mean?

How does this force us to think differently is going to be, I think, a crucial tool that we utilize within higher ed.

And that's just one more thing that faculty will have to do, is they're going to have to stay up on some of these developments, which is why they should listen to AI Goes to College.

And remember, you can do that by going to aigostocollege.com if you want to follow us.

It's AI goes to college.com follow all the major podcast apps are one click maybe two clicks to subscribe from there.

That's a nice segue into that Science article that we both read this week talking about AI misuse or students cheating with AI.

I'm very hesitant to call what the students did cheating per se, but that's how the article referred to it.

And what does that tell us about students and what they're using?

Can you talk about that article a little bit, Craig?

Yeah, this came from Science.

It was a massive study.

More than 95,000 students across 20 major public research universities in the US.

It just came out in May.

The authors were Chirikov, Smirnoff and Kizilek.

I can't pronounce that one, but there will be a link in the show notes.

And they were out of Cal, Berkeley, Cornell and University of Technology, Sydney.

And it's just so huge.

That's enough reason to pay attention to it in and of itself.

So here are kind of the big findings.

Their methodology was interesting because they didn't just ask, did you cheat?

They asked some kind of masking questions.

I think they asked like four or five questions, and one of them was about, did you cheat?

And they didn't actually ask, did you cheat?

They asked something to the effect of did you utilize AI in a way that wasn't allowed.

So they weren't actually asking the did you cheat?

Sort of question.

Yeah, they weren't quite that blunt.

So 2/3 of the students surveyed used generative AI in the past year.

The highest disciplinary areas in terms of usage were in stem.

Not a big surprise.

And this data was collected in 2024.

So this is two years ago.

This is how much people were using it.

I would argue if they did the same study today, we would see the numbers increased.

Yeah.

Although I'm always a little bit hesitant to read too much into this because there's using AI and then there's using AI.

I mean, I use AI, you use AI.

There are a lot of people that use AI.

So all usage is not the same, but still, it's a useful statistic nonetheless.

But about 9% of those users turned in AI generated work knowing it wasn't allowed.

No surprise.

The misuse is concentrated among daily users who cheat at four times the rate of monthly users.

Although wouldn't if that was proportional, wouldn't it be 30 times the rate?

So I'm not sure how to interpret that the inappropriate use runs higher in non STEM fields, even though the adoption there is lower.

There's a lot to look at here and a lot that they really don't get into in the paper that I think is worth discussing.

Just for the record, this paper is three pages long, so it's very much more about what did they do, what did they find with a few recommendations for what does this mean for or academia.

It was not a 30 page paper filled with all sorts of theory and suggestions and those sorts of things, which is pretty typical for a science article.

Although in an interesting aside, they have an addendum that's really long that goes into all the details if you want to dig into those.

But to your point, yeah, it's not hard to just get this paper and read it for yourself.

So 62% of computer science students reported using generative AI versus 24% in the arts.

Business and economics were right at 50% each.

So it's not really the STEM, non STEM.

Although business is so broad, some of it morphs towards STEM if you look at finance and information systems.

So it's a little bit of a blurry line.

What I found kind of interesting is this mix of inappropriate use and usage rates.

There's a giant figure in the paper where, for example, computer science, over 60% reported using AI, around 10% reported using it inappropriately as opposed to if we look at journalism, what looks like about 16% used it inappropriately versus about 36%.

I'm reading off of a chart, so I'm estimating these a little bit using IT overall.

So that's a huge proportion of the users that use it inappropriately relative to computer science.

Business was kind of in the middle.

I wonder, Craig, how many of those numbers reflect disciplines that we're more willing to adopt, that it's okay to use these sorts of things, because I know a lot of computer science science programs are becoming AI programs and have realized that they need to be adopting this and utilizing this.

I do wonder how much of that reflects it.

Where I talk to my marketing communications colleagues and they live in a world that appreciates more the human nuance to writing and the importance of that.

And so there's different views on what's allowable and not allowable in some of these disciplines.

Well, I think it's not just that you hinted at this, it's the maturity of AI use in those areas.

So you and I and the computer science folks and some others since almost the beginning of the ChatGPT public release have been trying to figure out all of this and how to deal with it, because our students were going to be the ones that grabbed onto it first, because that's what they do.

They're technology folks.

And so we've had to wrestle with this and maybe we've drawn the lines a little more clearly for students, but that's all speculative.

What I think this points at, maybe more importantly, is what it has to say for policy implications.

Rob, you're a big policy guy.

What do you think?

Yeah, I think to ban AI as a flat policy is going to create a world where more and more of this is going to happen and we're not going to be able to enforce it because it's nearly impossible to detect, at least without bringing in biases and different things.

So I think open communication about how to use it, where to use it, and being more nuanced in how you define the appropriate way it's used in the class and in higher education settings is going to be crucial.

Because if you try to put your head in the sand and just say you can't do it, you're probably going to have a classroom space where you're not sure who has learned what at the end of the day.

Yeah, I mean, that's always been folly to ban it.

Quick story.

I used AI to write something using one of my write like Craig skills, spent about two minutes editing it, put it into Pangram, which is one of the better AI detectors, 100% human generated.

So with just a little bit of skill and effort, it's pretty easy to get around these AI detectors.

And then there are all kinds of other problems.

We don't need to get into that.

But I think this is just one more indicator of what you said about policy, but also the need for assessment reform.

And one of the things that I read in this article was the need to invest in faculty development.

And I absolutely agree that faculty need to figure out how to do these sorts of things.

But I also know that higher education has never been terribly great at investing heavily into faculty development.

It's a learn as you go experiment.

If nothing, we need to create a safe space for faculty to be able to experiment with what they're doing purposefully, and a space where faculty can share with each other what they're learning.

So that way we're not all reinventing each other's creations and various different things.

How can we create this space where we grow together in a meaningful, purposeful way?

Although that's one of those that's easier said than done sometimes.

But to me, the bottom line here, it's not just the interesting mix of inappropriate use versus use.

I mean, that's interesting, but the big deal is that We've got to change how we do assessment because it's not just that students cheat or can cheat, it's that it's no longer a trustworthy signal of what they know if we keep doing things the way we've been doing them.

So that's a big job.

But we've been saying this for the entire time we've had this podcast we have to get after this.

And I would say if you're doing nothing else this summer, this is a space where you've hopefully got some free time to think about what this looks like without the day to day grading and different things that may get in the way that make it hard during the academic year.

So give yourself the freedom to step aside and maybe work with a colleague to see what is possible.

All right, my turn to segue.

Rob, you're doing some things to help ease some faculty time burdens over the summer.

Tell us about it.

So we've got a couple of student interns working for us in the college that I'm giving them some ideas that have been in my head and helping them develop some AI tools that I think will be helpful for faculty.

The first one that I've been toying with, and we're probably about 80% of the way there with what this is going to do, is to take what was our syllabus template and this goes out to faculty over the course of the summer and there's updates or changes to what needs to be in the syllabus.

Kind of the whole premise is here's what the new template looks like.

Please compare it to your existing syllabus and make sure that you're communicating things in the way that we hope you would from a university or from a college's perspective.

And one of the things that I've always, I guess, hated about it is it's pretty tedious, pretty time consuming to sit and take my five page syllabus or however long the syllabus is, and look at what's new or different or changed in a syllabus.

That's the new template and what do I need to change and where is it at?

And so to create a syllabus template agent which will help faculty can upload their syllabus to it and it will pop out.

Where are the places that a you've got alignment so it gives some positive reinforcement for places where it matches and then where it's not aligned with what the template looks like it give for how to update and where to update and make those changes.

So it takes that human comparison process out and allows for the faculty to very quickly and easily find places where they might need to make some changes.

I have an unfair question for you.

First of all, that's a great use because you said tedious.

It's mind numbingly boring.

But are you going to use this to evaluate.

No, I don't plan on it.

We're not going to store people's syllabi and we're not going to evaluate them.

I want to keep the AI space a trustworthy where it's helpful and not seen as evaluative.

I look at this truly as a tool that puts the human in the loop to change their syllabus as they need to.

One of my fears is, are you just going to have it write people's syllabus for them?

Which would be nice, but I would be afraid that we'd have a syllabus that the faculty didn't know what was in their own syllabus, and it would create potential problems down the road.

Well, although I want to push back on that a little bit.

So what if.

If it created the syllabus 80% of the way or 75% of the way?

I mean, that's kind of what we do, right?

Most schools have all this syllabus language that's got to go in there, and it may or may not have a standard format, but you've got to have these different elements and you've got to have these different policies and all that kind of thing.

I mean, if it can write that 75% of the way there.

Yeah.

And one of the ideas I'm toying around with once we get the comparison tool made is what if I tweaked the agent to have it ask a series of questions like what's your name?

Where your office located?

What are your office hours?

And from some prompts it would.

What is your question?

What is your question?

Some Monty Python Holy Grail in there, perhaps, but through some prompts and some questions that it would actually create exactly what you're saying might be a version B of this, but for now, a comparison tool is where we're starting.

Well, I think I mentioned this, but I'll tell you what something like Codex or Claude Code Cowork is really good at is updating your syllabus.

I mean, I always screw up some date because I miss a holiday or, you know, delete a row or something like that.

And it's really good at going through and doing all that kind of thing.

So another tool that we're making is around active learning and giving people an opportunity to a upload activities they might already be doing to give them some feedback and some suggestions for how they might incorporate that into a more active learning, engaged approach, or even uploading things like the PowerPoint slides they'd be using and say, how could I turn what might be a death by PowerPoint approach to teaching to be more active engagement getting to those same learning outcomes.

So really a tool to help brainstorm.

Kind of the same way I've used AI to help me do these exact same things with my own prompting, but to bake those into a tool where it's just, here you go, give me some suggestions.

And then helping to develop new ways of approaching those learning activities in the classroom.

Are you using Copilot?

I am for this, Yep.

So does Copilot have skills like Claude and Cowork?

It does.

That's something I really think people ought to investigate more.

They're really flaky in ChatGPT, but they work very well in Claude.

But you could pretty easily create an active learning skill where you could take some set of learning objectives, give it a little bit of context, and have it create a library of learning activities.

Just a little bit of an aside.

People should investigate skills because they're really, really useful and you can call them from within any conversation as opposed to something like a custom GPT or an agent where you have to be engaging with that thing first.

You can be in some random conversation and invoke a skill.

Yeah, I like that idea.

And then the one last thing that I haven't started working on that's on my radar is an AI resilience detector to help people, to have a partner to help them look at the activities, the way they're grading, the way they're doing assessments, and to give feedback for how this holds up in a world where AI exists.

Yeah, that could be huge.

You should make that one publicly available and help all of us.

Those seem like they're very useful and great ways.

I don't know if this was your intention, but great ways to help faculty see the potential, even if they're reluctant adopters of AI.

If you give a faculty member this AI tool and said, look, I know you don't like AI, but if you put your syllabus through here, it's going to save you an hour of reviewing your syllabus, yeah, that's a pretty high relative advantage.

So, yep, that's the goal.

One of the things that I think is important as we think about encouraging people to adopt AI into their roles is to not say this is what you have to do, but to say, here's some tools that can help you think about how to achieve what we're all trying to do, which is to ensure our students come into our classro learn what we hope they would learn in our classroom.

I thought what we were all trying to do is avoid committee work, but I'll go with yours too.

Maybe I could create an app that'll help us.

That's another strong contender.

Do our committee work for us.

There you go.

Virtual Rob.

Associate Dean's meetings.

I'm going to throw you a little bit of a curveball, Rob, because we're talking about agents.

Quick backstory.

We have a class, CIS125.

We started renumbering, so now it's 125 3.

So when you say we, Craig, you're talking about.

Oh, Louisiana Tech.

Sorry, sorry.

Pronoun reference problem.

Yes, Louisiana.

I dated an English major once that said pronoun reference to me an awful lot.

So this CIS125 class, all the freshmen in business school, a lot of freshmen from across the university take it and it's kind of Excel and some other stuff.

Not just hands on, but a big hands on component.

Well, there's a lot of institutional pressure to integrate generative AI in that course and we've done that to some extent, but we need more.

So I was thinking about this problem and thought this seems like a good task for an AI agent.

So I gave it to Codex, which is C O D e X.

That's OpenAI's GPT based coding agent.

We've talked about that before on the show.

I gave it a couple of syllabi from the course, explained what I wanted to do, set some parameters, gave it some context and said write this handbook for me.

And I literally one shotted it.

We had a little conversation back and forth and I said I don't want to do this chapter by chapter.

I want you just to write this thing.

And it was good, not great.

In about 20 minutes it was a 70 page handbook with learning activities.

We shaped it around skills here.

Like I can't remember, it was four or five skills related to AI.

It tied it into the classroom.

Like there's a heavy Excel component.

This is how you use AI to help you with spreadsheet based problems, that kind of thing.

I wasn't happy with some of the activities, so went back and forth on how to clean up the activities and in another half an hour or so had a much better Handbook that frankly, if we just gave it to students, it'd be fine.

But I think with a little bit more work where the human comes in with some true refinement, it could be a pretty substantial learning resource for our community.

But think about that.

You and I with Franz Melanger have a pretty successful textbook, but when Franz and I wrote the first edition, we spent like a year and a half on this thing.

Now, what Codex created isn't up to that level, but it's a big head start.

I mean, it would have cut a half a year off at least easily.

So I think people need to be thinking about, can you do a bespoke textbook?

Or in this case, it doesn't replace the textbook, it's a supplement to it that's really necessary.

So I don't know, Rob, what's your reaction to that?

I think it's one of those things that begins to change the world of knowledge creation.

So if we can do this with textbooks, all of a sudden can have very customized classes.

I know a lot of times I can speak to the MIS curriculum, but I would imagine this is true for most every class is what you do in a particular class is aligned with one of three or four major textbooks that put the chapter by chapter of what you're going to do during that semester.

So degree at Washington State University in MIS may be very similar to one at Louisiana Tech because we've adopted similar textbooks that take similar approaches to teaching these topics.

Well, in a world where we can very easily make books this way, we could customize and tailor what is the flavor at Louisiana Tech versus the flavor at Washington State University and begin to have less similar courses based on whether it's the needs of the states we live in, who are hiring our students and the different employers there, or even what pathways are we setting students up for their graduate school experiences or wherever their life is going to take them.

Yeah, the nuances of the curriculum, you know, we have different course flows and that sort of thing.

And the student body, you know, Washington State and Louisiana Tech probably have some differences.

So yeah, it's early days, but it's pretty intriguing to be able to do something like that, to plug a hole.

Well, and the other thing I would say is, Craig, because you created that the way you did, it's no longer somebody else's resources, so you could build class agents and tutors and bots around that very thing because you're beginning to own that knowledge content and what you've created.

So it's, yeah, really potentially paradigm changing.

For how we approach the classroom experience.

Absolutely.

And I will make this commitment right now.

When that is finished, I will make it freely available under a Creative Commons non Commercial Attribution License or whatever.

That one is like 10 of them.

So, yeah, it was kind of scary.

I want to close with a couple of pro tips, Rob, that I've kind of developed over time and experience.

Those are the handoff documents and what I call theoretical memos, but a normal person would just call memos.

So two problems that I've continually faced using AI is I need to move from one conversation to another conversation, whether I'm running out of context window or I want to use ChatGPT on something I've been talking to Claude about or whatever it might be.

So I've started creating handoff documents, and I literally just say, I need to move this to a different conversation.

Please create a handoff document that will give this new conversation all of the context it needs to pick up from here.

Just something that simple.

I usually ask for it in markdown format, which is one that uses tags to do the formatting because it's really small.

But if I wanted to make it look prettier, I could.

They're easy to copy and paste into a Google Doc.

There's actually, if you right click.

If you've got this enabled, you right click paste for Markdown.

It comes in nicely.

They are insanely useful because, you know, it's just sometimes you just need to move to a different conversation.

So, Craig, when you use these, do you allow any of your tools to connect to Google Drive?

So then it could actually look through a folder on Google Drive to tell you things as well?

Yeah, although that's really spotty in my experience.

I just used Gemini baked into Google Drive.

I haven't shared this with you yet.

To give us a list of everything we've talked about so far on the podcast, because frankly, I lose track.

We're setting up the episode notes for the next episode.

It's like, I don't know, did we talk about this, or did we just talk about talking about this, or did I just think about talking about this?

And so it was pretty useful and it did a good job of that.

But you can connect these.

I think all of them have a Google Drive connector.

But even if you don't want to save the handoff documents, it's really good just to copy and paste it over to a new topic.

I save them because they're so small.

To quote Animal House, another great movie, it don't cost nothing to save it.

So I just save it.

The other problem I've always had using AI is I forget what I've done.

I'll have this great conversation.

I'll go down a different path, or a dog will bark, or I get a phone call or a text message or an email, and then I come back an hour later or two days later.

It's like, no idea what we were talking about.

Now I've got to scan through this whole thing if I can find that chat session, because I also forget to rename them into something useful.

It's just a pain.

So I started doing theoretical memos.

So this comes out of qualitative research.

When you have some insight, you're supposed to create a theoretical memo.

Any normal person would just call it a memo.

And so as the name implies, these are fantastic for memorializing an important chat or an important idea.

So when I get to something I want to make sure I remember.

I say, write a theoretical memo about this.

And I might say about this last, turn on the conversation or this last concept so it doesn't bleed over into early stuff.

I usually ask it to add a date to the memo title, the file name, so I can find it later.

I use markdown again because it's easy.

But this is my number one tip for anybody who's using AI more and more and more and finds themselves forgetting stuff.

I call them theoretical memos.

Whatever, put them into Google Drive.

Google search is so good within Google Drive that it makes it pretty easy to find these things later.

Any of you who are out there listening to this handoff documents and memos will make your life so much easier when it comes to AI.

Now that sounds like a great skill even to work with our students on as they begin using more and more of these entering the workplace and those sorts of things is how do they keep their productivity up without having to go, have I done that already?

Or am I hallucinating that I even did that?

That thing?

You'll love this.

When I was talking about all of this and creating one of These memos, either ChatGPT or Claude said, well, this will help keep people from.

How did it say it?

Having all of these disconnected facts in a big idea garbage can that never goes anywhere towards doing something productive.

I said, you've just described my entire academic life, this big giant garbage can of these interesting ideas that I never do anything with.

It's like, oh, that was depressing, humbling, and scary all at the same time.

So, Rob, I think we should end on that note.

Do you have anything else?

I don't, Craig.

I'm not going to add anything to your garbage can.

All right.

Well, thank you very much for joining us.

Join us again next time on AI Goes to College.

Remember, everything AI Goes to College is available@aigostocollege.com and we will see you next time.

Thank you.