Hash Rate - Ep. 170: Vocence - Subnet 78

Added: 2026-05-23

[00:00:01]Hello everybody and welcome to hash rate. My guest today is special K of Bossense subnet 78. How you doing special K?

[00:00:10]>> Doing well. Um very special. My name is Special K.

[00:00:15]>> Well I call you by your real name but but like you don't like I was saying earlier like you don't call Thomas Anderson, you call him Neo. That's his name. Exactly. Right. And K is his name.

[00:00:25]It's not Jacob. It's not Jacob Steves.

[00:00:27]It's Kant.

[00:00:28]>> Yeah. Yeah. So, you're special K.

[00:00:31]>> Exactly. And then, you know, since we proved that we have been living in the matrix, well, let's just go with our, you know, quote unquote coat name. So, >> yes. Now, you >> What's yours, by the way?

[00:00:42]>> Um, I I don't really have one. I'm just Mark Jeffrey, and I'm not the same everywhere. It's sort of boring, but it works for me, so I'm not I'm not going to change it anytime soon.

[00:00:49]>> All righty.

[00:00:50]>> So, um, so you technically you you run a couple subnetss. Actually, you run two, right? But what is the other one you run besides Vosense? So 2026 it's called perturve. Uh so essentially it's applying some sort of noise to the data sets like a perturvation you know impact is this term used in computer science.

[00:01:09]So you insert some you know noise and then you're trying to attack AI by doing so and then therefore the AI can you can essentially detect loopholes of these models and then make it more attack proof. So you know the future of attacks you know I I believe that it's not going to just be limited to DOS right like it's going to be attacking AI agents more importantly like a network of AI agents like autonomous AI network such as Bit Tensor for instance right so and that's what we're trying to do uh to really make sure that the autonomous AI you know network is uh bulletproof essentially >> interesting okay I didn't quite get what it was but we'll let's let's put that on the shelf for now I want to talk about Bosense primarily today I So mostly because it's the one I understand the the most. Right.

[00:01:56]>> So what why don't you tell the people what what is Vosense? What are you doing with that?

[00:02:01]>> Yeah. Vosense I mean it's all purple by the way. So Vance as the name suggests is is voice AI. So it is uh you know bit tensor version of uh 11 labs if you will right it's a decentralized voice AI end to end. We're talking about TTS um you know speech to text uh voice cloning uh end to end uh agentic voice as well as you know text to music I know Mark you're you're big on music so you know through the Bosins platform you can actually get sunol like features essentially you know drop a file do a bunch of things remix and then all that so so all you know we believe that voice AI is not just voice but it's with intelligence meaning hey the voice has to or the agent has to listen the same time you speak. So how do we achieve the the the naturalness is essentially the end goal. Obviously want to achieve soda and then we believe we can do that this this quarter but uh the you know passing the touring test essentially is the ultimate goal.

[00:02:54]>> The voice touring test because I heard Sam say that on his on his pod that he did with you. So I do prepare for these things. So yes I watched Sam's interview with you. Yeah. He goes, "Well, you know, I've been trying to get voice going on the I call it the voice touring test where you know you if you can't tell that you're talking to an AI, right? And you can sort of tell even though the voice it sounds very good, it doesn't quite sound like a human, your ear kind of knows like, okay, that's like a robot."

[00:03:21]>> Yeah, exactly. I mean, you know, along the line of we're obviously uh creating a lot of um you know, infrastructure um essentially to to make that happen. So one of is I think I mentioned that in uh revenue search as well. So it's called style trajectory TTS essentially just to blur out things so that it doesn't stop immediately right like it it has the right you know naturalness essentially like how we talk so so that is coming by the way and uh yeah like Sam is going to be like our version of Will Smith I guess you know eating spaghetti but it's like >> can you fool Sam >> exactly >> that'll be very interesting so yeah so you now on that show you did say you thought you could do it within a month.

[00:04:05]Are you still feeling that confident?

[00:04:07]>> I am. I am. I am.

[00:04:08]>> Okay.

[00:04:09]>> We got top miners. Yeah. Very good miners mining for us.

[00:04:12]>> Yeah. So, now I've used your product and because your product is one, as you mentioned earlier, I am a heavy SNO user. A lot of people don't know this.

[00:04:19]Um I have like this whole other it's one of my hobbies doing rock music. And I use to basically reperform some original tunes that I performed live originally on a on a rock album that me and my old band made during COVID. Right. So, so I I spent a lot of time with with Sunno figuring out how to like twist it and warp it and talk to it, yell at it and make it do what I wanted. And I and I finally got it most of the way to where I wanted it to go. It wasn't 100% perfect, but it was like >> 75% there, right? In terms of the picture that was in my head and what it ended up being.

[00:04:53]>> Um, but I've also used the voice stuff also. So, because one of my other hobbies is writing novels, right, as you can see behind me here. and and I make audiobook versions of them when I do them.

[00:05:04]>> And and I tried and and after I after I did this one in like 23 or 24 is when this one came out, >> um I I I ended up doing the the audio book read myself, but I actually wanted to I I didn't want to do it because it's a lot of work and I wanted an AI to to read it. But at the time, and this was like maybe two years ago, the AIS would not perform the dialogue in a in an even remotely convincing way, like the the the computer read was good enough for like a textbook, >> but it could not do really dramatic reads. Um, and then I I actually took yours and I fed in some dial I fed in a passage. It had dialogue in it.

[00:05:43]>> It was far better than what I had experienced like two years ago.

[00:05:46]>> Yeah.

[00:05:47]>> Yeah. Can we do a demo? Can we do it?

[00:05:49]Can we show it? Sure. We listen. Yeah, let's do it.

[00:05:53]>> Oh, you're gonna do I don't have it. I don't have anything up.

[00:05:56]>> Oh, okay. I was gonna ask you if you have that in front of you. I mean, that's >> I'd have to go find it, but yeah. No, I just was I was just screwing around with it, but like >> it was actually not bad. It was better than I thought it was going to be. So, >> yeah. Yeah. And your your your voice is cloned there. So, like, you know, >> you did clone my voice. Yes. And that is actually quite good. It does sound like me. Even I would say, "Oh, that's me."

[00:06:18]>> Yeah. Yeah. Exactly. I I don't know if you remember like hers like that that movie where like Scarlett Johansson's voice and then Samman or Open AI was going to use her her her name. Uh I mean I think that there are a lot of you know um things you can actually do with the the cloning the voice though the regulation side of things we kind of have to be careful but I think that we're very close in terms of like you know you not being able to even detect whether or not that's coming from you right. So so yeah I think that the model obviously procession is is there that we're very close to soda. Yeah. No, that feels it feels like you've definitely achieved like these kind of clone stuff that you guys done. I did play around with that also. That is that is not close by in my opinion.

[00:07:00]>> But but the voice stuff that you guys are doing does seem to be cutting edge and soda state-of-the-art.

[00:07:06]>> Um so and so you would you would tend to agree and you you did call out 11 labs as sort of like hey we're doing a decentralized version of that.

[00:07:15]>> What is Okay. So what does that market look like? What are people using voice for now? What are people paying for?

[00:07:21]>> Yeah, I mean a lot of agent like uh you know sales calls essentially set up some workflows and I you know start calling people and a lot of it's really not natural. You can tell right Mark you mentioned hey you pick up a phone and someone is talking and the moment you you say something the the voice will stop. I mean you'll know that's AI. So most people use 11 Labs for that and obviously um you know adding voice to videos like influencers doing that. So they will use like uh hickface or uh you know some other you know platforms for the video content and then adding voice using 11 labs. So uh there are a lot of use cases there but I don't think that the local small business are doing that.

[00:07:55]So all you know influencers as well as some of the bigger you know I would say more AI forward uh startups and and and uh as well as companies.

[00:08:05]So mo so so so I think what I heard was that you it's mostly enterprise that are making phone bots that for for I guess inbound for technical support or outbound for sales.

[00:08:15]>> Yes both as well as call center right to an extent.

[00:08:19]>> Um all of those things.

[00:08:21]>> Okay. And um what about what about audiobooks? Is that like a big market or not?

[00:08:26]>> Yeah it is but not as big as call center I guess. You know it is it is pretty big. Uh but I think that the adoption is not there yet. uh it's just like you mentioned the quality is not there yet and then uh people want certain you know accent I that's like British very sexy accent right something right so I think that it's just it's just the quality is not there therefore I think it the adoption is not there yet but I think that once we reach you know you know good enough quality I think that naturally that that will follow but for now it's just mainly you know outbound calling you know setting up workflows as well as adding voice to to video content >> what does the adding voice to video content look like is it just sort like narration over the top of a video >> narration or just like you know uh an AI app or like influencers for instance right like you know doing something and then there's that voice component a lot of people use 11 lab for that uh so yeah I mean you're you know I think that everybody now is by by now is consuming a lot of AI um you know influencer content without knowing it and then the voice there is u chances are generated you know through 11 labs >> so I could just basically produce a video and then like, you know, instead of me actually having to go through the trouble of actually narrating something for an hour, >> uh, I could be like, "Here you go.

[00:09:40]Here's the text. Just read this over the video."

[00:09:42]>> Exactly.

[00:09:44]>> You got the sounds like me. Yeah.

[00:09:46]>> Yeah. Exactly. So, that's kind of how people set up these uh autonomous workflows that they don't have to do the work. The agents are essentially just plugging in everything, fetching data from APIs and again 11 Labs API, not just like the platform, right? So like it's all plugged into for instance cloud uh you know whatever like that they they have uh that that they use locally and then boom you get the the final products.

[00:10:08]>> Yeah. Yeah. I know and for a lot of people who don't know this and I I know this because I've done a bunch of audiobook work. Um after about like three or four hours like you cannot continue. Your brain just starts locking up >> and you you have to just stop and come back tomorrow. It's it's sort of like and and and not doing ums and o's like it's very it's harder than it looks basically, right?

[00:10:31]>> And I remember I remember when Jason Calcanis and I knew this because I'd done several of my books before as audio books. Um, and Jason Calcanis came out with his book Angel, right, that he got published by Harper Collins. And um, and then he had to do the audiobook version of that. And he was like, he was tweeting about the experience. He was like, "Oh god." And like first day he's all happy and stoked up. By like second or third day, he's like, "Oh my god, this is like harder than it looks." And like I'm like tweeting back. I'm like, "Now, you know, >> it's it's it's a real limit. It's a real sort of like your brain just gets too tired to do it.

[00:11:06]>> And uh it it is a real barrier to production. So if if you but a robot a robot voice version of me could go for 12 hours without any breaks, >> right?

[00:11:17]>> Nontop non-stop. I mean you will lose your voice. I mean I haven't done that myself. So I'm I assume that it's hard in your brain obviously your voice as well, right? Like you'll lose your vocal essentially for for some time.

[00:11:27]>> Uh I mean yeah we're here to address that that problem as well. And it's really just like help people scale when it comes to you know whenever you know that they they need a voice essentially AI or a clone version of how they sound right. So so that uh yeah I think I think that you know it's great that you're you're you're mentioning this because there's that physical limitation as to how human can do this and that's why you know AI is here to really scale that right. So >> yeah sort of analogous to have you seen the figure robot ch uh versus the human.

[00:11:57]Yeah.

[00:11:58]>> Yeah. Yeah. It's great. It's great to see. By the way, I'm like, why don't we add voice AI to that? The robots can talk right to each other as well as asking questions. I feel like that's another, you know, obviously frontier that we can explore.

[00:12:10]>> Yeah, they'll they'll do it. So, I'm an investor in figure, so I I I'm always cheering for them whenever they do anything. Yeah. So, so I'm like, this is great. So, >> yeah, I need 10 of those, by the way.

[00:12:21]Chores.

[00:12:22]>> So do I. Trust me, I' I've emailed them like the second you got one, I want one.

[00:12:26]Right. So yeah, but they're not deploying them to homes just yet cuz the way and the Brad Adcock, the um the CEO, the way he put it was like, you know, we have to make sure like, you know, the robot's not going to like fall on the dog and hurt the dog, right? Like they have to make sure like all those things are like it's perfectly safe, right?

[00:12:43]Yeah.

[00:12:43]>> So makes sense.

[00:12:46]>> So let's talk about the voice market again for a second. So do you and I you brought up 11 Labs. I don't really have a good sense of like how big that market is and how much of it 11 Labs controls.

[00:12:58]Can you talk a little bit about that?

[00:12:59]>> Yeah, I mean there are like big few big players but 11 Labs itself is like you know about the market size is like uh 11 11 billionish right. So as of now >> so and they're you know obviously say it's leading the pack and there's few other you know mid to you know smaller ones but essentially it's monopoly uh if you will and mystery is like doing one a few things right but I say when it comes to voice is like you know 11 months um and by the way all of their models are closed source like we don't it's a black box you don't know >> right you don't know anything about what's going on there so okay so 11 so it's like 11 billion market and did Did you say 11? So it's 11 11 labs 11.

[00:13:41]Exactly. 11.

[00:13:42]>> Exactly.

[00:13:44]>> It's a it's you know again we're living in a matrix, right? It's like 111. But I mean the the entire voice AI market right now is about you know 20 you know 22 billionish though you know it depends on the definition like what you include right it's like the end to end voice agents or just TTS right as you know speech attack like whatnot right? So, so I think that let's say it's just 22 20-ish billion, but by the end of this this year, end of 2026 is it is projected to be at least 100 billion.

[00:14:16]So, >> I buy that.

[00:14:18]>> Yeah. So, so I guess the the demand is there, right? It's like, you know, hey, we know that this is no joke. This is a real industry uh that is doing serious stuff.

[00:14:28]>> Okay. So, you decided to create this subnet to go and attack it. Um are you charging it for your product?

[00:14:35]Uh yes we are right now. So uh market you know that so essentially it is uh you can you know spend 20 uh sorry $12 for uh 4,000 tokens or $24 for you know 10,000 tokens. Most people will actually use you know I think that $12 4,000 tokens you can actually use it uh for a monthish. If you're like a heavy user you want to do the you know have the premium tier essentially. So all in all like you know if you're thinking about the annual subscription is like $150ish which is again a fraction of what you'll pay uh to 11. Yeah, that was actually what I was going to get to like what what kind of a fraction are we talking about? Is it like one/10enth?

[00:15:14]>> One10enth. Yes.

[00:15:15]>> Yeah. Okay. One10enth seems to be like the law of nature in bit tensor where Yeah. You know like I see I remember number just it comes up. It's like Fibonacci of of >> natural some sort and I'm surprised it's not 111 but yes it's one.

[00:15:30]>> Yeah it's one like it keeps coming up.

[00:15:32]>> Yeah. So yeah. So basically you can go you can you can overpay by 90%. Or you can use VSense and and get you know the the quality is comparable. Would you say >> it is? It is.

[00:15:43]>> Yeah. Okay.

[00:15:44]>> Yeah. And I've used look the voices sound great. Everything I everything everything checks out. Like I've used your product and I believe based on what I heard I'm like I don't know how I don't know how it could be any better.

[00:15:54]Like it feels like this is like as good as it could get. you know other than the voice steering test like that I you still know it's not quite human >> but but that's coming right so >> exactly exactly where we going to double down again is the infra side of things like the style you know trajectory TTS that's more like the the the backbone of you know how we want to train models moving forward and then the second one is really just the agentic stuff we want to plug in bunch of things and Mark you mentioned that alpha gap towel is like the platform that you've been using nowadays we're talking to that team by the way we want to create their version of the the agent plugged into hours. So essentially you can just tag it and then you can speak to it be like hey what are the top gainers and losers of the day and then give me a quick run uh you know rundown of like you know uh discord discussions or conversations for this particular summit whatnot like all of these sentiment analysis trending all of that can actually be done through us I don't want to give too much alpha but like we are working actively you know really proactively with other subnets so that we can create you know many many many voice agents based off what they you know what they offer Right. So, I get like a morning report.

[00:17:00]Like, by the way, I am using Alphagap now and I am paying for it. I did go for the paid version.

[00:17:04]>> Yeah, I love it. I love it so far. So, yeah, I think it I think they're on to something.

[00:17:09]>> Yeah.

[00:17:09]>> Um, and it'd be great to to have sort of like a good morning, Mark Jeffrey. Here's here's what's going on with the subnetss, the headlines.

[00:17:19]>> I mean, what would be your your preference? I don't know. Like, what would be like a female? I'd make it sound like NPR, like all things Considered, Noah, whatever that guy's name was.

[00:17:29]>> Okay. Her sort of >> Yeah. I would want >> getting a coffee kind of vibe. Yeah, >> I like that. Yeah. Or like Jeremy Jeremy Irons. I don't know if you know the the act too dramatic.

[00:17:44]>> I know.

[00:17:44]>> Well, the subject >> that'll be fun. Uh he's my favorite. I mean, >> yeah, you can have fun with those things. Okay. So, uh, so you got this you have this product where you can reasonably offer onetenth the cost, same quality. How do you, how do you reach that market? What is the go to market strategy for for starting to steal those customers away? Because oh my god, if they only knew you existed, they'd show up like there's 11 billion dollars waiting to be handed to you. So, how do you how do you go get that?

[00:18:15]>> Yeah. Well, obviously we have the enterprise pathway as well as retail pathway. Let's talk about retail pathway. So, I'm gonna start. I I live in San Francisco. I know a lot of um you know real estate agents. I know a lot of local business owners like they're not by the way, you know, using 11 Labs right now. I want to transition them first. I'm not going to even talk about Bit Tensor by the way. I'm going to just say, "Hey, here's a product probably going to use it. It just works." And then, you know, it's very cheap by the way. So, you only pay like $12 per month, right? And then the you know the things that they can do for instance they can actually be like hey find you know top thousand leads near me and start calling them one by one and report back for instance right like I'm going to give them something that that they can just you know use um and and that is going to be my go to market essentially from retail perspective obviously bit tensor ecosystem is what we need to tap into uh think about this right like 500 AR 500k AR is what we want to achieve uh by the end of this year by the way. So if say uh we're talking about 12 you know dollars per month so you know 144 uh dollars per per year per user um you know times uh 30 yeah about 3500 3500 users all we need to get to that target AR I think we can actually go a lot higher but say you know about 35 100 users I think that if we can tap into the the existing you know users within the bit tensor ecosystem through other subnets obviously right and that's how we can get to that number very quickly.

[00:19:43]Uh so I I think that that the pathway to that is very clear. We're going to do a lot of local stuff like you know inerson stuff as well as tapping into the bit tensor ecosystem. Um the enterprise pathway I do know a lot of big you know uh players who are using 11 labs like enterprise right companies. So I'm going to actually hit them up uh and then we're going to strike some like very interesting contracts uh very soon. So but the first thing is we need to achieve soda. we need to actually really achieve that naturalness that that that marker you're you're alluding to. So I think that there are a few things we kind of have to uh really get get um you know hash out before we we start doing that. Uh but it's very clear uh as guide that what we need to do uh in in terms of uh go to market.

[00:20:25]>> Wow. Okay. All right. That sounds like it's a great answer. So all right so let's talk about your subnet now. Um how what is what is the incentive mechanism?

[00:20:35]What exactly? How does your subnet work?

[00:20:37]What you know uh how much you handing out per day roughly? Is it like 2K 3K something like that is my guess? I I don't know. I can look today but um yeah talk about your subnet a little bit in the incentive mechanism.

[00:20:49]>> Yeah. Great. So so the way that it works for subnet 78 is that miners are tked to create the best models essentially. So the base model that we give them essentially it's one uh 3.6 27B by Alibaba which is like the open source state-of-art. So they have to beat that model obviously. uh by at least 2% uh in terms of all of the nine dimensions that we um that they will be measured against like in terms of like uh accuracy like tempo uh you know uh naturalness like you know all all of these things that that we typically use essentially to to measure the quality of a voice model. So and um um and then the best model by the way is going to be used for our product and then that is constantly be updated right uh because of the bit tensor uh incentive you know structure so the way um and and you know to answer your question um daily I've seen like 5k uh you know the the maximum was 10k so it is actually pretty substantial so can win that is pretty substantial so at some point our emission is actually you know over 1% uh and even more So um therefore the competition is actually pretty fierce. Uh you know people are doing a lot of different things and and doing different you know applying different strategies obviously more compute and all those things right but you know we don't care how they get there. All we care is that they provide the model artifacts and that it you know it is the state of art where it is better than you know the base model or the the other winning models the previous winning models and we're going to use the the latest one for our product and that's how we evolve and by the way the pace at which we're evolving is so fast because this summit got registered um a months ago >> and then we got the product you know up and running obviously mainet and product up and running and the models being like updated every day. So yeah, I I think that I'm optimistic for that reason because like think about how 11 Labs will actually evolve. I mean it's you know they're not releasing models every day, right? They're releasing models I don't know every quarter maybe. Uh obviously not not open source. They're not releasing the model but they're just hey updating the model. So I think that the the pace at which we are improving uh in our back end is uh you know a lot more faster than you know the likes of 11 labs and therefore I'm confident that we can achieve soda this quarter. Um uh but yeah that that's how it works. I will say you know the reward is there is substantial um and people miners are and and and top miners are really fighting hard uh to get that reward.

[00:23:14]>> Yeah. So 5 to 10k a day that's not that's not peanuts. That's pretty good in a day for a day's work. That's pretty great. And you have people from all over the world >> who are who are going after that. Right.

[00:23:25]So we don't know who they are.

[00:23:27]>> Um you know depending on where you live that could actually be even more substantial. you know, you know, in LA it's, you know, it's not bad, but it's, you know, if you're some if there's some places of the world where that's actually pretty awesome.

[00:23:37]>> Oh, yeah. It could be like a month's salary or a year. I don't know, right?

[00:23:42]Like Africa, I don't know how how much they earn, but yes, I mean, in San Francisco, maybe it's nothing or it's something, but it's not as much.

[00:23:48]>> But yeah, you're right. It's uh the I think that the incentive um structure is very very strong for the subnet.

[00:23:56]and and um >> yeah and that is why you know we're attracting a lot of top-notch miners >> and now if you were going to try to do this and you have all two 256 you have 256 miners all sort of competing with each other if you had tried to um achieve the same product without using a bit tensor subnet >> you would have had to uh recruit a whole bunch of audio machine learning engineers first of all you have to find them >> right then you have to you know interview them then you have to hire them and then some of them are probably dead weight so you got to fire some of them. Some of them might sue you, >> right? Like talk a little bit about because I I want to illustrate this part of Bit Tensor because a lot of people don't understand, right? You you could do it the old school way or you could do it the Bit Tensor way.

[00:24:39]>> What does that trade-off look like in your eyes?

[00:24:42]>> Yeah, I mean it's huge, right? I mean the the the difference is very stark. I mean the traditional way obviously like you said, right? Like you hire people, you got to find somebody, right? And then like going through like the resume, stack of resumes and then I've done this by the way for years for like the you know the traditional way in web two right and then you have to work with recruiter and then HR and then you know interview like many many many rounds and decide and that's one person and if you want 200 right I mean that's hard and again like you have to maintain them you have to motivate them you have to manage them um and as a manager by the way in web two like they really don't do any actual work it's like you know especially at the middle management right they just manage people uh which is counterproductive obviously and that's why I guess like they're getting laid off right but so yeah so in in that sense like bit tensor way is a lot more efficient because none of that you don't need to do none of that all you you care is the output can you produce better model than everybody else than the previous version so that's what we care and then just straight to the point I feel like you know cutting through all of these fat essentially right keeping it very lean and Then that there is that one objective function if you will uh that we're trying to optimize for.

[00:25:57]>> Yeah. So that 5 to 10k it's being handed out. So first of all you don't even pay for you you don't pay for bad work like you know if for most of the miners that aren't the winner.

[00:26:09]>> Yeah. Maybe it's not even bad work. It's not it's just not the best work.

[00:26:13]>> So the only thing that gets paid for is the best work of the winners.

[00:26:17]>> And you don't even pay for it. The chain pays for it. Right.

[00:26:22]But that said, you still, you know, it's in your interest to see your subnet token stay even or go up in value. You don't want it decreasing in value. So there there is so some subnets, you know, buy their tokens off the open market and it's far and and they happily do it because the the subnet is their product department and the care and feeding of the of the subnet is keeping the subnet token up, you know, up up and to the right as much as you can.

[00:26:52]>> And that is a fraction that is a pittance compared to what you would have paid if you'd done all the recruiting and the hiring and all the other stuff.

[00:26:59]Again, roughly onetenth. that number comes up again from what people tell me.

[00:27:03]>> Um, so how tell me how you think about your subnet token and are you guys doing buybacks and how does that work in your world?

[00:27:11]>> Yes, we will be right once we get revenue all of the revenue through the products essentially either enterprise or retail or go back right essentially to the >> Okay, interesting.

[00:27:21]>> Okay.

[00:27:22]>> Yes. Yes.

[00:27:25]>> We I mean will you keep none for how are you guys funding yourselves? Have you raised money also or is some part of this?

[00:27:33]>> Yeah, I mean we are obviously so right now the SB obviously is our you know sponsor um and um and then obviously emission is is a big part uh in terms SB is your sponsor. Sorry DSP >> DSP. Oh okay so thank you. Got it. Got it. Got it.

[00:27:50]>> Yeah the other mark the other mark.

[00:27:54]>> Yes. Yes. Uh, and then obviously our um, you know, goal is to achieve at least 500k ARR by the end of this year and then we use all of that money to buy back Alpha tokens.

[00:28:06]>> Got it. Okay, that's a great plan. All right, so can't can't fault you on anything there. Um, did you want to show us like can we hear some of these voices? Is that easy for you to show?

[00:28:16]>> Yeah. So, go ahead. So, I I brought up your site here. I I'll I'll drive but sort of walk us through what we're seeing here.

[00:28:21]>> Absolutely. So, Vosense.ai AI uh you know studio is is live and then you can see voice design you can design your own voice could be a serious male you know with British accent and then you can actually give it a name you can use it later so there are many ways for for you to uh to design your voice and then save it and use it later. So that's one a key feature and if you scroll down I mean there are a lot of uh um yeah so text to speech is another one obviously you can give a prompt and then have the voice that you just design or use one of the ones that already exist in the platform and and to speak essentially to get that audio um out. So I can't hear it though.

[00:29:00]I know you're playing Mark uh but I can't hear.

[00:29:04]>> Yeah. Can you hear Do you hear what I'm clicking on?

[00:29:06]I can't hear. I guess it's Riverside.

[00:29:09]Maybe it is, you know, treating that as a noise. I can't hear anything that you're playing here.

[00:29:14]>> Yeah.

[00:29:14]>> Okay. So, you did some voice cloning.

[00:29:16]This is hilarious. So, you got you got Michaela.

[00:29:18]>> You got Sophia and you got me.

[00:29:20]>> Got you. Yeah.

[00:29:22]>> Yeah. So, so let's So, I'm going to play the cloned and let's let's see what it sounds like. I I don't know if this is We'll see if we can hear this later, but we'll we'll try now.

[00:29:34]Well, that's me.

[00:29:36]>> I can't hear it, but I know how it sounds, right? Because I've done it. But can you add this like your your just like later like when you edit this like because I can't hear it. I think when you play it and then >> Yeah. I'll just I'll just have to see whether it came through or not. I don't know whether it did or not.

[00:29:51]>> Right. Yeah. And then also if you look at, you know, text and music, obviously you're big on music and there's so many things you could do. You can use genai to create the lyrics like you know genai there like you know generate lyrics with AI. Uh by the way I created bit tensor uh subnet symphony or seam thong. Uh I think you did that too with tensor. So we play that here as well but there's so many things you could do like style transfer retake da da right. So a lot of things uh yeah can be done uh through text to music.

[00:30:22]>> Yeah this is great. So we got text to speech and then this is you know you choose I guess you choose your voice.

[00:30:28]Yeah.

[00:30:28]>> Okay, there we go. So, that's where I choose my voice right there.

[00:30:31]>> Yeah.

[00:30:31]>> And that's sort of, you know, we've seen these voices earlier.

[00:30:34]>> Yeah. Um, >> and then the ones that you save can actually will show up here as well, like, you know, um, you know, cloned mark or, you know, special K cloned. Uh, yeah, stuff like that. But this is uh Yes. And then and then here you can type in essentially the prompt and then choose a style and and you know, generate the audio files.

[00:30:55]>> Yep. And then speech to text, that's you know going the other direction where somebody's if you got like somebody talking you can change that into text.

[00:31:04]>> Correct.

[00:31:04]>> That's the voice cloning where you can upload someone's voice and record. Do you ever worry that people are going to like you know you know use this for bad things.

[00:31:11]>> I mean that is always a concern right so obviously we have the disclaimer and everything. I mean that's something that we are actively working on like AI safety and all that but uh yeah for now uh you know you can clone but you know you'll have a watermark essentially in the file. Um, >> yeah, you're supposed to clone your own voice, not use others.

[00:31:29]>> Don't pretend like you're const um cool. So then playlist, by the way, is something that where playbooks is something that we we just added essentially. You can add your own songs here, make it public and share with others. And then we're going to actually do the the Spotify, you know, publish um on Spotify uh feature very Oh, okay.

[00:31:50]>> Yeah. Like sunolo like, you know, essentially experience.

[00:31:55]Very interesting. Okay. Well, we got so we got a good flavor of what you're doing there. That is fantastic.

[00:32:00]>> Yeah.

[00:32:00]>> Um let's let's talk about your other subnet. So tell us what you're doing there. I know you gave us a little flyby, but why don't we go in a little bit more detail?

[00:32:07]>> Sounds good. So 26 essentially is called perturb. uh in the world of computer science this perturbation or to perturb is a term used to insert some noise purposely to um a data set like image or audio file or or video right so essentially it's to try to distort the data set therefore um you know fooling uh AI or models um so and you know as human like when we see a picture of like panda for instance right we'll know even though some pixels are being uh altered or changed but AI because it is if not trained well uh it can know by changing few pixels it can actually see that as a I don't know polar bear for instance right so so that is where the attack vectors can really happen so we're talking about you know adversarial um you know uh attacks on the AI system or um autonomous AI networks so uh you know uh again like I mentioned the future is not going to just be limited to Dtos and all these like traditional attacks but rather AI attacking AI agents attacking for instance AI moderators through really surface bad content right so all of those things are essentially being you know can be prevented and and made um um you know by making AI models and agents a network of AI agents bulletproof so so how do we do that is essentially is through this um this centralized gain right like generative adversarial network so that their you know miners essentially you tasked uh by the validators to essentially try to fool trying to find the attack strategy to fool these top-notch models uh and send back the result and then validators will will do the validation and then be like hey this is a certified attack and therefore we're going to address it later. So so again we're collecting all of the attack strategies through this mechanism. So think of miners as the attack specialists or researchers, right? They're like trying to poke holes of these these AI models, >> the red team.

[00:34:09]>> Yeah. Exactly. Exactly. Exactly.

[00:34:12]>> Well, let's talk about Okay. Uh so let's talk about you because we haven't really talked much about you and your background. Yeah. Yeah.

[00:34:18]>> So you're doing you're doing these two very interesting subnets.

[00:34:21]>> Where do you where what were you doing before this and and where do you come from?

[00:34:25]>> Yeah. I mean I wasn't born yesterday obviously. I've been in space for some time. uh you know the space of AI and and data and crypto for over 10 years. I uh I did a lot of work and you know working as a full stack engineer, data scientist and analyst and you know I was a banker by the way my first job and then I I just remember why I left that world uh investment banker you know few months I'm gone right uh and then um yeah so I did a lot of work in the world of uh web 2. So essentially, you know, working uh for some of the companies like startups uh uh in San Francisco such as Next Door uh headsp space as the first kind of person spinning up the the the AI stack and the team and creating recommendation engines and creating you know feed optimizations and and stuff like that. So and then I transitioned uh into web 3 via this um this project called Avalanche. Maybe you know it's one of the L1's.

[00:35:17]>> Oh sure.

[00:35:17]>> Yeah. So it's the Able Labs Foundation.

[00:35:20]uh I got hired uh uh by the foundation and doing AI work essentially as the first person being a trailblazer and then fetching all the data the messy you know web3 data into the back end and clean up and you know creating dashboards and analytics and then later on uh my team was creating this essentially web3 version of AWS so blockchain as a service so essentially with few clicks of buttons you can get your blockchain up and running you can add a layers and etc. So uh funny enough uh you know a labs or avalanche had this terminology subnets and uh you know and then one day I was doing this like search right I was like hey subnets uh and then boom like bit tensor popped up I was like what is this project and I kind of start you know I I kind of went down this rabbit hole like reading you know white paper and doing bunch of research I just never came out of it I feel like that was the moment I fell in love I was like oh this is fascinating by the way I study math I study apply math for years and years. I was like, I can't make sense out of this. Like I have to like I went back 10 times to really understand. I still don't think I understand it 100%. Um I I just loved it so much that experience and then it really transformed me. Uh and I was like doing bunch of like research on the side and one day actually within a month uh a recruiter reached out to me be like hey have you heard of this project called Bit Tensor? I was like, "Yeah." Oh, yeah. The universe essentially just found me and then um but and then the the the the recruiter was like, "Yeah, the the founder is interesting. Hire somebody like you. Would you be open to that conversation?" Like, "Yes." Right.

[00:36:50]And then within a week, I was able to talk to Cons and hit it off. And then uh doing that meeting, he's like, "Oh, you're hired. Can you start tomorrow?"

[00:36:57]I'm like, "Not not yet. I I can't, but like I have two weeks, right? I noticed." So, so yeah, that's how we uh that's how I got into the the ecosystem.

[00:37:06]And then uh the rest is history. I have so much fun. I feel like, you know, every day uh I feel like a kid in a a candy store, right? Doing a lot of different things, but like combining different tastes and then, you know, mixing them and eating them like, "Wow, this is delicious." Right? Like every day is like a surprise. Every day is something new. Every day is, you know, getting to know folks like you, Mark.

[00:37:24]And then, yeah, it's just a it's it's a really just fun and and uh yeah, just just a a great experience thus far. And uh that's why I'm here. That's why I'm back.

[00:37:35]Yeah, it's a great that sort of mirrors my path here. It is a great and fun ecosystem. Everybody's awesome for the most part.

[00:37:42]>> Exactly.

[00:37:43]>> You know, maybe like 99 but 99% of the people are awesome. Like seriously, which is very unusual, right?

[00:37:48]>> Exactly. Exactly.

[00:37:49]>> And I and I and and it's sort of and I got the same feeling like you did probably, you know, the first time I encountered Bitcoin or Ethereum, >> like you just know that like it's going this way. This is a big deal. And we all kind of got that same little, you know, force quiver, if you will, from Bitensor that we got with Bitcoin. So, that's why we're here.

[00:38:08]>> 100%. So, >> all right. So, I think we've covered everything. Is there any uh Do you have anything you want to add or anything you want to plug?

[00:38:16]>> I wanted to play the song, do something there, but I don't know because >> Oh, you want to do that?

[00:38:21]>> Go for it. Let's try >> because people can't even hear it, right? Can I play the song? I I'm going to >> Okay, so special. Okay, let's let's let's so let's teee up your song. So, we're going to end the show with your song. So, why don't you tee it up? What what are we listening to? And then we'll play the song.

[00:38:34]>> Sounds great. It is a bit tensor steam song with all the subnets out there. Um powered by Vosense.ai.

[00:38:44]>> So, you wrote this song. You wrote the lyrics to it. I did.

[00:38:47]>> And then you used Vosense to create the song. AI created the song. And it's all about the bits or subnetss. All right.

[00:38:55]We're going to close out the show with your song. Here it is.

[00:39:03]>> Signals flow. Miners in the dark turn thought into light. Validators watching watching through the middle of the night. Intelligence. Nobody owns the flame. Every sub building something with a name. From language to vision. From code to art. A decentralized mind with a thousand beating hearts. Stake it. Rank it. Prove what you know. Consensus rising, rising while the networks grow.

[00:39:29]No gate to keepers. No eye for thrown.

[00:39:32]The future of AI belongs to the swarm alone star. Hear the single war.

[00:39:38]Knowledge on the chain forever more.

[00:39:42]Digital minds in a synchronized storm.

[00:39:45]Building the future in its centralized form. Sore. Let the sun sing. Rewards for the value intelligence brings from the minds to the bad strong. This is the network. This is the song. No single king like those through the wire.

[00:40:02]Competition spy minds connecting tonight. One chain many spine free but keep it time and open source intelligence. They're coming across the world. The jeep ignite. Millions of minds connecting tonight. One chain.

[00:40:18]Many voices evolving in time, turning raw data into something