Goal-First Workflow Design – Dr. Eric Perry

The practical side of artificial intelligence how real people use AI tools to get work done more efficiently, creatively, and productively. In each episode, we talk with professionals about their work and their scholarship, followed by an in-depth look at how they integrate AI into their day-to-day workflows. And whether you're a teacher, a researcher, or a professional looking for ways to sharpen your productivity edge, you'll find ideas with each one of the episodes that we have.

In in in my second episode, we welcome Eric Perry, educator, creator, co-host. of the Tech Savvy Professor Podcast. He and I do that together. He's known for translating emerging technologies into practical classroom ready workflows. He he brings hands on experience in content creation, course design, and productivity systems that help educators. He is like A really nice, gentle, thoughtful person. And the part that you don't hear about, Eric. That much.

But you you need to know he's like a tremendous parent and very much connected to his family. And so we talk a a little bit about that off camera when we're on when we're on Tech Savvy Professor and when we're on the circular firing squad, but it's real apparent that he is It's he is very much into his family and taking care of them and taking care of himself by being very much into his family.

Uh Eric has got some remarkable talents. We met many years ago through ACEs and doing technology together with them. And now He's off well, I'm gonna I'm gonna stop saying wonderful things about you and maybe just get a little background where you are and what you're doing.

So sort of and how you move from into this educational technology aspect and media production piece. From where you were to where you are now. So just let us know where your where your feet are planted right now. Yeah, I as soon as my cheeks stop burning. I appreciate it, Mauri. I I so right now I'm program director for an online counseling program based out of Virginia, Divine Mercy University.

I had a really weird move into counseling and training. I actually started more on the educational side of things. I worked in background investigations. My background is in criminal justice and was working on training background investigators and realized you know we were incurring a huge cost flying people in for three or four weeks.

To get them trained. So had decided we were going to start building out curriculums that they could use and learn remotely. And this was in Early two thousands ish. We were starting to do this. That's really where I got my start in instructional design, curriculum development, developing content and media, learning different ways.

that we could help facilitate education at a distance than bringing them on site for, you know, the the shorter part of their training. So that really started into that. I moved from that role after Finishing my counseling degree with a provider, Presley Ridge, who's

And I think seven states, seven or eight states now providing training for counseling professionals and working as a clinician there. So doing some of the same things. How do we provide education and content across the country essentially? So really just building technical skill from there and looking for ways to make learning easier, content development easier.

You know, and that's carried forward into my educational career, serving as an instructional designer, building courses and you know, now kind of overseeing curriculum and development and and different programs that I've worked in. Well you m you you got your master's degree and then you went on and got your PhD at where where was your degree from? So my master's was in Slippery Rock.

Yeah, that from Slipk Rock University and then my doctorate from Duquesne in Pittsburgh. Okay. All right. So Duquesne. So Pennsylvania focused kind of uh kind of education.

And how did the transition from the master's degree into the PhD influence your use of technology? And eventually we'll get to talking about AI, but I'm just I want to know that transition and then moving into the into academia with that. Yeah, I I think At the master's level I was really focused on building clinical skill and getting that kind of experience.

really wanted to work toward telehealth and and kind of distance work and things that we could do from a clinical standpoint. I think moving into the PhD, I knew I wanted to get back in the classroom. Uh I love teaching. I love the learning end of things. So moving through the PhD, I was really focused on refining those skills and thinking about how I transition from just delivering learning to delivering

more experiences. I think counselor education is a little bit different than some other content-based areas in that we're asking students to learn things that they're going to bring in relation to others at a distance. So that was something I think that developed over time and still is a really big interest area of mine. My dissertation was around counselor educators' feelings of efficacy in online education.

And really just looking at some of those factors that we needed to build in. And I think technology is a good conduit for that. How you know, how do we leverage it in a way that helps students really learn? Yeah.

So how tell me about your initial experiences with AI. So I ha AI has been it's been interesting. So I I've been following it kind of from the earlier versions of of GPT really. I think GPT has been the most accessible tool folks have used and there's there's certainly others out there that we could talk about.

Uh and I'll be honest, and I think this is what most people end up doing. I've wanted to find a way to source information easier. And I really thought GPT was a way to do that. And early on it it tended to hallucinate a little bit more than it does now.

But it was still a way to brainstorm ideas, to get thoughts out, to gather and collate information early on that was better than Google. You know, when you're using an internet search, and I think this is something that gets folks confused sometimes.

Something like Google or a search engine is gonna give you results based on the information that's out there. Something like GPT or a generative AI is going to pull those sources of information together and collate it and try and give you a response that it feels is Best answers your question. So it's not just about what you're asking for, but how you ask the question that creates specificity in the response. So it's it's just interesting.

So kind of uh as it's grown, I've I've started to use it for different things to help support my workflow, make things easier, and really gather information in ways that I I think help support processes and development.

Now, did you play around with some of the earlier chatbots like Eliza in terms of you know that we had in our field? And for those who are not in the counseling field. Eliza was a programmed Rogerian response. scripted kind of bot that you could say, I'm feeling really bad today. And Eliza would analyze that today and go, can you tell me more about that?

And as you would tell more to Eliza, Eliza would take fragments of your sentence and put it on to another question to kind of probe deeply. Did you play around with that? So only once. You know, I read about it, heard about it, and then got access to it. And y you know, if you remember like the little Microsoft paper clip. Mm-hmm. Yeah, Clippy. That was around forever, Clippy.

There are so many of these that were out there kind of prior to this that were like the earliest generation of kind of AI response bots. And they're they're taking information and responding based on what you're giving. in some type of curated way. They're looking for markers. And and I think that's that's where the earliest models kind of grew out of.

But when I think about that, I always think about Clippy. You know, I I had a an Atari twenty six hundred. That was probably my first gaming system. And there was a book that was published, I still have it down on my shelf, where you could write your own games. Well, you couldn't write your own games. What you were doing was you were putting in the code for uh for the games and the pages of code that you would have to type in to get the game.

But there is one for Eliza. And and I always thought if I had the time, and now it'd probably be easier because there's all sorts of ways you can get that input put in it. It would be fun to a and you read the code and you can see the affirmative responses that were written from person-centered Rogerian approach to to rewrite the code and mess around with a little bit. just to see if you could change the nature of the therapist by changing the response set that they had.

But when you started talking about, you know, interacting with AI, I love what you said about the differences between doing a Google search and kind of interacting with what AI is going to give you back or a generative a generative model will give you back and its purpose versus what a Google search is. It made me think of some of the some of the origins when I said what it what kinds of things have you played with.

When you get s when you got started on this track. Yeah, and I think you see a lot of these early on to the the AI chatbots for like customer service for sales. They got into this pretty early, kind of on the back of clipping, when you would go to a website and it would ask you, hey, can I help you with something? And and you could type in a question and it's looking for keywords. You know, so a lot of us have seen those and interacted with them and not even realized it.

Your autocorrect is a version of kind of an AI response tool. It's looking at what you've put in and basing it on a set of and kind of trying to see what the closest match is. So you know, there's some earlier generations of this that I think we've all had a little bit of experience with, but using AI now is a lot different, especially when we're talking about something like generative AI.

You know, one that I recently played with, I don't know if we've talked about Monsters Notebook LM, which can be a lot of fun too. Can create podcasts with fake Voices and breath sound, you know. Very, very cheery people. Too damn cheery. Excited about the weirdest things that you have put into your text. I don't I don't want to jump too far ahead, but

Hey Eric. Hey, Eric. How about you and I talk about manic depression? You know that's kind of the level of engagement. It's a ma for folks, it's amazing there, they think. This is pretty cool. For for people who've worked in this medium, it's like, God, that's so fake. Yeah, I think the attention to tone and and pace and those types of things, they do tend to get really excited.

But for me it was like the the little attention to the littler details. You know, they'll throw in little anecdotes and they'll have like like I said, the breath sounds and chuckles and different things, you know, so I think that's neat the way it can kind of create some of those things, even though it doesn't really fit most of the time. Yeah, I think I mentioned replica.

Uh it's an app. It's an it's a relatively expensive app, but it is essentially a more sophisticated Eliza with with a three-dimensional person that you can interact with versus text language that you did with Eliza. And, you know, there've been some press about how people have taken those things too far and have had some trying to s trying to deal with mental health issues using chat bots is not kind of a good source for that. for that process. But

It's interesting how it's grown. And so let's let's move it up a little bit. Have you is there anything that process that you're doing right now, which would be informative to listeners or viewers of this, that you ca you're willing to share in terms of how you do work. Yeah. So there's a couple of tools that we've been using. that I think have been really helpful. So I I talked a little bit about Chat GPT first and and I'll get into kind of process with that too.

But I've used the plod pin and Zoom AI companion to help out with note taking. I think this is extremely helpful, especially when we're generating ideas and we're starting to think about tasks and projects and different things that that we're doing. had a re meeting pretty recently where we were thinking about our enrollment growing and some of the processes that we had and and some of the course experience we have really aren't scalable for larger groups.

They're great for six, twelve, even twenty-four people, but when we start to get to sixty, eighty, ninety people, they don't really work. So we were thinking about ways in which we need to change those course experiences and courses and the notes can help. that generate both what we talked about and start thinking about workflow. I've taken those notes, put those into GPT and asked it to create tasks, workflows, and assignments.

so that we could start building out what it is we need to change and what we need to do based on our conversation. And it can pull that from the context of the conversation and the notes. So that's just one example that that I've used kind of patching in different systems, but I think it can be really valuable from beginning to end stage and and throughout a project to kind of help support. And that's where prompting becomes really, really important. Is there

You know, I'm gonna push you a little bit on this. Is there a specific step by step thing you could say, this is what I do, that would be something that a listener of your could could easily replicate within their own workflow? Yeah.

So uh uh I'll outline just a little bit. For those of you who are thinking about using it, I think the first thing to do is kind of set your goal. What are you going to use? AI for. You know, what's its purpose? What's its role? Some of the examples I can think of is it, you know, you're gonna create a syllabus or you're gonna create a rubric, you're gonna create an assessment, a test. You know, those can be really good examples of just things to try.

I think your second kind of step is to figure out which tool you want to use to do it. Tat GPT gets a lot of attention, but it's not the only one out there. There's several others that you can use and some of those that you may have used already but haven't really leveraged a lot. So, you know, Copilot, Microsoft Copilot is another example, Claude.

Perplexity. Yeah, perplexity. Yeah. I mean, so there's there's tons that you can kind of pick and look and see whether or not this is something you're comfortable using. So you have your kind of goal and task set. You want to then, you know, create your account. Some of these are very limited in terms of their use. GPT is a is an example of that. You know, it you can get some use out of it out out of a free account, but really to leverage it well.

you probably need a paid account. So there's a little bit of investment on that side. After that, you need to take some time to learn the tool. And this is where you get into to prompting and learning how to prompt. And the the best way to do that is to practice, right? Take some time giving it instructions. And I I You know, everybody has a different way of doing this, but there's tons of guides out there and best practices you can find for prompting and generating prompts.

I like to give it really specific instructions. So I I mentioned earlier sometimes AI can hallucinate. It can generate information that doesn't actually exist. One example of that is sometimes it'll it'll create references that aren't real articles. to support what it's trying to give you. And sometimes they'll call that a hallucination. You need to give it really specific and careful prompts. So with some practice, you can start to see how to get it to give you what you want to come out of it.

So let's just use the syllabus example. I'm gonna tell it I want a fifteen week syllabus with one assignment readings from this specific textbook each week and one discussion with instructions and rubrics to accompany it. And it's going to generate that for me based on those constraints. But those are the only constraints I gave it. If I give it more information, more prompt, it will then develop based on those instructions.

So you need to be careful about how you prompt and take some time to practice to give it the right constraints. If I want real references, I need to tell it that. I need to say, Hey, I need you to find me real sources that you can source online from reputable journals that are peer-reviewed and scholarly.

within the last seven to ten years. I have to give it that specific information. It's gonna kinda generate on its own what it feels like is best given the so Yeah, I I I hear people who say they have problems with hallucinations. My first question is, do you have a paid account? Because I don't see that in paid accounts. Also, it's probably the way in which they're asking or getting.

information like you suggested, being specific about having real links to the to the articles and then being able to check on them. I think that's kind of standard practice now to make sure that they really do exist. But I have less problems with that than And it might be because of the expectations that I put in prompts for Chat GPT to respond to stuff. Practice is important.

Yeah you can avoid some of that by feeding it the information you want it to use. Right. So if if you have a set of data or information, you know, I I only want these pulled, let's say, from counseling today, you know, or or you know, you're gonna give them a source. Give GPT a source or your AI a source, like this is the information I want it to pull from. You can also attach files you want it to source.

And tell it, I want you to follow this format. I want you to take this and make it better. So, you know, there there's tons of ways you can use the generative capability of this and and constrain it. So that it's using what you want, how you want. But you have to test and you have to verify. So what you talked about, Marty, is I think really important. You need to verify the information that's there. You can't just take the output that it gives you and assume it's okay. You know, I I worked on a

a course outline for a new textbook that I wanted to use and found out later that it got all the page numbers wrong because it was using an earlier edition because I didn't tell it which edition to use. So it just picked one. You know, so it it's just examples like that. I think when you're thinking about workflow, so once you have your kind of prompts, your output there, you've kind of verified the information, you see what it is. I think your last

step really is to look at refinement. And some of that of course you can do the old fashioned way, right? Get in and make the changes that you want. I think once you're kind of happy with what it's giving you, it gives you an opportunity to have templated prompt. keep track of things that you use regularly. And it also learns what you want. So GPT starts to learn what type of information you're looking for, what type of voice you want back.

So it it'll start to kind of create things from your frame of reference. Most of what I ask it about and I ask it to do is related to education and and counselor education tools, clinical work. So most of the time when I ask a general question now

It gives me that response from that frame of reference. My prompts need to be less specific now because it's kind of learned what it is I'm looking for. The caveat to that is if you're going to ask something outside that frame of reference, you need to give it more context and more information. Yeah. So keeping like a list of how have I been successful the prompts that I've used so that I can adapt them later.

can be a real time saver. Yeah, I I'm gonna give an example here and then, you know, we'll we'll wanna hear anything more about that you have. For this show and for most episodes that I do, not just this show, but also Tech Savvy Professor that we do. I've started to use AI to generate

Show notes. And show notes are sort of the flow of what we're going to talk about. And if I just say, hey, we're going to do a we're going to do a conversation about how Eric uses AI, I get less specific responses back from them. But if I take a previous show note and say, I would like you to produce this information for me, and here is a generative Here's a format that I've used in the past that works.

I'm adding that document into the workflow that I have with ChatGPT and it gives me the introduction to the podcast that we're doing. It gives me the tale to the podcast that we're doing. It knows that I have sections with hyperlinks in that if there are content that's available, that's on the topic we're talking about. It will find that and search it and put it out. If there are questions we have and we have a section of questions for response, it will generate questions.

that are related to the topic. So it knows me, it knows that I'm doing podcasts. And yeah, this time I and and other times I it's always good to throw a model in because it can pull from that information just instead of just saying.

Give me this. If you give it something that if you can give it something that looks like what you want, you get a better output from it. Well, and I think I think I have a good story that helps with this. Uh I'll give you the abbreviated version. My son had a little bit of a dandruff issue. So we bought him head and shoulders and we said, Here, use this, it'll help. Like three days later he came to me and he said, I'm out of shampoo. I was a decent sized bottle and I'm like, but it

Where'd it go? He's like, Well, I've been using it. You said to use it. And he was showering every day because I got basketball. I said, Yeah, but it's only been three days. And he said, I'm like, How are you doing this? You know, I'm thinking he's like filling up his hand with like half the bottle. He's like, well, it says head and shoulders. So I use it. He's using it on his head and shoulders.

instead of just using it as a shampoo because that's what it says. It says head and shoulders. And I'm like, Well, I guess I really didn't explain and he assumed from the bottle that it was to be used on the head and shoulders. AI kind of acts the same way. It's gonna give you what you ask for, but it's open to interpretation. You know, and there's a stand-up comic that talks about these same up same things.

Like if I tell my son to go put the laundry in the dryer, he will. I'd never told him to turn it on. He'll just go put it in the dryer, right? Um So AI kind of acts the same way. It does take some liberties to try and meet your request. It tries to fill the holes in, but sometimes it it doesn't know what you're asking for or what you want it to look like.

So the better you can prompt and give it instructions or you can attach content or point it towards content that you want it to use, the better it's going to be able to understand what you want.

So I just had coffee with a colleague and the colleague told the story of a mutual friend that we have at another institution. who was asked to write a demar departmental report and and it was on the work that the department had done and the faculty in the department had done over the years, something like that. And He chose to use a generative

LLM, probably ChatGPT. Had all he had data documents that he put in. He, you know, I guess you could put in people's Vitas to find out what they've published in the last year or what they've done in the last year. And i I don't know what content he specifically put in, but the end result was he had this marvelous report that of course he checked for accuracies and he handed it into his

supervisor or his dean or whatever. And there was rave reviews on how good this report was until he disclosed that it was written by Generative AI. And then suddenly what was really good turned bad. I mean he he's fine. I mean there's there's no problem with it. But I mean he made the point in some ways that you just praise something that and a process that I use that you would intuitively disagree with.

Because I should have been sitting in front of a computer for eight hours hashing this report together from data when I could do it in thirty minutes and make the adjustments I needed. And that's what I think is silly. I think, you know

Uh the productivity he got back from having a tool like this help him put information together. All I can think of when you were telling me the story was yeah, I've done this and yeah, it's really boring. Uh Like I I can think of about a hundred different things I'd rather be doing, you know, and and this is what AI is good at is collating information and presenting it in ways that are are easily digestible.

You know, so I yeah, I would be really excited that that happened that way. And, you know, I I think the attribution makes sense. Yeah, great. AI did it. But you got what you wanted out of it. So where's the issue? Yeah. My argument was a little bit further than that. We have to generate those reports in academia all the time. The reports go in and nothing comes back.

So it's like if it's a black hole that I'm throwing this report into and it it goes into an alternate universe and I don't get any feedback or it has no direct meaning back to me. From what I did, then I might as well have a robot write it for me. If it has no inherent usefulness, then why would I spend time on it? You know.

I like to think of this in terms of like just the productivity end of it, right? And I teach a lot of career counseling. I'm in teaching a career section now and we talk a lot about like lost productivity and what could this mean for folks who get to spend their time on other tasks. Th there's a lot of stuff that comes out of admissions and I think a lot of the admissions softwares that are the sales force and things like that are starting to integrate AI a little bit more.

So the types of data they can generate that they used to have to track and excel sheets and handwritten memos and things like that goes away. So they can work and focus on generating revenue, getting more students and focusing on the the personal aspects of their job. And I think that's

what some of this allows us to do. Do you have any favorite tools that your kind of go to? Yeah, I mean Chat GPT is still my favorite. I think Perplexity does searching details about people better. It goes through and and has a really kind of presentable format for like social media posts and things like that. So when I wanna see what somebody's up to, it kind of creates a a cooler profile, I think. So I think it does it does depend on what I I want to do.

which one I'll use. But I I think GPT is the one I lean into the most. I use Copilot for some things. I use the companion AI for Zoom quite a bit. It just really depends on the task that I'm getting into. But the kind of all around for me favorite is GPT. Yeah. I you know, there are models that you can choose. I made this statement and people were surprised who aren't don't understand AI it's as much as you and I are into it.

But you know, if I'm gonna generate a picture, I'm maybe not going to Chat GPT. I'm maybe going to Mid Journey, or I'm going so there are certain platforms that are better at doing certain things. And I think you said that with perplexity in terms of for information around people and and I've heard I have friends who love perplexity, but it's kind of the tool that I've honed in on, the relationship that I have right now. is more with uh chat GPT.

And what's exciting are there are some apps and platforms where you can load the model that you want to use and it will run that under that model and you'll get somewhat different responses. I've tried to do that with you know, I do some prompt engineering stuff and I talk about it. when I present and how to do that and I've run the same prompt engineer through different devices and gotten different output or different platforms and gotten different outputs.

From the different models in terms of what one would expect. One would take me down one path. And, you know, the other one would take me down a different path.

Just because of the way the models interpret what the prompts are that you're putting in. So anything else you'd like to share before we go? Uh just some quick kind of heads up on on prompts that you could use. Some of them that I think are really helpful. I like an acting as if I may say, you know, act as if you're creating a new course in, you know, career counseling. You know, what would that look like?

You can have it act as if you can actually have it argue and give dialogue between two opposing sides. So I've actually had a a back and forth conversation about whether to do one thing one way or another way. And rather than have a pro and con, I can read a dialogue. You can ask it to refined grammar. I mentioned earlier, make sure that it's looking for reliable, peer reviewed research research.

You can tell it the voice and tone you want it to use. Sometimes I want professional but casual. You know, I I need to make a request from somebody, but I don't want it to be like authoritative and I can't find a way to write it that doesn't sound mean. So it can help soften voice. Or it can help make things more professional tone. So I use it a lot for just kind of general email things. So prompts are really cool. Yeah. The the mostly the adjustments I make

is you've seen how I've written to you ChatGPT and the con the tone that I use. What you've just given me is way too professional. People will not believe that email came from me. So can you take what you know about the the way I write and adjust the email to make it sound more like it's coming from me? And trying to cloak a little bit, I guess, maybe of the fact that this was AI generated. But, you know, use the use the

the model to that it will what it already knows about you to be able to make uh make better output, particularly in scenarios where you wanna sound like you. You don't want to sound like a professional report you're handing into the dean. Um lots of good ideas. Yeah, and it's been good to talk about it. I it's A AI probably one of my favorite things right now.

Yeah. And and I'm finding all sorts of ways of using it now and integrating it into not just my professional workflow, which is what this podcast is about, but also kind of personal. ways in which I'm using it to come up with ideas and generate stuff. Uh primarily around podcasting and vi uh video casting, which is what my hobby is. So, you know, my next step will be to take the transcription from this and run it through so

I can produce show notes in the in the podcast or vodcast that will give users the step by step of some of the things that we talked about. So the The media also contains that information. in a easy access form and doing that by using AI to get it. You know, you you referred to using the plowed cloud AI as a pin for capturing meeting minutes and such. And that is what I'm gonna try and run on what we've done here to incorporate into that, that into the show notes.

So I really appreciate you doing this and welcome to come back anytime as we as we evolve and do new things with AI. And of course I'll see you on the Tech Sabbe Professor and On circular firing squad. For listeners or viewers who want to watch and subscribe to us, you can find us on YouTube, youtube.com slash at AI Productivity Workflow. That is the channel on YouTube where you can get the video and you can also pick up an audio feed from it.

You can also find the audio feed on Spotify, Apple Podcasts. And any major Apple or any major podcast platform. Everything you need to know about the show can be found at our website, thepodtalk.net. You can email us at the podtalk network at gmail.com. And we've already mentioned other shows that you can find on our website thepodtalk.net, circular firing squad, tech savvy professor. profiles. They're all there on that website for you to link with. Until next Keep expressing. Stay productive.