323. The State Of AI In Audio With Joe Allan Muharsky

In Episode 323, Sean and Andy huddle up with Joe Allan Muharsky for a wide-ranging conversation about the state of AI in the audio field — where it is and where it’s going, what it’s good at and what it’s not. This episode is sponsored by Allen & Heath and RCF.

Joe is a software engineer, AI practitioner, and generalist with 30 years of technical experience spanning Microsoft, Google, and a suite of small-to-medium sized companies building line-of-business applications. He’s also the creator of the TODD Framework — a proprietary meta-framework for AI-assisted development built on the philosophy “agents in a system that they can evolve to accomplish their goals” — and has contributed to academic content for a textbook on AI adoption in the events industry.

Alongside his technical work, Joe serves on the board of directors of the Washington Festivals & Events Association (WFEA) and is a part owner and partner at Teamatorium, where he helped scale a six-person team to more than 20. Music runs through everything he does: from audio engineering at venues and festivals to producing Band Together, a site at bandtogether.com for hyper-inclusive collaborative music.

Episode Links:
Joe Allan Muharsky On LinkedIn
iCitadel
Band Together
Joe Allen Muharsky Website
Episode 323 Transcript

Connect with the community on the Signal To Noise Facebook Group and Discord Server. Both are spaces for listeners to create to generate conversations around the people and topics covered in the podcast — we want your questions and comments!

Also please check out and support The Roadie Clinic, Their mission is simple. “We exist to empower & heal roadies and their families by providing resources & services tailored to the struggles of the touring lifestyle.”

The Signal To Noise Podcast on ProSoundWeb is co-hosted by pro audio veterans Andy Leviss and Sean Walker.

Want to be a part of the show? If you have a quick tip to share, or a question for the hosts, past or future guests, or listeners at home, we’d love to include it in a future episode. You can send it to us one of two ways:

1) If you want to send it in as text and have us read it, or record your own short audio file, send it to signal2noise@prosoundweb.com with the subject “Tips” or “Questions”

2) If you want a quick easy way to do a short (90s or less) audio recording, go to https://www.speakpipe.com/S2N and leave us a voicemail there.

Transcript

Signal To Noise, Episode 323: The State Of AI In Audio With Joe Allan Muharsky

Note: This is an automatically generated transcript, so there might be mistakes--if you have any notes or feedback on it, please send them to us at signal2noise@prosoundweb.com so we can improve the transcripts for those who use them!

Voiceover: You’re listening to Signal to Noise, part of the ProSoundWeb podcast network, proudly brought to you this week by the following sponsors:

Allen & Heath, whose new dLive RackUltra FX upgrade levels up your console with 8 next-generation FX racks – putting powerful tools like vocal tuning, harmonizing, and amp simulation right at your fingertips. Learn more at allen-heath.com

RCF and TT+ AUDIO.... Delivering premium audio solutions designed for tour sound and music professionals for over 75 years.  Visit RCF at RCF-USA.com for the latest news and product information.

Music: “Break Free” by Mike Green

[00:00:58] Andy Leviss: Hey, welcome to another episode of Signal to Noise. I'm your host, Andy Leviss. With me, as always, the artificial to my intelligence, Mr. Sean Walker.

[00:01:09] Sean Walker: Every time with those things. You get me every

[00:01:11] Andy Leviss: Uh, you know,

[00:01:13] Sean Walker: You got too much time in your heads, bud. How do you come up with a new one? Every, for every episode.

[00:01:16] Andy Leviss: I don't, that's why you often see the sheer look of panic in my face right before we start.

[00:01:23] Sean Walker: fair enough.

[00:01:24] Andy Leviss: That one, that one I had, that one I thought about like an hour ago, but like half the time I said, that's why I just suddenly go blank and I'm like, uh,

[00:01:32] Sean Walker: Fair enough. Fair enough. What's up with you this week?

[00:01:35] Joe Allan Muharsky: this

[00:01:35] Andy Leviss: oh. You know, just hanging in there. Getting, getting ready to, uh, head out to California next week for, uh, uh, head to the, uh, Yamaha mothership for a few days as we kick off the new fiscal year.

[00:01:47] Sean Walker: All right. Cool. Cool.

[00:01:48] Andy Leviss: about being in the corporate world is we get to celebrate New Year's twice

[00:01:51] Sean Walker: Yes.

[00:01:53] Andy Leviss: or for those of us of the Jewish faith three times.

[00:01:56] Sean Walker: Oh, there you go. All right. Cool. Cool.

[00:01:58] Andy Leviss: But, uh, what's, what's up

[00:01:59] Sean Walker: parties. Dude, I'm into that. That's cool.

[00:02:02] Andy Leviss: What's up with you?

[00:02:04] Sean Walker: Oh, I'm just on vacation with the family. We're in, I'm in our travel trailer in a, in a campground someplace on the free wifi, so I'm hoping it holds together while I'm, you know, set up on the living room table here with the Sennheiser. Podcast thing. Just trying to make the whole thing work. You know? Got, I stole my kids' little blue Walmart headphones so I can, you know, have headphones on this one.

'cause I left my Dark Matters at home like a ding-dong.

[00:02:28] Andy Leviss: Dope.

[00:02:29] Sean Walker: That's all

[00:02:29] Andy Leviss: They, they look cute though.

[00:02:30] Sean Walker: It's fun to get out for spring break and hang with the family and do some, you know, not work stuff for the most part.

[00:02:36] Andy Leviss: Right on. That's some, some outdoorsy time.

[00:02:39] Sean Walker: Yeah, dude. Totally, totally cool. And it's, you know, it's family time, man. It's important. The kids are, kids are the age where they still want to hang out with mom and dad, and that's going to rapidly come to the close as they get older. So we're going to enjoy that while we can, you

[00:02:53] Andy Leviss: And here I am like a jerk. Stealing you away for an hour.

[00:02:55] Joe Allan Muharsky: ski.

[00:02:56] Sean Walker: You set of a gun.

No,

[00:02:57] Joe Allan Muharsky: No, I can't

[00:02:59] Sean Walker: just kidding, dude. I'm just kidding, dude. How you feeling? Better? You've been under the weather the last few weeks.

[00:03:03] Andy Leviss: Uh, yeah. I mean just, you know, keeping an eye on the eye thing and. You know, plug plugging along. Um, I'm officially, uh, I take statins now, years old as of a couple days ago, uh, related to all that. So that's, uh, there we are. But, uh, if, if, if you need 'em, talk to your doctor. Take 'em.

[00:03:24] Sean Walker: All I need is sparkle Waters, bud, and a nap. I'm good. And coffee,

[00:03:27] Andy Leviss: that's, you just, yeah. You just have this better genes than me.

[00:03:31] Sean Walker: Great. Fair enough, dude. Fair enough. Well enough about us. Let's get onto our, onto our guest today. We've got a homie hanging out with us that's gonna talk to us about kind of what's on everybody's mind, but nobody really knows the answer. But our guest today, Joe Allen and Maki, absolutely knows the answer and is gonna talk to us about AI and how that's gonna take our jobs.

I mean, help us out and not take our jobs. Joel Allen, and how are you buddy?

[00:03:58] Joe Allan Muharsky: buddy?

Um, I I'm pretty good, Sean. Tha thanks for that introduction.

[00:04:03] Sean Walker: Yeah, dude, good to see you. Thanks for being here. What, uh, what can we learn about AI that's going to help scare not scared? Where are we going? What's this road look like for us in the audio industry?

[00:04:15] Joe Allan Muharsky: industry. I, I think the, it's both, like, it's exciting and it's scary, kind of for the same reasons. Um, I, I do think in some ways AI is a natural evolution of technologies that we have.

And in some ways, the way that it's intersected our world is like completely unprecedented and gonna be wildly disruptive. And, um, the, the outcome of that is. Its choices we get to make. I don't think it's, it's written in stone, but it's, it's gonna change things and we get to decide where that chaos lands.

[00:04:46] Sean Walker: Okay. What, uh, what kind of things has it changed already and what kind of things do you see it changing in the very near future? Like, what are the things we should be paying attention to and expecting to come down the down the road quickly?

[00:04:58] Joe Allan Muharsky: So, I mean, immediately the nature of how you interact with software has already changed. Like when chatGPT launched, the thing that was different about it was you can just talk to it. You don't find the right button to click in the dial and hit this, drop down and then say, do this thing. And what happens?

You talk to it like a human being and you ask it questions like a human being. Um, and it answers in the same way. Um, and so this dramatically lowers both the barrier to using it. And I think just the cognitive load, like even if I'm really familiar with it, it's so much easier to talk to, to like having a person that can do what I want it to do than any amount of user interface.

Um, and that translates through into the audio world, um, too, there there's tools right now that are highly disruptive. Um, like suno and AI generated things, and that part is scary and ethically questionable, but there's this other side where, okay, we've got your basic mix now what are you gonna do with it?

And this is two-sided. Um, it makes operating with software intentional. So, um. Like instead of tweak, you know, I'm gonna tweak this, I EQ here, and I'm gonna tweak these faders here. When you're using this new AI software, it's like, Hey, I need you to bring the vocals more forward in the mix.

[00:06:26] Sean Walker: Okay. Is it gonna be able to mix the songs for us pretty soon, or can it do that now? Like in your DAW or in your console? Is that gonna show up and be able to sort that out?

[00:06:37] Joe Allan Muharsky: Um, I, so I feel like the cons, the daws are already adopting some of these features. Like a really early on was split to stems, like a little bit of AI assistive technology and a little bit of AI assisted coding just made, taking a song like it, it became commoditized in what was I using? Pro Tools and I think, um.

Studio one, both sort of at the same time. Were like, yeah, just give a song and we'll split it into stems. Then you can work with them. Um, and that used to be a very complicated process. And if it worked with a tool, it was still, well, it's fuzzy and you gotta go fix it. And I had a friend who was like, well, I need a karaoke track of this song and it's in the wrong key.

And like, free tools cobbled together in like an hour. I did it, I took it, I split it into stems with something free online that could do it. And then there's a chrome extension that literally when it's playing back, will let you rekey what it hears. And the, the crazy part wasn't that I was able to do it the next time she needed it done.

I didn't have time. And it was straightforward enough that like a 65-year-old woman with limited technical experience figured it out and did it herself

[00:07:51] Sean Walker: Wow, that's pretty impressive.

[00:07:52] Andy Leviss: That's cool.

[00:07:53] Joe Allan Muharsky: Yes. Um, unfortunately, people like me who would normally get paid for this, and I did get paid the first time, don't get paid anymore.

[00:08:02] Andy Leviss: Yeah.

[00:08:03] Joe Allan Muharsky: And I think there, there is a fundamental truth that every time we democratize art, we devalue it for the talented, this is true in video.

This is true in painting. You know, art used to be only the place for the highly educated, the highly trained, the highly talented. If you wanted to be, you know, if you wanted to do this, you had to have a radio station, right? You didn't, you didn't cobble it together in your trailer. You had an army of things behind you.

[00:08:32] Sean Walker: I feel so seen right now.

[00:08:35] Joe Allan Muharsky: I mean, seriously. If,

[00:08:37] Sean Walker: No, you're you're totally right though. You're totally right.

[00:08:39] Joe Allan Muharsky: and if timing had worked out, I would be doing this from my trailer in the Olympics too, because technology affords me the ability to do that. Have a beautiful mixer out there. I can just have a, a recording studio in the woods. And that wasn't like a, you know, run huge power lines and everything.

It's, no, it just plugs in. You take it where you want it to go. Um, but again, if you've made your life. Doing this, it's devalued it, you know, COVID and these market forces killed record plant Hollywood, like this legendary studio in la. Um, and I don't, I mean, I know COVID did it, but this has been happening, you know, the first time I took, um, I have a, like a studio live 32 digital mixer and I took it to, um, a studio.

Deval where I lived at the time, and it was the first time this guy had seen a digital mixer and, and he's like, this is like a wall of my stuff, right? Of my old classic gear because whatever I have preempts on all my channels, they're, they're quite good, right? Like, and I could just buy like a better mixer will have better ones.

And so this is displacing, right? Every time it happens, lots more people get to do the art. But now 10 times as many people are supplying art and there isn't more demand for it. So all that can do in a capitalist society is devalue it.

[00:10:04] Sean Walker: Yeah, which is kinda lame.

[00:10:06] Joe Allan Muharsky: But I, I don't think that's actually a flaw in ai. I think AI is exposing a flaw in a system that forces us to capitalize our art and not create for the sake of creation

[00:10:16] Sean Walker: Sure. Totally. How is that gonna affect audio engineers? Our, our audience is basically live, sound, audio engineers and, and systems engineers. How is this technology going to. Help us over the coming years and what are the, what are the roadblocks we should be looking for that are gonna drastically change how you think we do our job?

And that we may be, you know, where do we hedge our bets and where are we gonna be Really excited that this is gonna help us a ton.

[00:10:43] Joe Allan Muharsky: Yeah. Um, the biggest roadblock is probably. Again, it's choices by people. It's not technology like the, the last time I looked, 85% of producers are rejecting use of ai. If audiences reject AI generated music, AI generated music goes nowhere. I think AI is best when it is your buddy doing a thing, and so I don't think any amount of suno, which is this ai.

Music generator will create something as vivid as like Jacob Collier on his greatest day, or Jimi Hendrix back in the day. But I do think talented artists armed with these tools can move forward in a way that you don't need to have the holy unicorn of, I'm a creative, I want to create, I don't have writer's block or the right, like I have tools that can help jog the thing that you know.

People who become superstars can do all these things at the same time. Right. You know, I'm a great musician and I can write songs. Well, I'm a great musician and I can't write songs, but I have ideas and if I can work with one of these partners, then I can get a song out of it. Asking it to write you a song,

[00:11:54] Sean Walker: So like you're, it's your best personal assistant and

[00:11:58] Joe Allan Muharsky: is, yes.

[00:11:59] Sean Walker: yeah.

[00:12:01] Joe Allan Muharsky: Um, and I, that's when I think it's at its best. Another roadblock. Okay? These things are built to solve your problems as a user to be very personalized. Out of the box, there's sick offense to a kind of dangerous level. Don't just ask it an open question. Hey, is this a good idea? It's probably gonna say yes.

There's a few safeguards in these days. In general, that thing is gonna say yes, but the minute you tell it, your job is to be like an objective assistant for me, right? And to like, check me when I'm wrong. And if I'm right, then gimme supportive evidence of why I'm right, not just blindly. Yes, sir. Yes, sir.

Great idea, sir. Um, again, one of the brilliant parts of it is you can tell these tools to behave in that way, but you have to right now, 'cause out of the gate they're very talented, but they're highly a DD and they're extremely sycophantic. And they like to make

[00:12:54] Sean Walker: Squirrel.

[00:12:55] Joe Allan Muharsky: up. Oh yeah. So much squirrel. Yeah. Um, there's, did you, have you ever seen The Good Place? Um, okay, so there's this episode and I, I play it now for audiences when I do talks, um, where Janet, the AI personality is glitching and Michael wants this file who she's working with, and she keeps handing him a cactus. And he is like, and she's like, I have the file now. Do you have the file? Yes, I have the file.

Is it actually a file and not a cactus? It is actually a file and not a cactus. Great. Gimme the file and she hands 'em a cactus. And I had so many have had so many days and it's getting better now 'cause I know how to build scaffolding around it. But if you're using this, that will be your experience that

[00:13:39] Andy Leviss: I have, I've had that with Claude the couple times. I've like the, it feeding in an error message from some, you know, uh, Python or Apple script it generated and, and it's like, oh, I know why that error happened. I need to do this, this, and this to fix it. Here you go. And then you run it and it gives you the same error.

Oh, okay. Hang on, let me actually fix it.

[00:13:58] Joe Allan Muharsky: right? Yeah. And so like I, I have an engineering background and so I've built scaffolding into these tools in these things called, they're called Ralph Loops. Um, if you use Claude Code, which is kind of a level deeper than Claude like, even if you're not writing code, you get a persistent file system.

It's like cowork on steroids. Um, but there's these things called Ralph Loops. We're named after Ralph Wigga from The Simpsons. I am in Danger and it just sits there and goes, are you done yet? Are you really done? Are you really done At the end of everything's loop to tell it, did you really do this? Did you just, just go back and tell me you, you're sure you did this?

I was like, it's just simulating you go. Did you really? Oh, no, I didn't really, I just sort of talked through it and then did nothing. Um, yeah, and for, for the record, I think Claude is the best at it. Like the, it's the best at not exhibiting this behavior and it still still does it.

[00:14:53] Andy Leviss: Yeah. Every once in a while.

[00:14:54] Joe Allan Muharsky: yeah. So, okay, bringing up the coating actually brings up another thing because also you're, you, you people working in the daw, right?

And the DAW is already a lot of technical complexity and. Again, are you the unicorn who's also a coder and also has the spare time to write plugins for themselves and not be completely consumed by work? The fact that you can now vibe code yourself plugins for your daws can be game changing for a producer or a ma mixer or a master or anything.

And again, that's, I don't think the software will mix for you unless you as a master teach it how to mix. For you, and then it'll mix like you. And in that vein, a bunch of people can be better mixers. Right now, the thing is, it favors the people who know what audio should sound like more than the people who know how to turn the DOB knobs.

And although people like to imagine that those are the same type of people, it's like, oh no, I've let 'em met a lot of people who know the knobs and really don't understand what they're listening to. We're trying to pull out of it. And I know people, and I used to be one that really know what they want out of it and have no idea what to do with this spaceship to, you know, make it, make those sounds and this, this new world completely favors If you're creative, if you're intentional, if you know what you want, you know, the, the idea guy, it, it really favors you in a Yeah,

[00:16:25] Sean Walker: Really a lot. Of the same thought process as the vibe coding, where you're like, I want to do this thing and it will help you do that thing. Is kind of the, is that the, the i for lack of better terms. That's the vibe you're talking about. Like,

[00:16:38] Joe Allan Muharsky: Yeah,

[00:16:39] Sean Walker: it's not gonna do it for you necessarily, but if you're like, Hey man, I'm really thinking I want this, this, this, that, and it'll, it'll be able to do that.

[00:16:47] Joe Allan Muharsky: I mean, certainly, and this is what my, my software startup is about and my use of it is people want to like skip all the work between, right? They just wanna sort of say, well do this big thing. Right. And in, uh, I'm an avid role player, so role playing turns, man, you're, you're just doing a blind role and you might roll a 20, you might confirm you're 20, you might actually get what you want.

If you're not an expert in the thing that you want, you have no idea how you got there. You actually don't know if it's right. You just know it did a thing and didn't break. And the process that we use is much more about. Like it's more systemic, but it's more the way that I as a consultant would've engaged with people.

Tell me about your problem space. Let's learn about it together, and we'll either I'll get from you or we'll assemble this corpus of knowledge about the space, right? And the rules of how to do it. So like I want to build a plugin to do magic mixing. Well, we're gonna go do research on mixing right? And what that is, right?

And general audio engineering and the DAW that you're using and then the language that we're gonna build it in and the best practices for the languages and the libraries people might have done. And I don't have to know it about any of these six things, I just have to tell it to research those six things and then take that collected information and then getting the tool you want is almost out emergent out of that corpus of knowledge.

This teaches you as you go also, right? You can use these same tools to like distill this information into densities and modalities. Um, you know, do you wanna see this as a picture or do you wanna see it, hear it in audio or, you know, increasingly a video, um, the emergent crazy world and the one that. Is is daunting from variety of copyright reasons, but I wanna learn about a topic and I want you to give me a very special episode of my favorite eighties TV show that teaches me about this topic.

Whatever, like just pick arbitrary topic

[00:18:44] Sean Walker: I can make eighties TV shows. Oh, I'm in, bro. Me and the, maybe it's. S me and the A team are about to learn about a lot of stuff. 'cause I love it when a plan comes together, bro. You know what I'm saying?

[00:18:54] Joe Allan Muharsky: yes. I, I will say one of the earliest signs that I was actually a designer, not a developer, and that I love to coordinate people, friends of mine in, I'm dating myself, but yeah, it was elementary school. Um. Huge fans of the A team and like my contribution to the group was to fire up an early spreadsheet and start like diagramming out how we would lay out the van.

[00:19:17] Sean Walker: More power, dude. That's how you lay out the van. More power,

[00:19:21] Joe Allan Muharsky: power. Yeah. But also where do you poop? Where do you poop? Became a very important

[00:19:26] Andy Leviss: Not in the

[00:19:27] Sean Walker: don't live in

[00:19:27] Andy Leviss: rule number one.

[00:19:28] Sean Walker: go, you know? Yeah.

[00:19:30] Joe Allan Muharsky: I know. I

[00:19:31] Sean Walker: there's no pooping on the bus. Everybody knows that.

[00:19:33] Joe Allan Muharsky: no, but. I will say no. Elementary school, I don't think it ever occurred to us that there was another place. Just it's them in a van. Right.

[00:19:42] Sean Walker: Oh, that's funny. I see what you're saying.

[00:19:44] Joe Allan Muharsky: in some hotels, but Yeah.

[00:19:46] Sean Walker: Yeah. Everybody knows you go poop at Walmart. Like, come on dude, go. You know, please. Not in the van.

[00:19:51] Joe Allan Muharsky: Not in the van.

[00:19:54] Andy Leviss: Now, I'm trying to think back through like every season of 24, did we ever see anybody go to the bathroom?

[00:19:59] Joe Allan Muharsky: Uh, there's a, there's a great. That was one of those role playing d and d skits too. It's like, have we ever done this before? I, I guess so. I've never said I've done it. Yeah. Anyway. Um, yeah. I, okay, so this is why AI's scary. It makes the possible trivial at scale to anyone who wants to do it. You kind of just have to be methodical, like I think good project managers or good organizers, right?

This could be a, a person who is simply like a homemaker but keeps like a detailed schedule and everything. This is really for you when you can think systemically like that and you can communicate to people, um, this tool is at its best and it's an amazing assistant and it's a. Ridiculous force malif for creatives who want to create and get the other stuff out of the way.

[00:20:54] Sean Walker: What about technicians that need to technician, uh, in live sound? It's probably not going to fly speakers for us, but could it mix the show for us at some point?

[00:21:04] Joe Allan Muharsky: Okay. Um, yes, and again, I mean, it could probably today and if I took the worst audio engineer and the best AI mixing tool. It's probably got em right. Um, and that's only gonna get better. But I do think audio is still subjective. Now there's probably a magical loop. Like if I'm really imagining, if I really think through the technology, security is big on this, cameras can stare at crowds and get send image analysis real time.

And I mostly think about this with my work in festivals and events because this is security, right? I wanna know where the angry people are. But you

[00:21:43] Sean Walker: said I was sorry. It was one time.

[00:21:45] Joe Allan Muharsky: You could, you could be doing real time sentiment analysis of your crowd to tell that like, Hey, people are getting overwhelmed by the noise.

I need to bring it down, or whatever. Right? We need to kick up the base and get the thump going, right? Like you, you can imagine a world where this happens, and I choose the world where we're sitting at the helm of that, but that's a choice because it's gonna be good enough. If people want to hit the Just Go Drive button and replace us, they can.

Um,

[00:22:17] Andy Leviss: they'll get what they get.

[00:22:19] Joe Allan Muharsky: and they'll get what they get. But again, if that's good enough, it's good enough, right? Like I do think mechanical licenses are screwed for, for audio people who create audio just so it can, like a tidbit can be played in a TV show. And the background, oh my gosh, no. Suno today can make better stuff than.

What, whatever be musician and you're already hearing it, I'll guarantee you a ton of those. Like used to be mechanical licenses, paying sort of commoditized musicians to make, um, scoring and, and little background, you know, jingles and pop songs.

[00:22:53] Sean Walker: All the gum and insurance commercials. Are everybody doing those is fired. Yeah.

[00:22:58] Joe Allan Muharsky: Right. And, you know, there was a big. Deal. If you want to work with the Actor's Guild, you don't get to replace them. But again, watching my phone, there's a clear difference between a person with all their little tweaks and it's like, oh, cheap little games. Suddenly you can afford like the most chiseled, right?

Like GQ looking man, confidently talking to the camera. So you don't have that kind of money. So you just, AI generated a dude. Um, lot of fake podcasts that look like this. And they all use the, the sure. They, they all use that Sure.

[00:23:31] Sean Walker: The SM seven b.

[00:23:33] Joe Allan Muharsky: Yeah. The SM seven B, and they're all a consistent distance away from it, and they're all pleasant.

They don't look like robots, but they're not animated in the way that a person is. But again, that gap's, it gets vanishing like week by week right now

[00:23:47] Sean Walker: Well, that's gonna be perfect because Andy and I have face for radio, so when we have to put this on video, we're just gonna do that. So we're hire some handsome looking studs to go

[00:23:57] Andy Leviss: if we're doing a video of, of this podcast, it's a cartoon man.

[00:24:00] Sean Walker: Yeah, a hundred percent dude. A hundred percent.

[00:24:03] Joe Allan Muharsky: and, and in general when I use AI for generating stuff, like I'm not trying to generate real looking things, reality's there, go enjoy that. Um, my, my spouse is a digital, digital and traditional artist, and they use it as a brainstorming tool. And, you know, our, our general principle, and again, this has to be a choice.

When we have made enough money with the things that we're using AI to do for free because we can't afford to pay people to do them when we've made that money, we will pay a person to do it, and we will still use the AI for the rapid brainstorming, but the end product should be made by a human. That's not a profit based choice though.

And in a system where everyone is like justified by, well, what saves me money, do good enough and don't employ a person, it's gonna be cheaper. You, you have to have a value at play that is greater than, is this profitable, or we as sort of the, the labor class that we are, right? Like just doing things to, to make money is kind of gone because AI's gonna do this faster and cheaper.

Now, if audiences refuse to go to places where there isn't a, you know, heartbeat, breathing, human standing at the audio board. That's a choice that can make that, you know, preserve that in live sound. Um, you know. Okay. But you, the other one was you said about flying, uh. Flying the speakers. So the Chinese, it, it was one of the festivals around their new year.

They had a crazy good martial arts team of kids out doing this full martial arts routine, interleaved with robots, doing them in extremely complicated, like back flips, tosses, all the things you would want them to do. Um, they are absurdly. Ahead of us. And again, if I have a robot who can do it, they never get tired.

And more importantly, if I even wanted to take a human values, they never fall off a scaffolding and die. That never happens with a robot, right? I guess they land on somebody that that could happen. So, but I don't. We built a world where everyone has to work to make money to live, and there were lots of sort of reasons that that had to be there maybe as a function of our world and lots of reasons that just, it was AI is rendering that obsolete and this is where the big choice comes in.

We can go, Hey, we don't have to do the bad manual labor part. We have tech to do for us. People are the point though. People are the point. If profit is the point, robots replace us. If people are the point. The robots just free us up to do more interesting things, right? To create arts for art's sake and to enjoy it for art's sake.

It, this is where it's the most disruptive because I don't see a different path. There's either forces that are gonna use it to make us obsolete, or we are going to choose that we were the point in the first place, and the technology just makes the toil obsolete.

[00:27:15] Sean Walker: So at what point does it become Skynet with Terminators everywhere and we're, we don't, we're running for our lives bud.

[00:27:21] Joe Allan Muharsky: So I, I will say AI architecturally doesn't. We choose what agency we give it. Right? Like technically it's already Skynet, right? If, because everyone watches like Skynet or The Matrix and we consider it, yeah. Horrible. But we're also like, well, yeah, I see your point. Right? Or the DA Earth stood still, right?

We don't watch those things and not get them, and this is why it's. This is why it's important and also what is entirely enabling ads to bias ai. It is important to put guardrails that aren't just pure logical thought experiments, right? That, that have bias and value at play, that human, human life is more valuable than something else.

Those are going in at different rates, uh, in different, in the different models too, that people are using and in different ways.

[00:28:16] Sean Walker: Fair enough so it, it can fly speakers for us and it can mix. Us. We just need to tell it what to do.

[00:28:26] Joe Allan Muharsky: You just need to tell it what to do and train it, what to do. But it is frighteningly good at picking up those patterns and then just doing them. And so there is a world that if you've taught it to mix the show and the show doesn't go wildly off the rails. Yeah, it works. It's also, okay, so the other thing to the Skynet point, they are.

Pattern matchers. They don't actually think, they're not actually intelligent. They're like pattern matchers. They're like auto complete at such an absurd scale that it appears to be intelligent

[00:28:58] Sean Walker: Oh.

[00:28:58] Joe Allan Muharsky: and it can seep patterns and it can interpret the space between them and sort of, I interpolate. That, but it's not actually creative and it doesn't actually know things.

You just have to bound it with patterns that tell it what would and wouldn't match, and it behaves in this way that appears to be intelligent, but it's not, it's not emergent except in the ways that you've explicitly told it to be so. So that's. That's the deep, that's the Oh, okay. It's not that bad. The downside is the tools that this enables.

The scale of coding and the scale of development that it has enabled almost certainly is the technology that will build Skynet. Right?

[00:29:39] Sean Walker: Right.

[00:29:40] Joe Allan Muharsky: I don't know, is there an example of someone building an AI and they're like, Hey man, everything worked out like the A, like we just took it, made Star Trek out of it.

'cause even Star Trek like. That's where I think this can take us. But if I look at Star Trek as a model, yeah, we have some rough times coming up on that timeline before it's all sorted out.

[00:30:00] Andy Leviss: We, we gotta figure it out as we get there.

[00:30:02] Joe Allan Muharsky: but I, I don't know, this is way off audio, but most of AI is about, to me, the dangers are about human choices. And we don't have to do this silly boom bust cycle that we do with our civilization, right?

Like we can get over our traumas and then just move forward in a healthy way. And I am ever hopeful that every time the wheel turns that we can grab it and go, oh God, stop turning. Just pick this direction. Pick that. It's not as exciting as the spin, right? As the. Sensationalist, you know, we, we build up this affluence and then we get sort of hubris about it and they're like, ah, I should have all the stuff.

And then we get over, you know, over-leveraged and over selfish and then it explodes and you know, it did it in the forties, kind of did it in the eighties, kind of did it pretty recently. But we didn't learn any lessons or make any structural problems and, you know, we're, I don't know, just the world is really moving hardcore in back into this fascination with.

Fascism and efficiency and profit and, um, that is depressing, but also aware that that wheel turns right, that it never stays there. Um, and I'm hopeful this year that it's,

[00:31:17] Sean Walker: history tells us that we make the wrong choice every time, right? So I like your optimism, but, uh, I by fear that we will continue to make the wrong choice. I.

[00:31:29] Joe Allan Muharsky: sure, I guess my why I have the optimism, I call it faith, is that you will either go towards what you have faith in or you have fear in, and we're way too good at manifesting our own realities that I'm like, if I believe that we can get through this and we're all in it together, that might get us through it.

If I believe that. No man, it's all screwed. Those behaviors will definitely make me correct. Right. There's no chance of getting out of it if we all believe that it's hopeless, and so that's like, that's my intentional faith there.

[00:32:02] Sean Walker: I like that. That's awesome, Andy, your thoughts?

[00:32:06] Andy Leviss: I mean, I, the, the thing I find interesting and, and that we, like, it kind of occurred to me to mention earlier with one of the things. That you brought up Joe, and then when you were just explaining a bit about how, you know, how these technologies work, I think that's where it's useful to kind of help figure out what things these various AI technologies are good for and what they aren't.

And I feel like the deciding factor there a lot is if you don't have, whether it's a creative endeavor or a knowledge endeavor, if you don't have the ability to look at the output from it and decide either this is good or this is bad, that's where it. I mean, in certain situations can become dangerous, but in others just becomes not as useful.

Whereas if when you can, then it can be really useful. So like for like, you know, some of these examples of like having it help you compose something or help you mix something, like you're still evaluating the output there. So there's the ability to go in and fine tune it. Whereas like we look at like, you know, Google's AI and chat GPT and when people go to that for.

Like, as somebody who works in tech support, this gets personal for me. You know, uh, not only for me, but on behalf of my customers, that that's when, like, you know, when you go to it about something that you don't know because it is, uh, just this probability machine. It's, it's guessing, but it's guessing in a way that sounds much more knowledgeable than. Than it is. And that's where it leads people into trouble when you don't have the knowledge to know that it's, and sometimes it's just, it wastes a bunch of time. I mean, sometimes, like I, I dunno if I've talked about it on the show before. I, now I've mentioned on the Discord, like I literally had a customer spend a thousand dollars on a network switch he had no need for because Chat g PT told him to hook up these two Dante devices together.

You need to have. One of these three models of switch and you bought a thousand dollars network switch and it was like, nah, dude, you're connecting like, you know, a stage box to a console. You need a cable, you do not need a switch. I promise you that is not your problem.

[00:34:04] Joe Allan Muharsky: Yeah. So. Alright, so like you got to like, what are these great tips to use? And one of them, like the, the most important thing you can do with an ai, the first sentence you give it, tell it what its role is. This thing pattern matches everything in reality. So you're like first rolling the dice on who you're asking.

You might be asking the janitor, what should you do? Oh, you know, I know a cousin, he did this right. And it answers you that way if the first thing you tell it is like. You are a professional audio engineer with extensive knowledge of like consumer hardware and you seek the simplest solution, and then you ask it your question, you're gonna get a very different answer than sort of the unbounded, how do I do this?

Right? Some,

[00:34:43] Andy Leviss: Uh huh.

[00:34:44] Joe Allan Muharsky: dude in the crowd raises their hand and exhibits stunning Dunning Kruger effect with their answer. Very, very authoritative,

[00:34:53] Andy Leviss: And I'm like, I'm thinking through like prompts and how yeah.

[00:34:57] Joe Allan Muharsky: Yeah. So. Um, you know, a, a great thing to cover when you're writing a prompt, like a simple vibe I want to plan, right? Or how do I fix this? Um, again, you're wildly rolling the dice on what you get out of that.

If you break it down into a structure, and the simplest one most people already know is the like, who, what, when, where, why, how? Right? And so the first is who, who are you? Who is this agent that I'm asking? Right. Um, and just tell it it's an expert in the thing that you're doing. Right. And tell it. If you are trying to have it be objective that it doesn't hallucinate these much, like, you know, the Pirates of the Caribbean line always comes up.

It's like, oh, they're more like guidelines, right? The code. Yeah. That's the way it is with prompts. They're guidelines, but it mostly follows them most of the time. Um, one of the reasons I work in Claude Code is I get to build scaffolding that is not. Guidelines. They are rules, they are boxes that it, it can't escape without human permission.

[00:35:57] Andy Leviss: Copy. That's interesting. Yeah. 'cause I mean, there's just, there's so many different models here and it's, yeah. For the, the average, uh, lay person like me or Sean, um, or, or somebody with better grammar than I just displayed in that last sentence. Uh, uh, yeah. It's, it's hard to know what is the difference in what's better at what.

So it's, it's interesting to, to learn some of that like distinction there.

[00:36:20] Joe Allan Muharsky: That answer is rapidly changing. I'll say like in the last month, I think even since I did the, the intake talk for this, like I don't use chat GPT, the only thing I will do with chat GPT is teach people how to not use it. Um, and there were like three, three concrete things that happened in like the space of two weeks.

I think, and one was him talking, Sam Altman, the CEO of OpenAI who owns chat, GBT, talking about how, you know, people talk a lot about how much energy and water it takes to do an inference, but they, what they're not taking into account is how much energy and water it takes to raise and educate a person.

And I'm like, are you gonna like, pull off your human mask and show us what, whatever species is making this argument right now? 'cause this

[00:37:03] Andy Leviss: I was gonna say, wait, are you Kang or CODOs?

[00:37:05] Joe Allan Muharsky: right, right. But, but, okay. So that was one, two, and this is why I'm happy 'cause I bet so many things on Claude, but people are the point. And so I have lines on this Department of Defense made a big stink about forcing Claude to allow autonomous kill without human intervention and just bucket surveillance.

Just hand claw the whole of whatever data you have and ask it to make inferences and, and anthropics like. No, those are against our terms of service. We're not making exceptions to that. They've literally been declared a supply chain risk by the federal government over their refusal to do this. And on the day that they refused, and this came out, Sam Altman is both writing a public letter of support for clo for, um.

Philanthropic and signing a $200 million deal with the Department of Defense, which, you know why he got that deal. Right? Like a detail not disclosed. Yeah. They're letting Chachi bt kill people. That's, that is certainly my takeaway. I have no insider knowledge of this, but, and then the third one was he's at some, like, imagine the future and he is so excited to imagine a future where intelligence is sold like water. And I just, I'm like, I just. It's so sad to me. It's honestly sad that people with that much resource and that much power imagining the future, can't imagine in something better than rationing out their power to us.

[00:38:34] Sean Walker: So he's actively working on Skynet where we're all replaced, and the other one's not yet.

[00:38:40] Joe Allan Muharsky: Right. And I don't hook man, no one's innocent in a late stage capitalist system. But I do think in my experience, um. Clawed and anthropic have been the most open about the challenges and the pitfalls. Um, they have this new model called Mythos that is insanely good at coating, and they have this, it's called Project glass wing, right?

Because we're soaring on these wings made of glass. And the idea is you have all these, these high tech consortiums that are gonna take the smartest model and patch all their stuff before the world gets their hands in that model that could go find the holes in that stuff. Right, because they know the minute they release it, these things are smart enough to find holes in all of our software, um, in really shockingly easy ways.

[00:39:27] Sean Walker: So is that finally gonna fix Pro Tools from Crashing and we don't have to have the Save Twitch anymore.

[00:39:32] Joe Allan Muharsky: anymore? I, so, okay. What's, what is disappointing to me, and it might be like hubris or just nervousness about companies. I, I was a coder for 30 years, right? And I have written more code in the last six months than I did my entire career, and I haven't written a single line of it, and I don't even read it anymore.

I have tests, I have lots of tests. I have lots of specifications. I have lots of different agents who look at it like I would have a team, but I behave like an engineering manager or a, or sometimes even now a user. Just telling it what I want. Right. And sometimes all the way down to the level of a project manager, but never a developer.

I dunno. I I have agents that write code. I, I behave like a project lead. Sometimes I'll give pointers. Um, and that isn't going away. And this is, um. This is true for audio too, for your audio hobbyist technician, and I do think there's a misplaced fear and you're sort of watching the world realize it.

There's some pullback from AI in 2026 on a lot of big companies because they all saw this tech and immediately imagined it was a replacement for us across the board. And I'll say if you're low level, if you're like turning knobs and doing things and you like type mails and send them, it's very dangerous if you make decisions.

If you actually make decisions and get things done and coordinate people, this is not a replacement for us. It is a, it's the crow that can do really, really, really impressive tricks, but it's still a crow. It's not a person, and there is human judgment that practically right now still has to be applied.

That may not be true over the years as we develop the systems to behave more and more acceptably, but right now. Companies are realizing getting rid of a whole bunch of their expert decision makers in favor of these little hallucinating crows was not a good idea. Um, and big tech is desperately trying to make this technology big enough to solve problems at its global scale.

Because of that, they've sort of missed the fact that it already operates at our scale. And it keeps getting better. And it is a really, really, really effective replacement for bureaucracy, which is what big studios offer. It's why it it, right? Like their ability to do marketing for you at scale. Their ability to, um, you know, provide you professional.

Production tools? No. If you're an individual artist who wants to be motivated, this can take the place of all of that for you and just free you up to be a creative. This is true for artists and it's especially for true Foretek, and I do think that most of big tech sort of doesn't realize that AI actually lays the groundwork for their own irrelevance because we can build better stuff than they do.

That is exactly what we want it to do. Maybe even the duh, right? I haven't played with this, but they're simple libraries for doing sound processing, right? You could probably in a weekend recreate, is it audible, is the open source, duh, uh, uh, audacity or Audacity. Yeah. I, I would be willing to bet that in a weekend you could recreate audacity.

Without knowing code, just knowing what you wanted audacity to

[00:42:45] Sean Walker: are you coming?

[00:42:46] Joe Allan Muharsky: and being very systematic about how you told it to do it. And then within a month or two, you've created a new idea, right? As you tweak that into what you want it to be, um, anyone is going to be able to build anything with only the will and like creative desire to do so that that is just coming.

[00:43:08] Andy Leviss: And I, I think what's interesting what you just said is, is that there's, you have to at least be able to describe what you want, which is something that, like, in theory, anyone can do, but in reality, not everybody's gonna be able to do. So it's not like it's, it's completely removing the skill, I guess. It's just shifting what the skill is or where the skill lives.

[00:43:28] Joe Allan Muharsky: Right. And so, you know, to that point you were saying about the music, like being a music producer or a mixer or something, or like, there's these tools that can, Ooh, it's gonna write a novel for you.

Yeah. If I'm a crappy novelist, it's gonna make me the most prolific, crappy novelist in the world. Right. I'm gonna put out so many crappy novels, but I don't know what a good novel is, so like that's all it's gonna do. Um,

[00:43:49] Sean Walker: And, and what you're saying is it doesn't know what a great novel is either, is that what you're saying? It, it knows the framework of a novel, but it can only do what you're telling it to do, or it does know what a great novel is, and it can write a great novel for you.

[00:44:03] Joe Allan Muharsky: it could. A great novel is more than like it could write a great technical specification

[00:44:12] Sean Walker: It knows the framework of a great novel.

[00:44:15] Joe Allan Muharsky: there is a humanity to a novel that is going to come out derivative. Or not have a point now. I, I think that's right now, but again, even as I think through this, well, there's a structure, I even know in my head, like with the, the arc of a story, right?

Like you take the hero's journey for a really simple one. Well, if I've built that in with the hero's journey and I've built sufficient little. Points, right? Like, do I have characters? Do they sort of have story beats along the flow, right? Do they connect so that there's, at the end there's this web of connections that have come together for a finale.

There's probably an algorithm that solves this, right? And then the end, you know, I don't know, like back to the Star Trek metaphor. 'cause it's the closest, if you have the holodeck, you still have your live music performances, even though, right? You still have real world. Choices. And some of that is just a choice, right?

This is going to be a dangerous world for people who simply choose to live in the virtual, um, virtual, like virtual, the nature of virtual reality in those virtual worlds too. AI is gonna usher in a absurd level of immersion in there to believe that you're in a better place. Um, yeah, so I don't know. Like I see like there's.

Like the humans is. Humans is nothing like I, I don't know. I guess Terminator kind of future or matrix future, right? We're only useful is like abstract batteries. There's that utopian Star Trek future. There's also the Wally future kind of in the middle where we all just get lazy and don't do anything. I don't actually mostly see that in people.

I see that when people. People being lazy like that are people who've like been crushed by a system that they don't believe that they can go beyond that and they're not there because they're happy? Some are and they should be free to be content, but I think a lot of people, if freed of their, like you could create or go do things or, or make meaning in the world, I think people would choose to.

And so I do think there's a somewhat irrational fear that. As that need to work goes away, which I think is inevitable anyway, that we will choose Wally. Um, I don't think so. I think it's more likely we'll actually choose one or the o like one or the other poles. We'll, we'll either go hard dystopia or we'll just be like, no, people are the point.

Technology's a tool.

[00:46:45] Andy Leviss: Hard Dystopia is the name of my nine inch Nails cover band.

[00:46:48] Joe Allan Muharsky: Oh, nice.

[00:46:49] Sean Walker: totally. I, for one, am not excited to, uh, shoot Terminators and run for my life with Arnold Schwartzenegger. So let's check the other one. Let's just choose the other option, right.

[00:46:58] Joe Allan Muharsky: Right. Yeah. Um, yeah. So like partners, yes. Replacements, no. Um, I've always felt this way about software automation, uh, all my career, but it's especially poignant now.

[00:47:12] Sean Walker: So as our, as our listeners are, you know, trying to put this together, how do we, uh, how do we clarify and simplify this last 45 minutes? Like what are the things that, as a working live sound. Engineer right now we should be keeping our eye on and what are some of the things that we should be absolutely prepared that are gonna change and we don't get to do anymore in the future?

If you had to, you know, rub your crystal ball.

[00:47:37] Joe Allan Muharsky: ball? Um. I mean, I've already seen the push away from physical, tactile surfaces, and I don't like it. I'll, I'll admit I don't like it. Every time a knob and a dial and a physical surface goes away and I'm left with a flat touchscreen with no feedback, I dial a little inside. but I'm pretty sure they're gonna move.

Like you're gonna be sitting there with a flat device in your voice, right? And you'll choose your voice. And the I, I do think that the idea of the physical console will go away unless ven venues or audiences simply demand it to a point where it can't. Right. But we'll probably go through this cycle of making it obsolete and then like the record, which is really cool.

And then the cassette, which is a horrible idea to bring back, but Right. These technologies, oh, you brought back the disposable one that degrades upon use. Cool. Um,

[00:48:41] Sean Walker: Cassette tapes were never cool, dude. Like, get out here. Like, records were cool. Eight tracks and cassettes were never cool. Like get outta here.

[00:48:50] Joe Allan Muharsky: that that chilling sound as the tape ate, as the machine ate the tape

[00:48:54] Sean Walker: Jesus, every time, dude.

[00:48:55] Joe Allan Muharsky: Um,

[00:48:58] Sean Walker: I, and I'm talking about cassette tapes. I don't mean two inch recording tapes. I mean, I mean

[00:49:02] Joe Allan Muharsky: Sure.

[00:49:03] Sean Walker: I mean, you know, the, the cassette

[00:49:05] Andy Leviss: He was about to get the analog studio guys all,

[00:49:08] Joe Allan Muharsky: yeah. No, no,

[00:49:08] Andy Leviss: his face.

[00:49:09] Joe Allan Muharsky: no.

[00:49:09] Sean Walker: No, I meant the Walkman era cassette tapes that make you wanna blow your braids out every time it munched in the middle of your jog.

[00:49:15] Joe Allan Muharsky: Yes. Consumer age, disposable crap. Yes. Um, also a problem here. It's gonna be a flood of technology and we are obsessed with high power, easily disposable short life technology.

Um, and it's gonna be disruptive, right? As people come out with, Ooh, here's our headless console, right? Or whatever, right? AI enables people to be creative in sometimes healthy and sometimes unhealthy ways, right? And it's going to enable lots of different industries to be wildly disruptive. Um, what else? I mean.

[00:49:57] Sean Walker: How disruptive, like you say, disruptive, but what is that gonna look like in in a day? Day-to-day job for us audio engineers, right? How is it gonna disrupt our day-to-day work from now until the next, you know, two to five years or something like that? Are we, we don't, we want to have massive consoles.

[00:50:13] Joe Allan Muharsky: I, I don't think you have massive consoles. I think the UI is going to radically simplify in favor of voice and prompt and I mean, already most of these things that you can prompt, you can just talk to and voice recognition also getting really, really, really good. So, you know, on the good side, you can get rid of some of your carpal persistent carpal tunnel.

Audio engineers have, um, on the other side, a whole lot of your specialized knowledge of how to operate pro tools or logic or these things go away in favor a person with intention. And then that means to me the disruption then of what are those companies going to do to produce software that is useful in this new paradigm.

Right. How does Pro tools differentiate itself from, um, you know. Studio One or Logic or whatever when it is, Hey, can you sweeten my voice and bring it a little forward in the mix, please? Right. And the, the base is too mushy, right? Um, yeah. So. mean, I, I guess for audio engineers, make sure that you are focusing on the sound itself.

Right? And I remember when I was in digital audio class, they'd do this like, look man, no matter how much you look at that little wave form, you're not making pictures. The only thing that matters is putting on your headphones and listening to it over and over and over again. And so that that world is going to favor audio engineers who have focused their abilities there and not the ones.

And I totally understand why in the old world you would've focused your abilities on the technical side of operating it. Right. And because we're finite, like we have finite time, that is at the expense of how tuned are you at listening? Um, yeah. I.

[00:52:06] Sean Walker: All right. Well that gives us, that gives us plenty of info to chew over. Is there anything that we haven't asked you yet that you'd like us to ask you? We're coming up here on an hour and.

[00:52:15] Joe Allan Muharsky: Yeah, uh, I, I would, I like my, my big high points are it is, it is not a human, never mistake it for a human, but it's at its best when you treat it like a human. Um, and it's not an expert in anything and it's at its best if you tell it it's an expert in something.

Yeah. And that it's just a tool and it's a really, really helpful, like transformatively useful tool. Uh, in fact. Okay, this is the other one that I would actually say for an audio engineer that's a huge deal. You can like upload a PDF of a manual and then bind, basically bind a brain to that manual and then start asking it questions.

So RTFM becomes the easiest thing ever because it reads the manual and then is the person who read the manual that you can ask questions of.

[00:53:06] Sean Walker: It's your best phone, a friend ever. Andy, I never have to call you again, bud.

[00:53:10] Joe Allan Muharsky: bud. So you can be like, Hey, I'm in this venue and I have like this model of mixer and this model of compressor and this isn't working.

Help me figure out why. And it like the good tools can not only give you answers, but they're citing click here to go in the PDF of why. Right? Like, here's in the documentation why I'm telling you this. Um, and so yeah, I missed that one. But that, that's a huge deal to people because I know as audio engineers, especially if you're like mobile and you're going like with a band, sometimes you're just like, well, this is the audio system they have and this is what I'm working with.

Um, and it can be, uh, just an amazing at digesting any sort of expertise but handing it, things like that. Here's a manual is so much better than just, I have a, I have a mixer. How do I do this? Now? Grab your PDFs, hand it your PDFs, and it can read them and it can cite sources. Um, that's probably the most immediate day-to-day for an audio engineer.

If you don't have a stable, even if you do have stable hardware, you might find cool features.

[00:54:19] Andy Leviss: Yeah. And, and, and I encourage folks to then take the moment to go back to those sources and just double check that it got the context right and stuff like that.

[00:54:28] Joe Allan Muharsky: Well, and so there's some tools, like if you use Notebook lm, which is a Google tool, it's only bound by what you uploaded it, and it puts citations in the text that will literally open the document that you attached to the place.

So there's some that you can use other ones, yeah, you always ask it. Like always is a follow up question. Hey, double check your work. You didn't make anything up.

[00:54:50] Andy Leviss: Yeah, and I mean the, I mean, the fact is like right now, like humans write those documents and sometimes the documents are not well written. So just like we can misinterpret them, sometimes these things can misinterpret them, which is. Yeah, which is why I said go back and root. 'cause like I've, I'm trying to remember what it's called.

I forget what engine it uses, but I wanna say it's like, it's like PDF DF Pal or something like that as a program that's at that, like you feed it one or more PDFs and basically the whole pitch is have a conversation with the pdf. So like I've fed, you know, the, you know, the owner's manual and a reference manual for like a Yamaha console and do it and ask it questions and Summit gets right and summit.

It'll cite me sources, but when you actually go back to the pages it's talking about, it's like, I see how you got there, but that's not what that meant. But also it's not the most clearly written manual. So I see how you got there.

[00:55:33] Joe Allan Muharsky: there. Right. I mean, in general I'm like, like I'm giving you an army of researchers and they're incredibly well read. They're semi stable and they're quite a DD, but they're almost free. There's a use for them, but that's not just like hand you the keys and blindly trust what you tell me, but.

You know, and especially if you bound them by, Hey, double check your work. Yeah, you, you do a couple passes of that, you probably have something that's accurate. So I guess be patient with it too. And don't be overtrusting, like don't just ask a question, get an answer and move forward. Especially

[00:56:09] Andy Leviss: one right there. Yep.

[00:56:11] Joe Allan Muharsky: Yeah. But I, again, to your point, like just pretend you have a very talented team that is sort of new and green and has a lot of intelligence and not a lot of wisdom.

[00:56:21] Sean Walker: That's an excellent, that's an excellent, uh, way to summarize that. That's awesome. Joel Allen, and thank you so much for coming and hanging us chatting about AI and getting us all hip to that for the hour. We really appreciate you.

[00:56:34] Joe Allan Muharsky: you. Yeah,

[00:56:35] Sean Walker: Thank you to Allen and, and Heath and RCF for letting us chat for the week about Audio and Nerdery.

That's the pod y'all. See you next week.

[00:56:42] Joe Allan Muharsky: week. Alright, thank you so much.

Music: “Break Free” by Mike Green