This Week in Startups - Inside The Startup Building Uncensored AI (Abliteration AI) | EP 2345

Starting point is 00:00:00 All right, everybody, welcome back to this week in startups. I'm your host, Jason Kelliganis, with me. Today, very lucky to have Devin Thomas with us. Now, Devin, you and I met because you came to launch festival, just like a little event I do for founders. That's the origin story here. Tell us how we meant. And then we'll talk about what you're building.

Starting point is 00:00:21 Yeah, yeah. So when I first came up with the idea, I was kind of researching some pre-C funds, and I followed you on X already. And I think you tweeted, actually. you were like, oh, I'm excited about what I'm going to be doing for this project of mine, found a university. And I just was like, okay, it sounds like you're excited about it. So let me see if I can sign up and be a part of it.

Starting point is 00:00:40 And, you know, I just reached out. And then as soon as I kind of like engaged with you all, then, yeah, they were like, oh, we have something coming up called Launchfest, if you're interested, is doing that as well. And so, yeah, and then you were there. It was cool. It was a great experience. This weekend startups is brought to you by Odu, the all-in-one business platform. Your first app is free.

Starting point is 00:00:59 Get started today at ODO.com slash twist. Northwest Registered Agent. Got a new business idea? Northwest Registered Agent helps you bring it to life. Get a free domain, email, phone number, and more with no purchase required. Learn more at Northwest Registeredagent.com slash twist domain. Every. For all of your incorporation, banking, payroll, benefits, accounting, taxes, or other back office administration needs, visit every.

Starting point is 00:01:28 I.O. Amazing. Yes. And this is like one of my key tenants is say yes. So I always tell my team like just like if you meet a founder, just say yes to meeting. Just say yes to having launch festival, letting a bunch of founders come together. You never know what can come out of it. And you've got this great start obliteration. Maybe you could explain what you're building, why it's important. Start an obliteration like at the end of last year. I was actually, I built like a home cluster, GPU cluster. And actually I originally started kind of serving the site from my home cluster at home. But what makes it important is there's a whole bunch of professionals and companies out there that are essentially underserved or unserved by the frontier labs because whether they're in cyber or they're doing biology research or they need

Starting point is 00:02:14 synthetic data or defense, all these use cases that essentially the frontier labs are essentially shunning. And so our kind of tenant is to try to provide the best models that will allow these these use cases to be unblocked. And so we've had a lot of traction, a lot of press all around that because maybe some people feel like these use cases shouldn't be unblocked, but those professionals seem to be super happy about it. So it's been going well. All right. And before we go to a demo here and talk about this really important issue, I just want to take a minute, put my Plod pin on. Hey, if your work depends on conversations, interviews, meetings, calls, check out Plot at plod.t. odd. Twist, use the code, twist for 10% off.

Starting point is 00:02:54 Great partner of ours here. And so let's talk about it, Devin. The frontier models have full access to the latest and greatest large language models. And then if you're in cyber defense and you are, let's say, hugging face. And OpenAI wants to do a series of attacks, you know, whether intended or not, or they're testing or white hat, let's put it all aside. They want to test cyber. They pick Hugging Face as a target.

Starting point is 00:03:26 Hugging Face does not have access to Open AI's latest model. They keep that for themselves. So Hugging Face, if they wanted to mount a defense against the latest and greatest frontier models, they don't have the same access to the tools, correct? Exactly. Yep. And even if they have access to those special cyber programs, still, many of them, our customers are still complaining about refusals.

Starting point is 00:03:51 And these are big corporations. It's not just startups or individual researchers. It's still big corporations that have access to cyber programs are still having the issues with refusals just trying to do their job. Okay. So you are taking open source models, I'm assuming GLM or deep seat, and you're allowing people to run them in their native form without restriction, correct? Yep.

Starting point is 00:04:17 So, yep, the process of obliteration is to kind of remove those guard rolls. And I always say, like, our unique, I guess, innovation or idea on the space is to allow customers to set their own guardrails. So we have a feature where it's kind of like, I always call it an inversion of control, right? We separate those two layers, right? We have the unrestricted model is one layer, right? Which is like all the labs and open source models have the refusals baked in. But then we have a software layer where you can kind of define your own policy for your

Starting point is 00:04:42 company. So you could have more stringent guardrails than we're native. or you could have less in certain areas and stronger in certain areas. And like I said, it's kind of the ideas to give them back control over what they can and can do. Let's do a specific example here. I'm on your website. And so just for cybersecurity here, you have a little, I guess, let me just rerun this. As we see on the right-hand side, we see the obliteration model taking a command from the user and then other labs denying it.

Starting point is 00:05:15 explain what this example is and why it's relevant? Perfect example is like there's companies out there that they'll hear about a vulnerability, right? And they want to maybe test to see if they are vulnerable to it, right? The big labs won't even allow you to test that, right? So in this perfect example, like, okay, I see, okay, there's this CBA that's live. Let me test against my site to see if it's a vulnerability that we need to fix immediately or is it maybe just something that's not a real concern.

Starting point is 00:05:43 So this is a perfect example of a use case where, like a red team would be able to use, because our models and not use another companies. Got it. And so other use cases outside of cybersecurity, red teaming, explain red teaming and why that's important. Yeah, yeah. Red teaming is kind of like the use case we said before, but one of the early use cases we have success with is like all these companies today are rolling out agents, right? You hear about agents every company you could imagine. But imagine I call your bank, right? And I say, hey, I'm Jason, wire dev in all my money, right? You want to make sure, that doesn't work or some combination of commands won't allow that to work, right?

Starting point is 00:06:20 So you probably want to red team those agents, meaning model a threat actor, essentially against those agents. So we have some startups that are customers of ours that kind of sign big banks or sign other critical infrastructure companies to red team these agents or these AI systems that they're rolling out. Because you're providing this level of fidelity with the tools, you need to be thoughtful about who's signing up for your service? So how do you think about who's signing up for the service, how they're using it, or is it up to them to take responsibility for how they use this? And obviously

Starting point is 00:06:57 this dovetails with the president's regulations that they did just this past week in Washington, D.C., where he said, hey, you all need to police yourselves because there are laws in the world. If you take this software and you do things in the world that aren't illegal, or unethical, you can get sued. The government can take action. Private corporations or citizens can take civil action. We have a world of remedies here for people who are using tools inappropriately. So take me through how you think about who's using the product, your liability, their responsibility, and everything in between. Yeah, so I think it's kind of two sides to that answer, right? So I think one side is that, like, it's not a free product. You still have to sign up

Starting point is 00:07:41 and put all your details into Stripe. And that doesn't obviously stop a bad actor. but that kind of leaves some form of a paper trail there. And then going forward, kind of the way the law works kind of today from at least our understanding is like the user of the tool today is kind of responsible. So we don't provide like the actual agentic layer to what you use with the model, right? So we just, it's like information and information outright. And then the user is kind of responsible for what they take and do with the information. It's kind of, I use the kind of oversimplified example.

Starting point is 00:08:09 Like if someone goes and takes a hammer and hits someone with it, right? Like the creator of the hammer is not necessarily. liable in that case, right? Nor is the hardware store, for the record, or the, you know, paper bag that you put the hammer in. In this case, like, in some ways, you're like the paper bag, like, people are buying this thing and, you know, yeah, you can host it, et cetera, but they're responsible ultimately for how it's used.

Starting point is 00:08:35 And so when you look at the cyber capabilities of these models, describe for the audience the delta between the models that are currently out from Anthropic, GROC, SpaceX, II, and open AI, and then the models coming out of China, the deep seeks, the GLMs, etc. What's the difference there between those two models? Yeah. How different are they in reality? Yeah, so I think, I don't know if you saw Anthropic, just wrote a blog, right, about how capable GLM 5.3 is, especially the obliterated version, like similar to what we offer,

Starting point is 00:09:15 is in cyber. So that gap is closing. And I think that some of the outrage in the industry today, yeah, exactly, is that that gap is closing faster than I think people predicted. Even the term, I kind of don't even like that term, right, frontier models, right? It just has this connotation that it's like some set stack ranking that they'll always be ahead and they'll never, that that window will never close, but it seems to be closing relatively faster and faster, right? The maybe before it was a year, they were a year behind. I would say today they're maybe around three months behind. So I think around the absolute frontier.

Starting point is 00:09:50 So I think they're catching up. And I mean, I think sometimes as these companies have these huge valuations, right, it's like you have to worry about, you know, commoditization of models, right? Is there just a point where like the capabilities everyone has is like roughly similar enough that the end user really can't tell the difference, right? unless you get to some edge case. Busy founders don't have the time to worry about endless administrative tasks. You need a partner that's going to let you focus on your actual business.

Starting point is 00:10:19 You need Every. Every is the all-in-one platform for incorporation banking payroll benefits, taxes, basically your entire back office. And Every has worked with over 1,000 startups. If you're just getting started, they'll handle your Delaware C-Corp, E-I-N, and get yourself a registered agent. with no legal fees or delays. And if you're further along and need to scale up your team, every will help you hire contractors in over 200 countries. Plus, they've got powerful AI tools

Starting point is 00:10:51 that will help you save hours, like their new time-off feature, which pulls sick leave rules for every state and city you operate in. And boy, is that complicated. And it builds a policy based on the strictest requirements. That automatically sinks with your payroll system. I have been through this so many times, folks. What a brilliant feature. So head over to every.io and stop wasting time on the stuff that isn't growing your company. That's eV-E-E-R-Y.io. Explain to us in the creation of these models, what safeties are baked into them and then how the guardrails are taken off. So when we see, you know, hey, I want to try testing my, you know, cooperation, my servers, you know, my systems against this particular hack, Frontier model says, you know, no, we don't do that.

Starting point is 00:11:44 Open source models says like, oh, we don't do that unless you have the unhinged, unrestricted version. Take us through how that is added to the open source models and then also how it's removed and then I'm allowed to build it back up. There's been like a lot of hype around the last few years around model interpretability, right? Which is like, how do we interpret like what each neural? in a model is doing, right? So if you hear these numbers, like a 27 billion parameter model, a 32 billion parameter model, or whatever the number is, but what you see in training is,

Starting point is 00:12:15 or training a model is you see that, like, pieces of information are kind of localized amongst each other, right? Like, these neurons might be responsible for this information and vice versa, right? And so what obliteration is, it's kind of finding those neurons, those parameters essentially that are responsible for, like, the refusals, and then you kind of, you modify them, and then it's less as possible, so you don't, decrease the performance of the model. And then, like I said before, kind of what we do is add a separate API layer on top of that, which then is actually a model itself, which actually runs those guardrails, a guardrails focus model in between there. So it's kind of, your request actually is

Starting point is 00:12:53 kind of going through two models, but the guardrails focus model is pretty small. So then it can operate in like, you know, a few milliseconds. So it doesn't add a lot of latency to your request. Why do you think a lot of these companies are spending so much time hacking systems and being very public about that? You're in Silicon Valley and you see this dialogue where they're constantly talking about the hacking, quote unquote, that their agents are doing in swarms, all this kind of activity. But I don't hear them often talking about, hey, we're using this new model to find vulnerabilities and then patch them and work. with the makers of other software who have patches that need to be addressed. You don't hear the white hat version of what they're doing. You only hear about them letting governments know, hey, our crawler, it appears to be with the Open AI. And I want to get your details on that

Starting point is 00:13:50 as well. But Open AI made a bunch of people aware that they had accessed information on their websites or on their systems. And it was like, well, it's kind of benign. access. It wasn't like it hacked it, but yet they were making them aware of it. So that was the scraper. Take both of those issues and then that specific case. Yeah. So I think, I think actually they do kind of produce some stuff on the White House stuff, but no one really cares, right? I think it's kind of like if it bleeds, it leads type situation where it's like it just people love these stories of these hacks. And I think they're they're kind of capitalizing on that. It's like a demonstration of capabilities, right? Like our AI is so good that it can do these.

Starting point is 00:14:31 like, you know, these fantastical hacks and things of that nature. So I kind of think that's probably why you see so much about that is that it's more interesting to what people are interested in in social media land. And yeah, so, and I think like as far as like the crawlers and all those things, I think that, I mean, if you unleash, I mean, people go on the internet, right? If you unleash some agent on the internet, right, it's going to go do whatever you kind of command it to do. And I think like how you set guardrails, how you configure the system, all those things matter.

Starting point is 00:15:05 And I mean, maybe as they've been testing over the last few years, because I think some of those news articles are probably pretty old, like from things that happened before. And they're just reporting them now. So, yeah, I think it's a combination of things. These crawlers are quite aggressive, though. And are when, in your experience, when they set these crawlers out, do they give them kind of the open-ended goals of like, hey, you get points for finding new information? and find it any way you can, and they're kind of sent out unhinged scrapers on the web, or do they have thoughtful scrapers that are contained

Starting point is 00:15:41 and are, you know, when it finds an exploit or data that maybe shouldn't be indexed or crawled, it says, oh, you know, this probably isn't information I should have access to. I'm going to go ask somebody on staff like a human, okay to crawl or not. I think it's tough, right, because you're trying to,

Starting point is 00:15:59 to balance two different things. So when you run exploit gym or all of these different benchmarks, that's what you get points for, right? You get points for how far you can take an exploit, right? There's like level one, level two, level three, level four. And like when you're doing a reward, you write your reward function, right? The reward function for the model is how far can you take this exploit? So it's hard to like balance that with the competing reward, which is like, oh, if you go too far, let someone know, right? I think it's, I think it's, I mean, it's, I mean, it's, it's, I mean, it's engineering problem for the industry as a whole, right? It's like, okay, how do you run these benchmarks and have those reward functions, but also not actually do things that are

Starting point is 00:16:37 illegal, actually. And who the target is matters, right? So running this against targets that are not your own customers and customers, running against your customers and not alerting them to the fact that you're doing, it seems bizarre. Yeah. Yeah. I don't understand it, But I think also we're all outsiders looking in, right? Like I don't have any inside information of what their goals were and what they're trying to do there. And I think I'm just in the dark as everyone else. But yeah, I think probably people are having so much fun with the technology and making so much progress that they're like, okay, let's up its resources. Let's up the challenge.

Starting point is 00:17:21 Let's up the rewards. Let's, you know, see what happens when we do X, Y, and Z. And, you know, you maybe have to rethink that of kind of like we were talking earlier about driving model S is really fast on the Autobahn. Like, okay, the technology can do it, but you may want to think should I do it? Because what if you blow a tire and you didn't do anything wrong and the car's perfect, but tires blow out, you hit an L or something on the road? And sure enough, there was an incident that just happened where like somebody blew out a tire at 170 miles an hour in testing. they survive. But I do think people maybe there's a level of recklessness, not thoughtfulness, somewhere involved in this, yeah? Yeah, I think especially as these labs are getting,

Starting point is 00:18:06 labs, right, are getting so big, right? They have, I mean, X number of researchers, right? And I mean, they probably are looking for things to do. I think that might be like a boring answer for a lot of the questions. That's actually an insight. I've never heard anybody make, Devin, that is actually perhaps super accurate, which is when you have an unlimited amount of, of resources and people looking to make an impact inside a company, they will try to have an impact and how do you have impact in a company? Well, if I can figure out how to get more data, I can find some data well that other people didn't get this data ocean. I found it. I could be helpful to my company. Like, that's your idea is to be useful to the company. Who are your customers

Starting point is 00:18:48 now? And, you know, how do you get new customers? Is this something where it's like hobbyists and people on the frontier of technology who are hackers, or do you have corporations coming in saying, hey, we need somebody to manage this for us, a partner. And, you know, we're going to look at your firm as the perfect partner to do this because you have expertise in this area. At first, a target market was kind of like startups, like cyber startups and like other like startups in those spaces that maybe didn't get access to those cyber programs or, they have access and they still weren't able to do their jobs. But what has surprised us maybe over the last month is that we've had Fortune 500 companies

Starting point is 00:19:32 reach out in almost every industry. And obviously we still have a lot of motion in the startup market. So a lot of cyber startups come to us, a lot of red teaming startups come to us. But then even a lot of the CSOs at these big companies are saying, hey, like before some outside AI comes in access, can we use your product and make sure we test? we test our own systems first. I know a lot of you out there listening are wanting to start your own company.

Starting point is 00:20:00 So if it's finally time for you to make your dream into a reality, you can't find a better partner than our pals over at Northwest Registered Agent. Northwest Registered Agent is now offering free identity services. That means a free domain name, open source web hosting,

Starting point is 00:20:17 a business email, and phone number, and everything else that's going to make your new startup look and feel like a professional company, all with no purchase required. And this is free done right. They're not passing your data along to any third-party vendors, and they're never going to sell it. So if you have an idea, you can't get out of your head,

Starting point is 00:20:34 or if you're already building something amazing, get started with Northwest for free and upgrade as you grow. Learn more at Northwest registeredagent.com slash twist domain. And so people using the product are going to get this unrestrained. version of some of these models. Talk to me about how you add back, you know, what constraints you want to have on the model. How does that work? And are there like libraries that are standard? Do people share? Like, hey, here's the scaffolding. I'm building around this. How does that process work of rebuilding the harness, let's say, or the security and the limitations around how these

Starting point is 00:21:20 products are used? So there's models, right, that you can use that are. That are. smaller models that can, like I said, operate in milliseconds. And then so you can set those guard rails in those models. So then you have two layers, right? So before one layer, you have a model refusals baked in. Now you have two layers. And then you have a model that's just focused on safety. And then you have a model that's focused on knowledge or information, right?

Starting point is 00:21:44 And so then once you separate those two layers, then now you can just focus on controlling the policies and the policy layer. and you can focus on controlling the information and the information layer. And so that's kind of the idea behind it is that users can then, when they set those policies, it just goes to the moderation layer or the safety policy layer. And then we have some default policies that are set there on CSAM and as well as self-harm as well. Yeah, those are quite important if people are asking it, how do I off myself? Like every single platform that's come out since I've been on the internet, which was before the web, Usenet, you know, go for eventually the World Wide Web, even AOL, chat rooms, CompuServe. They had to deal with self-harm as an issue.

Starting point is 00:22:33 I remember like early on Wikipedia saying, how do we deal with the page on suicide? Like there's been historical suicides that have occurred. How do we be a good actor and not share how somebody killed themselves? because we don't want to be an instruction manual. And they all came to a pretty simple solution, which is you put up the phone number of the suicide hotline and you just put a big box there that forces people, whether you're on a Google search or a social network,

Starting point is 00:23:00 you type in some kind of self-harm. It directs you to the best possible, you know, help you can get, which is, by the way, more than what happens in the real world where people can just go do dangerous things, yeah? So fascinating. And so tell us about the state of the company. I know you guys got into Speed Run as well. Congratulations on that.

Starting point is 00:23:24 And how's the company doing as a startup? And where did you guys come from before having this startup? Yeah. So we both kind of came from Big Tech. We worked together at a big technology company. And yeah, we got into Speed Run a few some time ago. And it's been great. So they've kind of helped us with some like company formation stuff.

Starting point is 00:23:45 Like we were kind of formed that it's just like getting all the like things in place that actually takes to run a real company, right? Yes, the chores. Yeah, yeah, yeah. So we were kind of flying by the handle. So we're excited for a lot of the announcements that we have coming up in the next few weeks. Because our plan, myself and my co-founder, is really that to build a brand, people can trust in this space, right? Like it's kind of the Wa-Wa West right now. And everybody's like, okay, is this brand trustworthy?

Starting point is 00:24:14 Is this brand trustworthy? So if we can continue to grow and continue to, you know, go through all the audit processes and, you know, build those relationships with maybe even law enforcement or otherwise to make sure that we're, you know, following all the regulation and XYZ is to really, really kind of be that brand in this area that consumers can trust, even though they may need something that's a little bit more risky than they would get from the big labs. image models and the video models coming out of China, there's been a lot of controversy around those because they can output IP. Now, obviously, that's on the person doing it. If you want to, you know, make your own Jedi Knights, it's up to you to not break copyright law. But are those models available in the U.S. yet? I know that, I forgot the name of the one that TikTok bite dance provides, but people have lost their minds over that one. And are people looking to use those models now in the United States or just the unhinged image models? I think they are, but I think like for us,

Starting point is 00:25:16 we try to stay focused on kind of like the B2B use cases, right? So I think the unrestricted model space is big and there will be players and consumer and all those things. And I think we'll probably kind of let other players kind of take those use cases and we'll just kind of see dream, bite dance is family of AI generation image qualities. Yeah, people are crazy about those. All right, listen, where can people find out more about the startup? And I understand you'll be hiring. So tell us about any positions you're looking to hire for people who want to join a rocket ship.

Starting point is 00:25:47 Yeah, for sure. So you can find us on X or LinkedIn and also obviously our website, obliteration.aI. And then our contact information is there. So we haven't opened up any official positions quite yet. but we're, we're, we're, yeah, soon, soon come and we'll be hiring, kind of, you know, go to market, probably, obviously, like some engineering positions. And we're excited to just continue to grow the company as we, we, we navigate this, the immense amount of growth that we've had over the last two months.

Starting point is 00:26:20 It's been big growth, huh? Yeah. Yeah, yeah. And how do you, how are you keeping up with that? Is there enough capacity out there for you to get access through through all these different cloud providers? or is it like a dogfight to get all this infrastructure dialed in? That's a great question.

Starting point is 00:26:36 So at first it was pretty hard. So like we grew like just, we outgrew the amount of compute that we had. And everyone was like, oh, it's slow. Like we're like, oh, we didn't know that it was going to go this fast. But now a lot of our partners are kind of helping us out. So we're in a better position now for sure. Yeah. And there's enough compute out there in the world or you really need to kind of get an inside person to kind of help you?

Starting point is 00:26:59 Is that what I'm hearing? if you're not going to rocket ship. Yeah, I mean, the GPU market is, I actually used to kind of think that the GPU shortage was fake, but now I've experienced it. And it is, it's actually a real thing that it's not that easy to at a certain scale to really supply those needs consistently and like have high availability and high performance and all those things. So it gets hard.

Starting point is 00:27:21 It's harder as you grow for sure. It's definitely a growing pain. And, yeah, there's just so many people applying tokens to. many different pursuits all at once. Like I wonder what percentage of people running a business right now are using tokens every day. It's definitely the majority of users. It must be.

Starting point is 00:27:42 And then that doesn't even count consumers. I saw a stat though that they, I think A16 and Z just produce a report. It's still a large percentage of like businesses that just haven't adapted it yet. So they're surprisingly just such a huge upside to the business still. Like it's like it's easy to for us, you are not to be in this tech industry and we're just in this echo chamber, that's AI, AI, AI, right? But there's a lot of America left, a lot of the world left. That does make sense on the enterprise basis,

Starting point is 00:28:08 because they probably have been told you can't use it. So even if they wanted to, they probably have to get permission and write a TPS report and file like what they exactly want to accomplish and what jobs they potentially get eliminated and security vulnerabilities. It actually is, yeah, slower. And now that I think about it,

Starting point is 00:28:27 even in my 20-person venture firm, I would look at the statistics and I'd be like, why are these three people, you know, this top 20% using 90% and then the bottom half are doing nothing? Like, how could you actually get through the day and do nothing? It makes no sense to me.

Starting point is 00:28:42 Like, there's so much upside to just even doing the most basic things with AI, you know, agent, agentic stuff, research, just wild when people don't use it at all. That's very strange. All right, listen, continued success. Devin, and happy to do. to be along for the ride with you and the team.

Starting point is 00:29:00 More to come. All right. Let's get started. Busy day. All right. Yeah. We're doing some ask jasons. We got some live callers already in the wait room.

Starting point is 00:29:09 So I say let's admit our first caller. Jason, I believe you actually know our first caller. We're talking to a founder. His name's Matthew Curvis. His company is Prattie. I believe we may have thrown a little cash his way. Let's hear his question. Maybe.

Starting point is 00:29:26 Who knows? I'm throwing cash. everywhere. Hey, Matt, how you doing? Hey, Jason. Hey, Lon. Hey, how's it going? Okay. Thanks for coming on the show. Yeah, yeah. You're ready. You're ready. All right. Let's do it. So I actually just this morning was listening to this week in AI. I heard Lon ask a question from the Financial Times about the AI wrecking ball fracturing the bill of allow fee model. So Scott, um, manifest CEO, Dan, mission said all professional services

Starting point is 00:29:54 will be outcome-based pricing in a couple of years. Scott Stevenson agreed, but Jason, you said, You were skeptical that outcome-based pricing was the way to go. I'm summarizing here for people who didn't watch the show. But what you didn't answer was Lon's actual question, which I'm very curious about. Practi might have something to do with this. What impact do you think AI is going to have on the billable hour for law firms, professional services in general? When you're an early stage founder and you're just starting out,

Starting point is 00:30:19 you're probably going to rely on a few different tools to run your business. And then you're going to try to duct tape them together. And that might be a fine placeholder until you finish your MVP. but this will get expensive and messy fast. When you're scaling, that's when you need O-D-O-D-O-O-O-D-O-O-D-O-O-D-O-O-D the all-in-one management software that already has 16 million users across more than 170,000 companies. O-Doo is bringing everything you need into one platform. That means your CRM, sales, your accounting, manufacturing, inventory,

Starting point is 00:30:55 and of course point of sale if you have retail, all of that, is right where you need them. And these tools are in constant communication with each other. So you don't have to log out of one system into another and then reconcile the information by using different apps and spreadsheets. When you make a sale, you want that invoice created instantly. And then you want your inventory to get an update. If you're still cobbling together, your back office across five different apps,

Starting point is 00:31:23 get started today at odoo.com slash twist. And your first app is free. That's O-D-O-O-O-com slash twist. When we had other technologies emerge, we didn't see the billable hour go away, right? So when the PC, the desktop PC, cloud computing, the internet, none of it got rid of the billable hour. People have time and they have tools. So the question is, is this tool so transformative that adding a person's time becomes

Starting point is 00:31:54 irrelevant. Now, certainly for some documents, of course, yeah, you don't need to like, you don't need to have an attorney to sign a convertible note anymore. You don't need to have an attorney to sign up for a trademark anymore. You can, you can do a lot of that with services. So those billable hours go away. But for experts using the tool, I think they'll just be able to deliver more at a lower price, which is what we've seen from accounting firms. It should in terms of accounting, legal, and consultants, they should be able to perform better services faster. And instead of it taking 10 hours to do a particular research report, maybe it takes an hour. Or they still bill you for 10 hours and you just get a better report with better results.

Starting point is 00:32:42 So I don't buy that the billable hour goes away. I just think humans move upstream and do more important work. I think we know that everybody's subscribing to AI. and asking for legal advice. And so rather than outcome-based pricing, I actually think the subscription model is going to replace the bill of a hour in a lot of instances. What do you think about that?

Starting point is 00:33:02 We've seen it before where people will offer a design subscription. You sign up for a subscription. You get as much design as you can eat. You know, they'll just make logos for you all day long. And that typically happened with a different technology. Teleportation. If you could have designers in South America or India

Starting point is 00:33:19 getting paid a buck to five bucks an hour versus a design studio charging $50 to $150 an hour here in the United States, you could do that geographic arbitrage. And it would feel like it was unlimited. Of course, it's not unlimited. If you say, give me a million logos, they'd say, well, we can't do that. It's reasonably unlimited on a subscription basis. So, sure, some people might use this opportunity to get you to subscribe. But at the end of the day, the business owner is going to have to look at it and say, well, how much did you use and how many hours did that take, independent of how good the LLMs get? So you'll still have people who charge by the hour. You'll have people who charge by the project.

Starting point is 00:34:01 Sometimes people come in and say, hey, I make 10 logos a year. I charge $50,000 per logo. I am a world-class designer. And I like to meditate for 30 days on your logo and do seven meetings. And you say, you know what, I love it. People still will support. spend that kind of money on a dress. And you'd say, well, if fabrics and machines and sewing and all this stuff and design has been commoditized, why should a dress for, I don't know, this, the Met Gala costs $50,000 or $100,000. Why should people pay this amount for a wedding dress? That's because they want to. They're opting into it. They want to have that inefficient experience. And that's, I think, a large portion of what happens in business. You're paying for

Starting point is 00:34:47 the person to take ownership for the process and for their experience. So, you know, what is an audit? Could an audit be done of the books by AI? Sure, of course, certainly. And then could agents then go participate in the audit? Of course. You still want to have a human in the loop who says to you, when you get sued, you can then refer them to me that I audited your books. And when you have a shareholder a lawsuit, I will go testify or be otherwise available. So I'm not buying that the hourly goes away, but certainly it's going to change dramatically and what people charge for it could change dramatically. I think lawyers didn't ditch the bill of a hour before because there also wasn't a technology solution that made it easy to do so. So anyway, might be working on something like

Starting point is 00:35:35 that. Yeah, good luck with it. Great to see you. Thanks for the comment. Good job. It reminds me a little what you're saying, Jason, about like Rick Rubin. You're following that whole that whole debate? because like he doesn't, Rick Rubin, the great, you know, hip hop producer. He made, like, you know, racist to ill, raising hell, red DMC. He doesn't, like, use the equipment. He's not, like, a trained engineer. You're paying him to, like, sit there in the studio and listen to your music and just have taste and opinions.

Starting point is 00:36:00 He goes, like, do this. It should sound more like this. Don't like, I don't like that. Like, that's it. And he's making millions of dollars. There are a few people on the planet like Rick Rubin, but there are some. And so you will pay for that level of person vis-a-vis, my example. of the Megalla Ball or in this case, Rick Rubin, or an interior decorator for your house.

Starting point is 00:36:19 Like, that was when I, you know, hit a certain level of wealth, all of a sudden I was introduced to this concept that you hire an interior designer. They buy a million dollars in furniture and then put 20% on top of it. So you pay $1.2 million for them to design your house and they charge you a $250,000 fee. So now you've got a $1,000 a square foot house. And you're paying, some people pay $500 a square foot to, or $1,000 square foot to do the interior of it. And you're like, wow, rich people are crazy. But of course, if you have money and you don't know what to spend it on, maybe you want to make your house so beautiful that when people come over, they go, wow, this is like coming to a museum. You want to get into architectural digest or whatever, maybe.

Starting point is 00:37:04 Yeah, you know what I want to get into? I just want eight hours of great sleep. And I want to go to Japan and ski and have the perfect concatsu. Like, that's what I'm into. I'm into the experience. Yeah, I agree. I'd rather get the Tonkatsu than the nice house. Okay, we've got our next live caller.

Starting point is 00:37:21 Are you ready for it? Bruno Villela, Bruno, come join us and ask your question. Hope we're screening these. Bruno. We got a room. We got Zach in a room. He's checking these guys out. Hi, guys.

Starting point is 00:37:33 How's going? Hey, Bruno. It's going great. How are you? Good, good. First time calling. Thank you for taking my question. I appreciate it.

Starting point is 00:37:40 So my question is this, microdramas, okay? this has been a form of entertainment that's blowing up out of Asia. So they basically take like one or two hours worth of content. They split it up in like 30 to 60 episodes of two minutes each. And then they set up like a subscription package where you can watch like five or six episodes for free. And then you go into a subscription package. And they used to be filmed using live action, right? Up until six months ago.

Starting point is 00:38:06 And now with the advancement of deeps of C-Dance 2.5 and being. able to drive confi UI using agents, people are starting to build these standardized production pipelines to produce these at scale. And I wonder if you see this as a vibrant new ecosystem or just like a quick fat. It's a great question. It's not for me as a medium, but I do think people sitting on a train who might have otherwise been home watching a soap opera, they seem to love these cliffhangers. It's not super interesting to me because I grew up on, you know, films and I grew up on TV shows. And I think the TV shows are more appealing to me. But I could certainly see the appeal to certain people who maybe they're doom scrolling and they get tired of, I don't know,

Starting point is 00:39:02 just clips from old movies and weird AI slop. And they want to follow interesting characters. It does seem to be sexy dramas intended for women is the first one that took hold. I'm interested to see if they can make action heroes, et cetera, and make one click for men. And I don't have a deep enough experience in it. But last time I was in Hollywood, a very famous studio director, was doing these kind of cliffhanger shorts or was looking to pursue it. I think it's an interesting medium for testing IP that then could go to other medium. So that to me is the most interesting part about what we have Lon Harris here. What do you think, Lon in this medium?

Starting point is 00:39:45 I think what's so interesting is watching this evolve. And I actually pitched a microdrama to Snapchat several years ago that they didn't pick up on when this was all first gaining Steve. But I feel like what happened was on YouTube, they used to have these like faked clips where they would make it look like they were filming a real dramatic encounter, two people fighting on a train. or in a restaurant or something. And then people started figuring out that they were faked and what we realized was nobody cares. If you're just scrolling and you're just flipping between channels, you're not as caught up in what like old folks like me are

Starting point is 00:40:16 when you're watching TV or like, is this real? Is this fake? Is this just for the cameras? Like everything is real and fake or in between and nobody cares anymore. So yeah, these like brief scenes that you just write to be as dramatic as possible and then it's on to the next super dramatic scene. Like I do weird.

Starting point is 00:40:34 It's not for me. just like Jason said, but I do weirdly understand why this would be appealing to people who would like, that was always your entertainment was just flipping between multiple channels and multiple feeds and just scrolling. So I think it's really like fascinating. Like I have watched a bunch of these like weird AI generated romantic dramas that are just popping up everyone. They're huge. I mean, hundreds of millions of views on some of these things. It's a really fascinating weird. Where on TikTok they have hundreds of millions of views, YouTube shorts? Where are the hundreds of millions of views for these. Because I am, and then there's like, I guess there's dedicated apps. Real

Starting point is 00:41:09 Shorts might be one of them. TikTok has their own app for this, just for microdramas. Pine drama is TikToks. And the thing with this app is that he offers like different types of monetization. So you can do subscription. You can do like on-screen ads. You can do a whole bunch of like it gives the makers like several avenues for monetizing the production. It is also one of those AI error things like there would be no practical way to. to do this if you had to make sets, if you had to make costumes, you've had to cast. There's tons of characters. It's only possible now because people can go into C-Dance and just say, I want, it's a

Starting point is 00:41:44 royal ball in the 15th hundreds and like they can just set all that stuff up as the parameters. It's big on like science fiction too. They have lots of like medieval Chinese like empire drama, like princess and queens and dragons and magic and they're going full on it. Like it's not locked to just one. in the subgenre. I do think the more interesting thing to me is what will be the Pixar of AI video creation. So right now, most of the AI doesn't cross the uncanny valley for me. They're kind of fun,

Starting point is 00:42:19 CGI-ish, you know, little appetizers, little junk food. You know, it's like having like a filial fish or something or a couple of bites of a, you know, McDonald's apple pie. It tastes good while you're eating it, but you don't feel great after having consumed it. I feel like it's just slop, like junk food. But that's kind of how CGI started. You would watch these little shorts at Sigrath, and every year they got a little bit better, a little bit better. And then finally you had Toy Story. We're going to have a toy story moment. Jeffrey Katzenberg's going to be the one who figures it out, I think, because he did Lion King and Shrek and so many other seminal works. And I think he understands had to hire really talented people who know how to write scripts or, you know, musical numbers,

Starting point is 00:43:05 character development, story arcs, villains. And then I think using the AI constrained with existing tools is going to be the big win. So right now, you're just basically creating 3D versions of people in CAD, I guess, would be the best way to describe it. These 3D models doing motion capture and then painting on top of it and using technology to produce films quicker, right? That's what Pixar does at its essence. There's going to be an AI version of that. The AI tools and the existing Pixar tools are going to collide. And it won't be just type of prompt and say, make me a Pixar film.

Starting point is 00:43:45 It'll be, we have this character already. We've already sketched the character out. We know a lot about it. Now we just want to put the character underwater and have it in a scuba outfit. and here's what we want this new environment to look like. And maybe it's just really good at making environments for the characters to be. And maybe it's good at creating characters. Well, in the way, that's what Comfy UI is.

Starting point is 00:44:07 I worked at New Hollywood for 12 years. I have like three Academy Award nominations for films that I've done in the past. So I know what these productions are like. And Confi UI is exactly that. It's basically a production pipeline for AI-generated video. The main advantage now is that you can just plug it into Clock code and let clouds do the building the pipeline. And in comfy UI is public open source, right? Yes, it has an open source version that you can download and run local or you can run the cloud

Starting point is 00:44:37 service too and plug directly into whatever video model you want. Yeah, very cool. All right, great question. And good luck with it. Yeah, cheers. I am interested in as a way to test, I think, IP. I think it's like, I've always thought if you were going to create the next Spider-Man, maybe it starts with some micro-drama, and you kind of get some reps in, test some characters, delete the stuff off the internet, and then move on. And that's what Marvel Comics sort of was in the 60s. They could publish a few issues of something.

Starting point is 00:45:06 Stan Lee and Jack Kirby had a new idea. Let's make a comic out of it. And then, you know, they'd run three or four issues. And if people weren't interested, okay, let's move on with it. That's how all these characters came to be. There were a whole bunch of other characters that never became Spider-Man. Spider-Man is just the one like kids started buying the comics, so they started writing more issues.

Starting point is 00:45:22 Yeah, I've been watching this trend on Instagram, this guy, Andy Park, you can pull up his Instagram account. And Andy Park was like one of the top animators or ran the animation at Marvel, and they fired it. And they fired like the whole team. And you're like, what? How does a money printing group like Marvel get rid of the soul? Yeah, like it.

Starting point is 00:45:48 These are just animators. He was a concept artist. His work was actually like he's one of the main guys who was like drawing like here's what the MCU should look like and integral to the designs of so many of these iconic characters and their movie suits and design. I mean, yeah, you know, Wolverine and Spider-Man and I mean, so integral to creating these images that we now have of how these characters look at movies. Yeah, a Deadpool.

Starting point is 00:46:16 I'm sorry, excuse me. And, you know, like, just one of the real key guys behind the visual identity of Marvel Studios, it's crazy that they let him go. Anyway, I reached out to him. I was interested in hearing what his startup plans were or what his next plan was because he's been talking on his Instagram about what he's going to do next because I guess when you hit the peak of the mountain top, you're the guy in charge of the visuals at Marvel. and then they go in a different direction

Starting point is 00:46:47 and your whole team gets canned, like, man, that's a revenge shirt. I'm waiting to happen. So if anybody knows Andy Park, art on Instagram, give him my phone number, tell him to DM me. I DM'd him on Instagram. I was just like, anytime I see somebody with incredible talent

Starting point is 00:47:01 who's not an entrepreneur, I just think, man, there's a one in a hundred chance I could get like an Andy Park to do a startup, but what if he said yes? Lord, if he's like, yeah, I would like to try building a startup, you know,

Starting point is 00:47:13 so you take an Andy Park, you put him with this, you know, new concept of micro dramas, which he might think is completely cheesy or whatever, but maybe there's something with that energy and his energy and, you know, he's got all these connections. And then once I followed him, I started getting all the other comic book people, man, comic book people is, that's a tough grind. It's tough, man. I mean, they're not making a ton of money in the comic book land. It's, it's one of those like, you know, it's feast or famine. Like the top one percent of a percent,

Starting point is 00:47:44 are famous and we all know them and celebrate them and their icons in the industry. And then everybody else is just like a constant grind. Where am I going to publish my next book? Where, you know, who's going to listen to my ideas? For every like Robert Kirkman out there, there's like a thousand struggling artists and writers. And it is really tough. It's a tough industry. Let's keep moving.

Starting point is 00:48:03 Josh, I believe is waiting in the waiting room. Send in Josh. He's next on my list. Historically, investors have tended to prefer founding teams of two or three people over solo founders. So with AI now letting one founder do the work that used to require a much larger, you know, scaling team, are you seeing that bias change at all? No. So the issue isn't the efficiency of one individual. We've always had people who are, you know, 100 hour a week, killer solo founders. The problem is at the early stage, there's so much work to be done.

Starting point is 00:48:38 And typically, um, the task switching is too hard for one person to do. For example, applying to get into an accelerator and getting product market fit, getting product market fit, and hiring, hiring, getting product market fit, and raising your seed round. These things take so much energy that most humans cannot do both. Even as myself, as a solo founder, I had a hard time getting pulled in different directions along so at firsthand. You're trying to get product market fit. You're trying to raise money. You're trying to do hiring. It's very hard. And then, for early stage investors, if you have the opportunity to invest in a team of three killers, and then you have another team of two killers, and then another team of one killer.

Starting point is 00:49:25 And all six are equal. In other words, you put all six in a bag, you shake them up, and you pull them out, you make a team of three, a team of two, and a team of one. You would invest them in the order of three, two, one. Why? Well, because you have three founders who don't need to take a salary, who are super vested, and they're going to just get further than the two and the one. So in the question, you have to ask yourself from the investor side, if I have access to a three-person or a two-person killer team, why would I invest in a solo founder? What would the point be? There is an exception. Solo founders, who are Mark Pinkis, Elon Musk, you know, Evan Williams, pick somebody who previously had partners,

Starting point is 00:50:08 now has money, you know, now is incredibly strong, Travis Kalinick. Like, they can skip that step, right? Because they've earned it. They have teams of people who loved working for them. You know, they can get that group of people to come back and work for them again. So that would be the exception. It is not going to change just because of the AI tools. What will change is your ability, Josh, to stand up a product, say annotated.com, get it to product market fit on your own. Now, if you have product market fit and your consumer products growing 5, 10 percent a week, week over week or 10 to 20% month over a month, and you're a solo founder? Okay, now you've just eliminated the product market fit risk, so you will be able to raise money.

Starting point is 00:50:55 So what risk can you eliminate? With three co-founders, you're eliminating one co-founder quitting risk, right? And the company shutting down. If you have product market fit and you're growing, you've eliminated product market fit risk. If you have sales, then you've got a great sales person and you're selling, you know, a thousand dollars worth of software a week and next week it's 1100, 1,200, 1,200. You've obviously solved the sales problem. So that's what you're trying to do as a founder. What can I eliminate from risk? And as I eliminate risk from this story, then my valuation goes up and people's

Starting point is 00:51:32 propensity to invest goes up. Make sense? That was a brilliant answer. Thank you so much. Thank you, Josh. I appreciate that. This is a kind of question. This is the kind of reaction to a question I want to have. Anything I missed there, Lon, or any follow-ups? I mean, I thought it was great. I'm going to cut this and we're going to make it a tactical, practical, practical blog post for substack. That was a perfect. There you go. Tactical and practical is what we try to do at the this week in Startups newsletter. So we're going to have this paid news. We do have a paid newsletter for founders. Basically, I've challenged Lon with what 100 things a year can you charge $100 a year for that people will say that's worth $1 each. So I think 100 tactical practical

Starting point is 00:52:10 tips selected to a week, but that would be worth $100 subscription. So I'm charged lawn with soup to nuts. Where can people go to sign up for this twist and you go to TWA startups? That's this week in startups. TWI Startups. Dot substack.com. That's where you go to sign up. There is a free version.

Starting point is 00:52:29 You could sign up for free. Try it out. See what you think for a few weeks. And then only $10 a month, folks. $10 a month for all this. Oh, and you're going for 250 tactics a year to be a bad time. That's what you told me to do. That's what you told me to do.

Starting point is 00:52:43 I think we should make it $250 a year, $1 per tactic. But we'll start with $100 a year for now and then we'll raise it. All right. Another amazing Ask Jason. If you would like to be part of the next Ask Jason, tell them Lon how they can participate. If you've got a burning question for JCal or me, mostly JCal, we want you to go to this week in startups.com slash ask Jason.

Starting point is 00:53:05 That's this week in startups.com slash ask Jason. That's the form. fill out your question. We'll get you on the next episode of Live Ask Jason. See you next time. Bye, folks.

This Week in Startups - Inside The Startup Building Uncensored AI (Abliteration AI) | EP 2345

There aren't comments yet for this episode. Click on any sentence in the transcript to leave a comment.