Subscribe
Share
Share
Embed
David and Gbohunmi discuss the band world and our favorite nerdy topics.
Welcome to Bandwidth Unplugged, where we talk about band stuff and a little bit of nerdy stuff or a lot of nerdy stuff and a little bit of band stuff. We'll see. But we're your host, David.
Gbohunmi:And Gbohunmi
David:And today we have a special guest, Trey. And, Trey is here because, well, he's a good friend of ours, but also you're gonna be talking about AI stuff with us. That's kind of our, our show on the nerdy side today is AI. And Trey, you work for an AI company, don't you?
Trey:That's correct. I work for a company called Abridge. We are a AI company that helps, clinical generate clinical notes, that are based off of a conversation between a doctor and a patient. So you go to the doctor, the the app records you, not only gives you a transcript and a history of your conversation, but also takes that and generates what's called a clinical note. So it's a, it's a framework that all doctors have to use to put your note into your medical record.
Trey:And then also generates, on the other side, generates some patient notes too that you get after you go to the doctor. So
David:Right. So you were telling me that helps the doctor because, the doctors don't have to focus on writing all their notes and then they miss things. They can focus on just the patient and then let the AI and and all the transcript and recording kinda happen in the background, so then they can review it later. So they're not they can just focus on on the patient, which is what we all want. Right?
Trey:That's correct. That's correct. Can do some really kinda fancy things with that. Some of the stuff that we're working on right now is, we have a process of there's a process inside of health care called pre authorization. And so what that means is your health care company, if they're gonna if your doctor's gonna make a referral, for example, for like knee surgery, there are certain things they have to provide to the insurance company as part of that.
Trey:And so because we're in the middle of this conversation, our our app can now give the doctor suggestions based on what they're talking about to let you know, hey, don't forget to ask this or do this because you need it for the prior auth. So
David:Right. Well, Bahomey, what are you doing in, the ban world, like with AI? Like what have you gotten to use it much or are you allowed to? I mean, there has anybody taken a stance on that?
Gbohunmi:So we have a required training for chat GPT in a couple of weeks for us to go over how to use chat GPT as a teacher. It's included in all of the software that we get as educators now, where people have to pay for the full version of ChatGPT if they want the advanced features. We get all that for free now. So the way I'm using it currently, I use it to write longer emails that just require a little bit more information. It's all my original thoughts.
Gbohunmi:I just want it worded in a specific way so that it makes sense to the family so that way there isn't any confusion with it. I also use it for creating schedules for kids. There's sectional schedules specifically for you non band people. Sectionals is a instrument specific kind of class that we do before or after school, where when they were in beginner band, they were in a class full of every single kid's playing the exact same instrument. But then when they get into the older band, it's all of the instruments together, and they rehearse together all day.
Gbohunmi:So we have sectionals to get that instrument specific information to kids even when they're not in beginner band anymore. The kids will fill out forms for all of their conflicts, and then it was my job to take all those conflicts and figure out a schedule that would be the best for every single kid. Now you can't do it so that every single kid could be a sectionals because it's the probability of that is is very low. But with IGBT, it was great because I gave it the parameters of, hey. Here's all the data in this Google Sheet.
Gbohunmi:I need you to create a schedule where, like, specifically, it was like, they all have to be in the morning. They all have to be like, these sections have to be together. And I I even told it, you can't stack a Woodwind sectional on top of another Woodwind sectional. It has to be like a Woodwind sectional and a brass sectional, like, can put them with one another. I gave it the parameters.
Gbohunmi:So I did that, and it figured out most of it. And then I did have to give it some follow-up, like, I need this instrument separated from this one. I need this instrument separated for this one. And then it ended up creating the schedule that we use right now.
Trey:So Well,
David:and and so I used to have my assistant handle that. And then he would have to go through the whole thing, literally chart everything out. He was he'd use paper or Excel or whatever, and he would have to go through and figure out, okay. Well, this sports ball thing is going on at this time, so we have to coordinate this. And most of the flutes are in volleyball and their games are on this day.
David:You had to kind of, you know, bend that in. So that you're saying you go through and it just kinda handles 90%, 95% of the work, I think. And talking, you know, and I think we all use this. We're talking go back to, like, emails and communications. Right?
David:And I I feel like there's something that's just, like, disingenuous about letting AI communicate for you. But I think people that's because people do it wrong in the wrong way. So sometimes you can have, AI I mean, what I do is I will put out the, you know, whole email that I'm that I'm using and I will have it then kind of go through like a really good word check, you know, where a spell check, I mean. Right? So it's going through and then it phrases things differently.
David:That's more clear. So it takes my original intent, phrase things differently, move things around. But I always go through and edit it after that. So it is my original thought. Right?
David:But then sometimes I do notice if somebody's, you know, Trey, have you ever seen this where you can tell when you're getting the they put in like two, maybe one sentence and then there's four paragraphs and you're having to read through all of their AI stuff.
Trey:Yeah. Pretty much. And, you know, you get into some situations where you just have AIs talking to each other, you know. But, no, I mean, I think and I I agree a 100%. I'm the same way.
Trey:And even when it comes to so I'm my title at a bridge. I'm a staff platform engineer. So what I'd work on is a lot of the infrastructure that runs our AI workloads and those kind of things. And, you know, but we still I write some code, you know, on most days. And, you know, the what AI comes out with is is can be good.
Trey:But at the same time, it's kinda too exactly like you're saying with actual text and and language that's coming out. I'm a firm believer in I'm gonna rewrite whatever it gave me, but it's great for like a a sounding board of, hey, how else might I say this? And then, you know, you get some words back and then you can kinda, you know, I I still think you can tell if somebody's actually just throwing you AI slot back or if you're actually, putting some thought based on what it gave what it gave you.
David:So What what do y'all think about this? So I, I started using, you know, ChatGPT and, most of ChatGPT for the last year very heavily. Right? And it's getting to the point now where like, even in my personal life, I mean, I don't Google stuff that much anymore. I just use ChatGPT and say, Hey, feed this to me, give me this recommendation, whatever.
David:And a lot of the sources that it gives me about questions are things that I would have Googled anyway. And I felt like there was a good skill to learn how to Google something in the right way to try to get the thing back. Right? But a lot of the sources it's coming from, a lot of it's from Reddit. A lot of it's from my brother is in the tech space.
David:He was talking about Stack Overflow. So that source has now been stopped because everybody might go to ChatGPT instead of going to that as a source. So have you all seen that where where maybe the sources are old?
Gbohunmi:For me, it what so I use ChatGPT fairly regularly now. My action button on my iPhone actually goes to ChatGPT now. I just changed that. And then I'll use Google if it starts off a sentence with, I think it's this based on blah blah blah blah blah. When it uses that prompt of I think it's like, oh, okay.
Gbohunmi:It doesn't even know if this is the right answer. Oh, I I have one
David:for you. Sorry. I had it the other day say, my gut feeling is or like
Gbohunmi:Exactly. So stuff like that. That's why I'm like, okay. This is no. If it's like a very niche specific question, I usually go to Google, and then I try to scroll scroll past all the Gemini stuff that happens at the very top first to see what I actually need to find.
Gbohunmi:But for, like, very simple things that I search on ChatGPT, I'm like, I wanna eat blah blah blah blah blah today, and I need to stay under blah blah blah calories. What can I eat at this restaurant that'll get me there? And that one's pretty accurate because it'll just scan the Internet of, like, what the calorie counts are that are like, they don't change. They're the same every single time, allegedly. So it's stuff like that.
Gbohunmi:So I have had instances where it'll say, like, my gut feeling or I think, and I'm like, okay, we're not gonna use this right now. We're gonna go to a different source. I
David:use, one of the use cases, I guess, health wise. So here's the really nerdy stuff. So I set up a, Bahomey, what are these things called? The little little things at the bottom. Right?
David:Those those The widgets? Yeah. The widgets? Pumps, widgets, we call those?
Gbohunmi:Shortcuts or whatever. Yeah.
David:Okay. Fine. So I
Trey:set up one
David:of those and it went directly, into a shortcut and all I did when it with it was started to input data. And that got added to an, a note, an Apple note. And at the end of the day, I just had all an Apple note with all this stuff that's timestamped in it. Right? So if I, so what I would do health wise, I would say like, protein bar, you know, or I would say apple or feeling tired or had Chick fil A sandwich, blah, blah, blah.
David:And I would just say something really quick. So it's really easy for me to do that. Then at the end of the day, I would take that note, copy everything that was timestamped and throw it in the chat GBT and a GPT I made for my like health tracker. And it was like, oh, okay. Hey, you were feeling a little low about this time.
David:That's because, of what you ate here and what you ate here. So this is really cool, like personal trainer style thing. Started to get my preferences and started to know like, oh, the protein bars that I like and everything. So it was a really cool, like, just, you know, got the data in there. And I started to, you know, okay, well, maybe I'll some cottage cheese right here.
David:It'll help me kind get through this part. I don't know. It really kind of helped, do some stuff. But what are, like, Trey, what is what is a personal, you know, interesting little, chat GBT or AI thing you use?
Trey:Yeah. So I'll use the most the most recent example I had is I was having some we're having some Internet issues at the house, and and it's been going on for months. And so I decided to take my if you ever logged into your cable modem, there's a bunch of logs in there and numbers that make no sense to anybody that's not a cable technician. So I took the logs and then the other metrics and stats I had for my connection, put them into I use I'm a Gemini user. Put them in a Gemini.
Trey:And not only did it give me kind of response about what this means, but then also said, if you're talking to tech support, here's the exact message to give them. And so I did that kind of once through, got somebody to come out the house. The techni the the actual technician asked me, he's like, how did you know what to say? And I like, well, I'll show you, you know? And he's like, wow, this is pretty, pretty good.
Trey:It's kind of scary. He's like, as soon as there's like, it can physically come out to the house, what do I have a job anymore? But but even, I mean, even took it to the point of I had somebody on tech support that said, oh, you need to replace your modem. And so I told Gemini, hey, the technician's telling me or my, you know, customer support person is telling me this. He said, no, they're wrong.
Trey:Tell them this instead. And so kind of again, use them. And I would imagine on the other side, I could have been talking to an AI at that point. But, so, yeah. But it it it was a pretty cool use versus use case.
Trey:I was gonna kind of roll back just a second. You you were talking about, you know, the the point in time of which these models are trained. So in my role, what I do a lot of is work with cloud infrastructure, and and these things move very quickly and can change very quickly. But a lot of these models, like, I know Gemini and Claude, I think, like, January 2025 is kind of the cutoff. Anything that happens has happened after that date.
Trey:It doesn't have a very good, I mean, it can't. Right? There it's they have to cut off these models to do the training. They have to go through and refine the models and then clean them up and to to release them. So in my job, the the term of hallucination comes up a lot, where not only is it gonna give me it may give me code back, but this code doesn't actually exist.
Trey:It doesn't do anything. I had it give me the names of, like, cloud resources that don't, you know, might, you know, Google or, you know, AWS don't actually have. But it's it's a definite issue, as kind of and kind of to your point about Stack Overflow, you know, is that the knowledge that's out there in the Internet gonna slowly decline because it's just a bunch of AI now. And it only has this point of time before it, which it knows what to do.
Gbohunmi:So that leads me to add two questions for you. The first one being for your job, in AI, have you seen instances where AI is replacing workers at your job to do more things for AI?
Trey:Yeah, I wouldn't say replace. I would say the the big well, I would say the biggest issue potentially for for that is for like junior level developers, you know, folks who are just out of school, because that's the kind of work that there there are some cleanup tasks that typically you might give to a junior level person. Now, you know, Gemini Chat GPT, Claude, you know, they can help do that for you. But, you know, I think at at the same time, when you see some of the stuff that comes out of it that's so incorrect, it makes you feel a little bit better about being a, you know, someone who's been in the industry a little bit longer to go. Well, I still I'm still gonna have a job for a while.
Trey:But but, yeah, I mean, I think it there's a that's been kind of the popular thing out there is, like, is our people's jobs going to go away? I think in large in part, no. Some of it's just going to be, it's a different tool to apply to your work versus just saying this is going to completely replace somebody.
David:It seems like a really good calculator. Right? Okay. So if you, I mean, if you know how to use it and you know how to use that calculator, then you are going to be better at what it is. And it's just another skill set to have.
David:And one worry though, that I have, especially in the education world and in the young workforce world, so that junior staffer or, you know, that high school kid phones it in, uses AI too much as a crutch, then they're not learning the hard work. You know what I Because, I mean, I remember what teaching, I just told my kids, hey, practicing is hard. This is the only thing that you do in all of your school where you have to do it over and over and fail and fail and fail. You have to do that. Some of the, in other parts of the school, don't have to do that, like maybe math problems or something.
David:But, that's one of the worries that I have, as, you know, like I said, people going through things, is they're not doing the hard stuff, the meaningless stuff that, you know, ends up making them better and they realize the mistakes. They realize those little things going through.
Trey:Yeah. I think, I think there's, there's been a lot of studies that there's a, I know there's a big study that came out a couple months ago talking about cognitive decline of, of, of just, I mean, that uses AI frequently that you're, you're not thinking as much because you're just, you're just feeding prompts into something and it spits you out an answer and you just don't even think about what you're what it's giving you. You know, we've you know, I I know not sure what where this would fit in like music education, but for in the the software engineering world, we have things called a I mean, a lot of these services have things called agents, right? Where I can, I'm no longer writing code. I'm interacting with an agent that's doing something on my behalf.
Trey:And that's great until it does something it's not supposed to. Like, let's say it deletes a service out of one of our, you know, our one of our Kubernetes clusters. Right? Like if it if it does something that it's not supposed to, you know, and some people are just like, well, what happened? I don't know what happened.
Trey:And it's like, well, because the agent is not giving you enough feedback. And in some of these models will give you like the thinking that goes on or what about what it's doing. But, you know, it is kind of one of those it's dangerous that it can go off and it do its own thing. And, you know, like we've seen we have a chat, we have a Slack channel at work and people will talk, you know, we have a we're our CTO is very much a believer in dog fooding AI products, right? So we are an AI product.
Trey:We should be using AI products all the time. And it's it's funny to see kind of see some of the responses. Like if you get in a code base of some point, the AI just decides I'm going delete everything because I've I've gotten I've gotten myself in such a corner. That's the right answer is just to delete everything. It's like, well, that's that's what the that's what the language model thinks it wants to do.
Trey:But at the same time, you know, not really what
David:In intend for a movie that we all know.
Trey:That's yeah. Yikes.
Gbohunmi:So my other question was so you mentioned Claude and Gemini and Jai JibT. Those are the big three that we've talked about. Mhmm. From your knowledge, can you tell us the differences between the three of them, and then why you exclusively use Gemini versus the other two.
Trey:Yeah. I think differences, they're all very, very similar, for, you know, I think certain, you know, so I think when you look at it as a whole, so like, let's say we look at at Gemini. Right? So Gemini has got, you know, you've got Gemini Pro two dot five was like the most latest one they had. But several weeks ago, they released, it's like two dot five flash that has an image generator in it.
Trey:And internally in Google, the code name was you'll see Nano Banana out on the Internet. But it was all about image generation. And so, you know, we kind of all think of it as that that's like the the umbrella product is Gemini. But, you know, that's one thing where it's kind of left a little bit ahead of the other two. Whereas like Claude, for a long time, like Claude Sonnet is like a coding model.
Trey:It's really good at doing software engineering. And so and then, you know, ChatGPT has always been like this general purpose. It can kinda do everything really well. And so, you know, for me, why do I use Gemini? I I'm a Google Cloud fanboy.
Trey:Like, I've born and raised in Google Cloud. That's kind of where all my experience is. And so, that's part of my my reason. But also, it fits in my workflow. So, you know, they've got a CLI tool that I use inside of a terminal that that I can work with.
Trey:And so that that's partially there. But I will say I use Claude a lot if I want it to generate. So we use a tool called Notion. I'm sure some of y'all probably probably use it before. Claude does the best job of generating Notion pages, right?
Trey:So I can give it a whole bunch of things that I'm working on say, hey, can you help generate a Notion page so I can go copy into Notion? Gemini does an okay job. Chat Chat GPT is okay. But Claude does a much better job of being able to directly copy and paste. And so, you know, I use Gemini the most because because it, you know, again, Google.
Trey:But at the same time, you know, each one of them has different use cases, and kind of we and it's I'm lucky enough to work somewhere where I have access to all of them. But so, yeah, it's just kind of, again, kind of to David's point, it's the tool belt and like what tool do I want to pick off the tool belt to do the job?
David:I think, one thing specifically that I've liked using is setting up my own GPT so that it's all already set, that all of my preferences for that specific purpose and use case is already set up. And, I have, I have two cases, they actually both have to do with kind of car things, I guess. But, one specifically I've set up is my, mechanic for my car. I have two Lexus cars, like, I don't know. So I have a Lexus mechanic, you know, thing.
David:And I programmed in there, you know, Hey, I uploaded every document I could find, service manuals, everything. I said, you know, here are my preferences on things, find this stuff. And then one, one tip that you should always give for your overall, GPT settings, Chad GPT settings, or whoever it is, is I say, show me the math. Whenever you give me math, whenever I want you to show it to me, give me the formulas. So I can kind of double check it.
David:But there's so many times when I'll do something and, you know, I will go in and say, like, you know, I kind of know the math on it. I, but I, and I could figure out the formulas or I could do that, but I don't want to. So I just kind of throw it in there. But, one really cool case where it saved my tail was, I was driving to, Colorado on a camping trip with my brother and I picked him up at the Albuquerque Airport. When I did that, I pulled into the airport and you know how airports are busy, you know, everything.
David:When I pulled in, all of a sudden my dash is lighting up with all these warning lights and I'm like, Oh my gosh, what's going on? So I quickly opened up ChatGPT, said, hey, this light, this light, this light's on, no other symptoms. This is happening, blah, blah, blah. Gave it all the symptoms. And it said, hey, here's like the easiest versus the worst case.
David:And it says worst case, it could be this horrible engine thing and failure, whatever. And I mean, if that happened, you know, what do I do without ChatGPT in that situation? I gotta go to mechanic. I gotta hope it works. I'm on a trip.
David:I'm camping, so I'm gonna be out in the middle of nowhere. My car could fail. That's a bad idea. You know what the easiest thing was? It said, hey.
David:If it's this and this, then it's probably the ECU and the ECO thing or whatever. Why don't you check your check your gas cap? So I walked around the car, closed the gas. That was it. Turned the car back on.
David:No issue. That would have either cost me a lot of time or ruined my day, had a whole another thing. Some of the guy would have ripped me off in another town would have been awful, but that's just a great use case. And the other use case right now, I'm working on is, I have a little project at home and I'm going through a lot of electrical stuff and a lot of hardware stuff or whatever. And I programmed in all of these things.
David:So I'm just kind of feeding in data. And then here I am at Home Depot and I'm like, Hey, what about this? Oh, you should try these. Oh, okay. That's on this aisle.
David:Oh, Lowe's has these. So it's just like this little assistant that you can kind of use on projects. Then when you go back to it, you don't have to, you know, it's not like another Google prompt where you're trying to find where the thing is. It already knows. You're like, oh, hey, remember like two days ago I said this?
David:You know, here you go. I don't know.
Gbohunmi:Yeah. I think that's the I think that's the part of it that people don't necessarily understand that they can do that. Because, I mean, it's a virtual assistant. Just like any other assistant, like, hire a new assistant, you have to tell them, hey. These are my preferences on things, and this is how I navigate my life.
Gbohunmi:So I think it's really smart that you're configuring your GPT to be specific for specific use cases because it's you're doing the work. You're telling it all of the things that you want it to do versus it just being a blanket. Hey. Tell me this thing without knowing who I am at all and having no context on anything. Exactly.
Trey:So The
David:context is kinda weird. It'll pull in some weird details. And so, like, I'm working on something, and then it's all of a sudden, like, it'll pull in like a work detail. And I'm like, what where did that come from? You know?
David:So it kinda it does limit it maybe and and put bumpers on it. But
Gbohunmi:Right. So my question is with that, when you create GBT models, are they separate, or do they learn from one another? Does it learn, like, your like, what responses you put into it and how you word things specifically so that it could understand you faster when you create a different GPT model? Because I don't know that.
David:So so as I understand it, and and maybe Trey, you can correct me on this. So there's overall GPT preferences that you can enter in, in your like main settings. And so I have a couple of those in there and like specifically there was one that, early on I was like, I was talking to my son or whatever using the, you know, the Christmas one. That was really cool. It came out with Santa Claus talking to him.
David:And so that was really funny and whatever. But I was like, Hey, don't remember my kids' names and you know, program that in. And then, but you can also add memories. It'll do that and it'll log what all your memories are. So you can go through and delete stuff or you can add stuff and say, hey, I want you to remember that I work for this company.
David:This is what I do. Okay. I'm a Clarinet person, whatever. Then the individual, I think the individual conversations don't necessarily remember each other's conversations. Right?
David:But if you do a GPT, your own GPT, then everything within that will kind of remember that same conversation. So that's why I kinda have them sectioned out.
Trey:Yeah. And to a and to a point, right, so there's a the concept in in these LLMs is called model context. So you'll see, like, there's a big thing about tokens. Tokens are actually think of them as like letters. They're part of the the language.
Trey:But but as you're like, you you get these models that, like, they talk about a million context or a million tokens that they can have a million points of information in which they can start to do this. And so you do kind of and as you some of these tools will will show you how many tokens you have left in its memory. And the closer you get to that number, that's when things start to fall apart typically. But kinda to to David's point, like, what I would what I would do in Jim and I call them gems, but it's basically like a GPT. And so I have I have a a gem that actually generates a model or a prompt for the other the other gem that I'm gonna use.
Trey:And so for an example, like I've got one that I use as part of my workflow for we, you know, I work for a very fast growing startup. We do a lot of interviews. And so I have a I have a prompt where I can give some of my notes and thoughts, and it makes it a consents format, you know, the consistent format that I can do to submit to when we do our interviews. But what I do is I have a, I have a, let's call it a GPT, right? I have a GPT that will create the prompt that will I will then use as another GPT to generate those.
Trey:And so as I've had more feedback, I can help refine that prompt even more and more. And I think that's kinda that's a pretty good hot, like, power user tip is like, use an LLM to help generate a a better prompt, because that prompt is really what gets you your better, you know, example. Right? I can just say, hey, I'm gonna ask you a question about this tech, you know, I'm gonna I wanna know how do I write a better Python script. But I can also give it the context to tell it, hey, you are an experienced Python developer.
Trey:These are the things that you need to remember as you're going through this. And then ask your question and your output gets tremendously better. Cause you're just, you're helping kind of guide the guide the LLM along about what it, where it needs to start to kind of start thinking about it.
David:One thing, if I was still in the band world, what I would be doing would be taking everyone, like, set up a JET GPT, mean, a specific one, and take every PDF of every method book that I have and everything and feed it into that one thing. It has so much data to kind of go off of. And then I can go in there and be like, Hey, my clarinet classes in, you know, kind of this point and this point, you know, out of all the data that I have, because sometimes I would forget, you know, oh yeah, I have that worksheet that would be really good at this thing or that, that, you know, part of this exercise or whatever. And then, I mean, heck, I wish there was some books that I had that I could just scan in and I would just, you know, get a PDF of those and just try to get as much context as I can in there, upload as many things and you can be like, okay, so, I I don't know if there's enough data in the band world yet, but if you give it more data, then say, hey, I'm having this problem with my trumpet class.
David:What are some of the suggestions? And then I always say, and give me the source of that, so that I can look it up myself if I want to or get the context.
Gbohunmi:Yeah. It's it's getting there. It's not there yet. Someone on Facebook put in a prompt into chat. GPT, give me a rehearsal plan that Eddie Green would write.
David:Right.
Gbohunmi:And it had, like, all of this stuff that you would do plus, like, why you're doing specific things for all this stuff. And it was it was the craziest thing I'd ever seen because it was so it was so detailed with the context of, like, all of the books online that talk about what he did and all the articles and stuff. But I think
David:the danger there I'm sorry to interrupt you, but the danger there that I found is that so that might be the source might not be from his full book and all the, because there's some copyright issues. So what happens is you then get summaries of that book or chapter summaries or whatever, or you'll get like a grad student's paper that interpreted that source. So sometimes it's not exactly that. And I'm not saying in that specific case, that was a busy situation, but you do have to worry about that is like, where is it actually getting all this information from? But
Gbohunmi:Oh, absolutely. But like the blanket, like production of it and like the level of detail that it gave makes you believe, oh, okay. So this is actually real. This is awesome. What would be very fascinating to see in the next decade or so is you being able to type in a prompt saying, hey.
Gbohunmi:My trumpets are bad at x y z. Can you find me a worksheet slash fundamental sheet that can help them out with this? And it creates that context, and it finds something. Now I think if it were to do something like that, it would have to preface, hey. This is one recommendation of many others, and it might not produce the results that you need.
Gbohunmi:But here it is. Because a lot of what we do is subjective in the music world where, hey, you think that this exercise might make it better for you, but it could make it worse. And it depends on how you use it. And that's another thing that I would probably have to say is, like, use this exercise this way doing these things in order to get the result that you need. But that would need like mountains and mountains and mountains of data.
Gbohunmi:And that means that people need to write more things about these exercises and workbooks on the internet. So that way the info and it has to be accurate information. So that way, these LLMs can grab the information correctly. Because like Trey was saying, I mean, it's becoming more and more of, okay, A lot of content on the Internet is just generated by AI now. And now AI is pulling from AI.
Gbohunmi:So what is what is real? What what is fake? What what what are we doing? So
David:Well, and in the music world, there's, five ways to do the same thing. And, for instance, each student, I that's why I love to I always go back to clarinet. Clarinet was a bag of tricks kind of class. And so, student A needed trick one and two, and student B needed trick three and four for the exact same thing to get the same result. Of course, that can happen over classes and what exercises and what I think is maybe different than what you think.
David:And in the end, we kind of get to the same result. We have maybe different styles, but, music is a, is a really good example of, like you said, this subjectiveness, but but also there's many ways to skin a cat, you know.
Trey:So where here's where it'd be interesting. And this is coming from someone that hasn't really been in the band world for a while. I was, you know, spent plenty of time there. But, so, like, thinking about the large language models, right, we know what those do. But other types of AI models that potentially could do, like, let's call it like a virtual clinician.
Trey:Right? So things like, you know, you know, pitch accuracy, like, you know, rhythmic precision, like those kind of things, dynamics and articulation, things that actually you could probably put in the model to spit out. Right? So you could have one side that could do that, and then that would allow, you know, you to kind of the the subjective part of music. Right?
Trey:How is the interpretation or whatever, would be interesting. Like, that'd be a very kind of interesting use of of AI to say, hey, I'm gonna feed my recording of my band. I'm going to contest. Right? I'm gonna go feed my recording in here.
Trey:Where else do I need to improve? You know, that maybe because, know, I think I would assume that as your band directors, as you go through these pieces to, you know, heading into contest. Right? You've heard it a million times. And so, like, why do you go to clinics and do other stuff and have clinicians come in is to do that.
Trey:So it'd be cool to have a virtual clinician that can come in and say, hey, I'm gonna give you some objective feedback to what you might be doing. So I don't know.
David:Or if, like to go to another phrase, like machine learning style, like Yeah. If you uploaded 10 or 20 recordings and say, these recordings are, have been chosen subjectively as more desirable than these recordings over this turn. Now, here's my recording of the same thing, analyze that. And I I'd be curious to what something like that would spit out and say, you know, what an AI would think, you know, this is some, there's some learning that we don't about ourselves that we don't know Yeah. Why we like this or why we like that.
David:And, of course, you know, as musicians, we do. But, I I think that would be an interesting way to go as well. Yeah.
Gbohunmi:Yeah. I feel like that gets hard because, like we talked about, so many things are subjective. And there are only so many objective things in music of like, oh, the dynamic is supposed to be forte here. It didn't sound like forte in this spot. Maybe you should do something differently.
Gbohunmi:But then again, there's the other context of composers don't necessarily write the dynamics in the way that it needs to be for whenever ensemble is playing that piece. Because as you notice on the score, it has like, okay. They wanted a clarinet a trumpet to play this instead of, like, maybe, like, the four or five that I have in my ensemble. So Right. It's gonna sound a little bit different.
Gbohunmi:So there are so many more variables that need to be factored into it, but it would be super interesting to just see what it would come up with and what it would say.
David:There's a grad student out there somewhere that's like, ah, great. That's what I'm gonna do my thesis on. Thanks, guys.
Gbohunmi:Honest to goodness, I'm a man. Wild. Yeah.
David:Well, let's take a break for a second here then. Where do you think we want to go here? Do y'all have anything glaring that you want you know, piece into this or, we wanna go to picks of the week? Or
Gbohunmi:Let's go to picks. I thought that was a really good Yep. Just dynamic conversation. I thought
Trey:I loved it.
Gbohunmi:I thought it was good.
David:Tight half hour. That was pretty good. Yeah.
Gbohunmi:That's good.
David:All right. So let's go to our picks of the week. Let's start with our guests. Trey, what is your pick of the week?
Trey:Yeah. So I just got a new three d printer at the house, after some, after a few shipping problems, but other than that, I got, I got one that that came in and worked and wasn't damaged. But that was that's a different story for a different day. But, yeah, it's it's really cool. So I I'm upgrading from a printer that was probably, like, eight years old technology wise to one that's kind of brand new state of the art, and, it's it's fantastic.
Trey:It's been I got it on a let's see. I got it set up on a Friday, and it's been printing all weekend. And it's just like, what else can we do? And, it's pretty great. It's the it's the bamboo h two s.
Trey:Right? Correct me if I'm wrong, David. That's the right model one. Okay. Yeah.
Trey:But it's it's pretty cool. The the the the fact that it can be automatic is is pretty awesome. You know, I haven't shown my wife how to print yet, but I'm sure once I do, I'm gonna be spending a lot of money on filament. But it's it's it's pretty slick. I mean, it's like even the fact that I can print so, you know, you get these rolls of filament.
Trey:But one of you know, when you order directly from the company, they'll sell you a refill. And so but you can print your own spools. Like the fact that I can go print parts that are gonna go on this printer, just kind of create your own thing and then, you know, just yeah. I'm excited to start around and play with it. It's it's much larger so I can, you know, you'll see me here with like a Master Chief helmet from Halo at some point or whatever.
Trey:But, you know, that kind of stuff. But, definitely a great pick if you're in in the mood for a three d printer and, yeah, go for it.
David:It's the big one. I have one of those too, the previous generation and I suggested it. And it's, what I love about it is that it's just so it's got the kitteness out of it. It's not like this kit yet to tweak and all that stuff is, like you said, like my wife is not interested at all in three d printing, but now I've actually printed more recently for her than for me. But it's just something you can just pull up real quick, click go and it works.
David:And like my son is really into it. Pretty soon, you know, he'll be able to print whatever he wants. I know like my, brother and his kids, they, they print a lot of stuff all the time. And I mean, I think that's where like that kind of thing is going where like band halls will eventually have three d printers as standard. Because you can certainly print parts, for that.
David:Especially, I mean, in like twenty, thirty minutes, you can have something and that's not like future tech. I mean, you could do that now if you wanted to. And, yeah, we're not talking about like, you know, a buffet clarinet base, but hey, if you're trying to figure something out and you need this screw or this little thing or whatever, instead of having to wait a week for your instrument supply guy to get in, it's entirely possible for that to happen.
Trey:So, and it's got a great, a great AI functionality where there's obviously a camera in it that can monitor how your print's going. And if AI detects that there's a problem, they can they can end the print for you. You don't have to babysit, you know, my old one, you had to sit there and babysit it, which is
David:pretty cool. That rarely happens though. I mean, I've had mine for two years and I've had maybe four prints go wrong out of a thousand.
Trey:So David, here's one to print. I really want to print a saxophone reed.
David:Yeah. There we go.
Trey:Yeah. Could you print, could you print a saxophone reed?
David:Yeah. Yeah. I, I had a kid that printed, his dad had one and printed one, and it it worked pretty well. And I think it would 100% be, if you got the right print and, you know, tweaked it whatever way, you could absolutely do that. Not a clarinet read.
David:These clarinets are a little bit more sophisticated. Nobody cares what the saxes actually sound like. That's I'm just kidding. True.
Gbohunmi:But if
David:Johnny's a sax player, know what this Yeah. Well, Bohoma, your pick of the week.
Gbohunmi:Well, new Apple products dropped recently, and I got myself an Apple Watch Series 11. I'm coming from the series five. So it was a massive upgrade for me. The screen is larger. I had a 44 millimeter before, but now I have a 46 millimeter.
Gbohunmi:And it's like edge to edge screen now, which is crazy. Just like how much more like, I can just easily read everything on the watch. It's been off the charger since 10:30 last night because I sleep with my watch. It's my alarm. And we are currently at 65% battery still, which is really, really good.
Gbohunmi:I had to charge my old watch twice a day just to, like, get through the entire day, which was super annoying.
David:Would you ever have, like, the charger underneath your wrist while you're trying to wear it or something?
Gbohunmi:No. I mean, it won't sense things if you do it like that. So that's yeah. The the, like, the double tap functionality is really cool and, like, the Flickrists functionality to dismiss notifications.
David:What is your double tap? What what do you do with that?
Gbohunmi:So if I, like, if I could prompt the pop up, if I just double click like this, maybe it'll work. Maybe it won't. Yeah.
David:Maybe because the way you're There we go. See? Oh, okay. Cool. You
Gbohunmi:can go through my smart step
David:like which
Gbohunmi:is super awesome. And it doesn't matter what finger you use. You can use any of your fingers. It's super awesome. So and then you can just flick to, like, go back to the home.
Gbohunmi:So it's pretty great. Super awesome. But, yeah, I was long overdue for a watch. My watch died at the gym when I was halfway through a workout. So mad.
Gbohunmi:Not not like it's about the numbers or anything, but, know, it's a great motivator. And I was like, okay. And I that that watch had been through its paces. It had been through half marathons, five k's, 10 k's, all these different things. So it was it was just time to get a new one.
Gbohunmi:So this came out on Friday, and I got it on Friday. So it's nice. I like it a lot.
David:Nice. Cool. Well, I'm gonna suggest something that I think we've all kind of maybe dabbled with, but I'm gonna suggest an app today and it's a Mac app. And it's Raycast, r a y c a s t. Okay.
David:And so, if you've ever used Spotlight on your computer, and I mean, honestly, just using like, it's command space, you know, it pops up and you type like the app in there or whatever. And apparently in iOS 26, that got better and they're doing different things. And some people think Raycast is gonna be, you know, gone. But it's really not. Raycast is far superior and it's great.
David:I actually replaced my spotlight with this and I use it all the time. It's an app launcher. Yes. But it's even stuff to say, I mean, like math problems. I mean, just putting in like, okay, you know, like 42 times 11 or something like that.
David:Sure. But you can add in, you know, percentages. Like I have to do a percent a lot. Right? And so I could say like, 30, or, you know, like a dollar amount plus 35%.
David:Well, instead of having to do the actual equation, it comes out. Or, 11AM CSD in Eastern time, yada yada. Things like that that are just kind of natural language it'll adapt to. But then the real trick is when you start getting into their, apps, like within the little extensions or whatnot. So there's an extension for just almost anything.
David:By the way, AI extensions. I don't typically use those because I like to just stay in the interface, but you can do that and you can have your ChatGPT, API LinkedIn. So, you know, it'll be using your version of ChatGPT, whatever you wanted to do. But use it for that. Use it.
David:There's also a thing called snippets, which is something that, you know, other, you know, companies have done in different ways, but I think it's really well done where you can, you know, program in, you know, groups of texts or links or whatever you want. It'll pull up those snippets. But the real trick is when you start using it a lot, then it's the keyboard shortcut. So it's all keyboard stuff. So instead of going keyboard, mouse, keyboard, mouse, if you're just doing something really quick, it's really great.
David:I have a lot of my, files that I use all the time just already programmed in there. So there's a color file that you use all the time and I just hit, command space and I type in colors, enter, and then that file pops up. And it's just so fast to just go quickly through those, set that up in my own preferences. I highly suggest if you ever do check into it, use the tutorial, to kind of go through and see what's what's possible and you can start getting in your head what you can do. And there's also a, iOS app that it does sync.
David:I think you have to have a certain level, which is kind of a bummer, but it does sync. So I use those snippets when I, that, that can sync, you know, across my device. And so that's really easy to use. It's all stuff that in general is a Mac and iOS and they're, they're kind of in the system or something that you can Google whatnot, but it's so easy to do. And if you just got a really quick, you start using it more and more and then it pops up.
David:And it's just at your fingertips, literally. So it's a speed thing rather than a convenience thing. So it's definitely something to check out. So, yeah. And, Trey, I think you have used that before?
Trey:No, I haven't used it. I'm actually just looked it up while you're talking about it. This is pretty cool. Do you know if it can, so like, can you provide your own, it looks like, you know, there is an AI feature where it can have models that are part of that, that workflow. Do you know if you can provide your own like models?
Trey:Do you have to have like, does it, do you have to purchase their, like, do I have to use ChatGPT through their service or can I, you know, own ChatGPT? That's a big one. Yeah.
David:I think you can set your own, but you know, Hey, our listeners go in and check that out. I didn't check it out that often that, that well, because I, like I said, I kind of use my own separate thing. That's pretty cool.
Gbohunmi:Imagine so because the new Spotlight and Mac OS that you can, like, program shortcuts for it to, like, do specific thing for specific use cases. And that's, like, very, very down down with, like, what Apple and challenges can provide. So I have a feeling that rate casting probably like you can provide your own stuff in it and make it happen.
Trey:Because I would assume, know, Bahamie, you mentioned at the top of this about ChatGPT for Is that like, you know, because we have the same kind of thing. Like most of our accounts we use have to be enterprise accounts because our data needs to stay our data. Is that the same for edu? I would imagine, you know, you're probably highly discouraged from using just off the shelf. Yeah.
Gbohunmi:Yeah. Yeah. It's in, like, we have this hub called ClassLink where all of our apps for anything we use in a district is linked to our district email. So we can't just, like I mean, I could probably load it on my phone right now, but, like, it would be monitored by my school district with anything that I search out. So
David:Right. Well, that's
Gbohunmi:smart to clear.
David:That's nice because then you can load in your data. I remember, I had a family member that, was in the healthcare industry and could not at all use that because I guess they didn't have enterprise, whatever, set up where they could kind of box things and silo things off. And so she would come up with, and I did this when I first started. I wanted to come up with something and so I came up with their names and I used an Excel formula to scramble their names and I kind of knew the code to unscramble it. And so I kind of had my own little mini encryption and then I put all their stuff in there and all their data and it was like an end of the year, you know, whatever.
David:And I used ChatGPT to do that. And then I used the opposite formula to undo it. So I wouldn't put any private names out there, you know? But now it's nice that you can kind of do that and silo that off into your own set. And so I'd really be interested to see at the end of the year, if you could do some data on that, with your, end of the year auditions.
Gbohunmi:Man. I agree.
David:Chad, cheapie teeth as your first chair.
Gbohunmi:I don't want to think about end of year auditions.
David:Well, that's a long way off. Well, that's our show today. Thank you, Trey, for joining us. We really appreciate it. We'll have to have you on again soon.
Trey:Yeah. Thanks a lot. Enjoy talking.
David:Yeah. Yeah. All right. Another episode of bandwidth unplugged.