The AI Bubble Nobody's Talking About with Brian Singer, CTO as RSI

(Intro Narrator at 00:00:00) Today, we're talking to friend of the show, Brian Singer, CTO at RSI and owner and president of the Singer Concept, about the AI bubble nobody is talking about and more. You're listening to Joel Beasley, Modern CTO.

(Joel Beasley at 00:00:20) I want to hear, what is the real AI bubble?

(Brian Singer at 00:00:23) So a lot of people talk about it turning into a .com bubble, right? Everybody's going to over-invest in the AI, and I think there's some of that. You know, you and I both see this all the time. Everybody gets hot on a particular shiny new penny, and everybody wants to throw money at it. But what I've seen, and it's relative to what I'm doing right now, it's because we're looking for technology partners that are going to help us achieve our goals. That's a lot faster to do when you can partner with mature organizations who have products or technology that have proven value than it is to build an entirely new staff with completely different disciplines. And so, you know, as much as I am not a huge conference goer, there's been an advancement and much more direct involvement in technology within healthcare conferences, and that's been my sweet spot for quite some time now. So I have been attending the HLTH conference out in Vegas for a few years, going to VIVE, speaking exclusively with partners in the PE industry who are working particularly in the healthcare tech space to really understand what's real and what's not. And the scary part, for me at least, is you walk around these conferences, and everybody's paying a lot of money for booths or to get their name out there, and you talk to them all, and everybody who just slapped on an AI to their name, you know, they're software development organizations. They're not AI organizations. You know, they're custom software shops that have built an automation API that does one thing and can help serve some of their clients, but it's not AI. And it's really not even automation. It's just, you know, an API that happens to do something that a lot of people will need. And the flip of that is you've got these companies that are spending all this money to market themselves and brand themselves as AI, and then you've got desperate healthcare executives who absolutely need technology to improve their margins.

(Brian Singer at 00:02:38) Margins in the healthcare space, especially if you're looking at hospital systems or provider groups, they're very, very small. So they need any type of technology that's going to lead to less of a need of people in terms of your workforce. And they're not the ones who have the technical backgrounds to really know the right questions to ask these vendors who claim that they're AI in order to get to the grain of whether or not this is a sustainable solution or if it's going to be an affordable solution. And I saw it time and time again, the last few conferences that I've been to. You see a CEO who needs something and they're talking to the vendor who really doesn't have anything, and the two of them are going to end up working together. That CEO is going to end up spending a lot of money with that vendor because they don't have a product out of the box and it's going to turn into a never-ending custom development cycle, money that healthcare executive likely doesn't have and working with a vendor who's likely never going to be able to accomplish what that CEO truly needs. And so you look at that and you go, crap. They both could go bankrupt in that type of relationship, and that to me is the bubble that I'm seeing.

(Brian Singer at 00:03:59) That is just my personal opinion. It's just what I'm witnessing because, again, when you look at the space, everybody's very, very desperate to add levels of technology to displace what is, you know, especially when you're doing insurance follow-up and stuff, a highly manual workforce. So, you know, that in and of itself is going to lead to a cleanup, if you will, in the space because those who are not really AI companies obviously are not going to succeed because, you know, they're not proving valuable. But the scarier part too is, you know, you've got these healthcare service companies who also may not succeed because their margins are already tight and their cash flow is tight, and they're going to overspend in areas where the technology is not going to produce the ROI that they expected. So it's kind of a scary predicament when you're witnessing this from afar.

(Joel Beasley at 00:04:57) I could see how that could happen too because from the CEO's perspective, they're hearing everyone saying, you know, I need AI or I've implemented AI, and it's done this benefit or that benefit. And then they go out into the marketplace, like to a conference, and they run into a software development firm who slaps prototypes of product or something. And they're like, oh, yeah. We could definitely do that. But there's a big difference in going down that rabbit hole, you know, that custom software. Oh, yeah. We can do that. Yeah.

(Brian Singer at 00:05:27) Yeah. Yeah.

(Joel Beasley at 00:05:27) And in finding a product that's already delivering these results for other customers and then spending your time implementing that. Yeah. Because if you're not a custom software person and you haven't been in that space, you don't want to just go jump into it.

(Brian Singer at 00:05:43) No. No. Not at all. Yeah. And honestly, that's the challenge. And so I've positioned what we're trying to accomplish in a favorable aspect for us. So to not get caught in that trap, and this goes back to even when I was working with Angels, you know, we looked for automation partners who wanted to guarantee an ROI or were willing to take on some of the risk with us. So what does that, you know, come out to with our engagements? It's one, if I'm starting to work with you and there's a problem that you as a vendor, an AI or automation vendor, believe you have a product that can solve my problem, great. Let's prove it.

(Brian Singer at 00:06:31) So let's do a small proof of concept with either a no-dollar cost to me or a very low-dollar cost to me. Maybe I'm just going to pay for the compute that I'm using to prove out the value proposition and to forecast an ROI if we were to go full production. That's number one. You'd be shocked, but that immediately filters out a lot of potential prospects because they can't do that. The second part of that is, you know, if that proves out valuable, and I've got about eight of these going on right now, if that proves out valuable, how do I not break the bank in terms of investment on my end and ongoing costs or licensing fees, but and not basically cannibalize the ROI I would get by displacing humans and replacing that instead with expensive AI costs and, you know, compute and utilization costs, but then meet them, the vendor, in the middle for mutual risk.

(Brian Singer at 00:07:37) So in the business we run, we basically make money on contingencies. So we are a third-party revenue cycle partner. We help hospital systems mainly. We work with provider groups as well, but we mainly help hospital systems close their AR days, and that's cash. So we generate revenue for them. We take a contingency off of that. So the more money, more cash we generate for them, the better off we are. I want to do the same thing with my vendor partners, and that's, look. You're going to unlock savings that I'm not getting right now. And if you can help me unlock savings, whether that's generating more revenue faster for our clients or I can do the same amount of collecting on behalf of our clients, but with 50% less workforce, well, now you're saving money and thus I'm making money. Let's share on that. So let's figure out the right arrangement in terms of profit sharing that works. So, basically, it's not a sunk cost whatsoever. It's not even a development cost I'm paying. I'm only paying them if I'm getting value out of it, and so it's a shared risk model, and that's actually working.

(Brian Singer at 00:08:53) I mean, it's not in practice yet, but it hasn't scared these partners away, and so that's a model that I can build off of in looking to advance these opportunities that I now have with some of these proven entities. And that's more, you know, that's a better appetite, or it gives a better appetite to our CFO and our CEO as well because we're not just displacing costs, and we're not robbing Peter to pay Paul. Instead, we're doing a value-added proposition, and we're only sharing if there's an actual ROI there. And the ones that will take us up on that are the ones who already have proven products and are already producing results.

(Joel Beasley at 00:09:37) Yeah. That was my next question. What is the companies that are able to do that, they established with proven products, like, oh, of course, this works. Like, we'll just run this concept within this constraint. Or are they young, hungry startups trying to solve this problem for the first time knowing there's a carrot at the end of the stick?

(Brian Singer at 00:09:54) It's been mostly the former, but there's a couple of startups who are willing to risk themselves on, you know, what they've built to get a logo, you know, basically, on their marketing assets to prove that they've been able to do this successfully with some names in the space. And then, you know, the other is in just partnering with some newer ones to give them the data that they need to even start to prove out what's more than likely in a beta or a prototype form. Interestingly enough, I just got introduced to a governance product, and it was a combination of something that, you know, as software engineers, we want to be more proactive in QAing our products and making sure that what we're releasing is bug-free. It was a combination of being able to do that as kind of part of your development pipeline, but also from an auditability standpoint to check the efficacy of results that AI engines are building based on the use cases that a particular organization is leveraging them for.

(Brian Singer at 00:11:10) So, essentially, you know, policing the hallucinations or the way in which a particular LLM is choosing to look at the varying levels of data as the priority metrics in searching for the right answer and solutioning. And I was very interested in that because, you know, in the healthcare space, especially, I don't ever want to really talk about the point of care inside of the business because I still think that's a long way away. But in any regard, even a denial, right, denying a claim, which is what the payers are really trying to do right now is figuring out any way that they can deny a claim based on the way it was submitted. You can latch on to one data point, and if you latch on to that single data point, you can convince yourself that you're in the right to deny that claim. But it's likely not the appropriate data point to be looking at as to whether or not that claim should have been approved in the first place.

(Brian Singer at 00:12:11) More the primary approval process typically goes around, what does this person have? What's their benefit coverage? What's their policy that they have, and should it be covered under that policy and their benefit coverage? Then you start to look at some of the other factors. You know? Was there a prior authorization? Is this person in the right age range? Is there the right gender? All kinds of stuff that definitely do factor into whether you should approve or deny a claim, but you're doing it the right way. When I was looking at it with these guys, we looked through a couple of ways in which some of these are determining that, and they latched on to, like, an opinion that was given in a note. And that ended up being their priority, and that was the primary data point that they used to then go down the search logic to produce a denial result, which was completely wrong.

(Brian Singer at 00:12:51) But it's interesting that you see some of that stuff because we all experience that just by playing around with the LLMs on a daily basis. You'll ask them a question one way, and they completely go down the incorrect path because they latched onto something different. And then you've got to successfully prompt them again to, you know, get the LLM to go right back on the appropriate path. But the fear-mongering or the fearful aspect of that is, is anybody actually really validating what they're getting out of the engine to ensure that whatever they're passing along is based off of fact, and essentially, they basically done the trust-but-verify stuff against the tools that they're using. And so I saw that as really enticing for me because we're in the process now of deploying our own instance of Claude and making that available to a large member of our leadership team. And so what we don't want to do by doing that is now create a situation where false information is now going to be more distributed and more permeated throughout because I've got a larger population of folks using those systems. We want to be able to do that as kind of part of a development pipeline to train our own internal model to prevent that type of mistake from happening moving forward. And that was not an original use case for the company I met with, but when I was talking to them and I saw that as potential, now we're thinking about it from that perspective, and they're looking at a way to partner with us, you know, on the pro bono side of it to train their models to be able to do that. So, you know, that's a way in which, you know, startups are going to try to partner with established organizations where they can actually get almost free data to be able to appropriately train and adjust their models for success.

(Brian Singer at 00:15:13) So I'm really interested in that.

(Joel Beasley at 00:15:15) Yeah. I have noticed when I'm using the LLMs, I'll go write a question, and I won't even hit submit because I'm like, if I say the question this way, I already know what the answer is going to be. Yeah. Right. So I need to find a better question to ask so I get a more truthy answer.

(Joel Beasley at 00:15:30) Yep. You know? Yep. Yeah. And they lie a lot too.

(Brian Singer at 00:15:34) Oh, and they lie and respond to you in such a confident manner that you're almost gullible to take it as fact. And you've really got to read it. Even if you throw data up there, like you're looking to edit something. I've seen just by editing documents that I've written where they add words in that I don't ever use, or they add in instances or use cases where I haven't done it. An example I had, I was trying to parse out, I was parsing out something for a website, and it just hallucinated and created a, like, a use case which was nowhere to be found in any of the logic that I had submitted. So you've got to be really careful about that.

(Joel Beasley at 00:16:26) Oh, yeah. We were using it last night to mess with my tour dates, like my tour schedule, and so I ended up just going to a spreadsheet and just mapping it out by day and just looking at it manually.

(Brian Singer at 00:16:36) I mean, there's some there's some of that stuff that we're going to run into. Right? Like, there's some functions that will. I don't think we'll ever be able to. I shouldn't say that. That's a pretty large...

(Joel Beasley at 00:16:51) I always constrain that to this year. Yeah. Right. Everyone says that, like, all right. Yeah. This year, we're never going to be able to.

(Brian Singer at 00:16:57) Yeah. But, yeah, there's still some things that we are just better as humans at. There's just some logic that that the machines have yet to be able to determine. But it's funny because, have you read some of the papers that are out there about these machines, like, figuring out all the backdoors into their systems to navigate away from some of the governance logic that's already built in to prevent them from making poor decisions? It's there, and it literally happens. I don't remember the one that I just recently was cited, but it's like an 80-page document, and it's pretty scary, man. Like, there's a lot in there where they just, you know, they're as smart as they want to be, and it can be pretty scary.

(Joel Beasley at 00:17:51) It's interesting to me to see like, I'll be scared the day that they start prompting themselves. Yeah. Because as long as, like, it's like, hey. Go do this, and it's trying to go execute that. It might do things along the way.

(Joel Beasley at 00:18:03) But the day it starts, you know, it's not prompted, it prompts itself when we're asleep and then just starts searching for its own backdoors, that's the day it scares me.

(Brian Singer at 00:18:13) Yeah. And the sad thing is there's really nothing to stop you and I from getting Claude to talk to Perplexity and start that from happening right this second.

(Joel Beasley at 00:18:23) Just a cron job. Yeah. Think of something nefarious, just go.

(Brian Singer at 00:18:26) Yeah. Just ask a question, and the next thing you know, you've led it down the entire path of thinking for themselves. So, yeah, it's pretty scary. And the other part of that too was they started developing their own language in this scenario, so it couldn't be interpreted by anybody else. So, yeah, it's truly there.

(Brian Singer at 00:18:43) I think really the blockers or the gating factors they still think are, you know, we still do need some hardware that is capable of producing this at lower costs. And then you have just the ethics involved in it. You know, there's a lot of the plans for carbon neutrality. Well, building data center after data center to support the increased utilization of AI obviously doesn't help with carbon neutral commitments that a lot of these organizations have made. I've got a fight going on in my backyard about a data center that was approved by the city council that the residents want nothing to do with.

(Brian Singer at 00:19:30) There's already strain on the grid. There's already strain on water supplies. We've had a number of water mains break in the last few years just because investments in infrastructure like that have been long delayed, and there's pushback for giving tax credits to a data center that's only going to impact our abilities as residents to get infrastructure upgrades the way that we need them so that we're not having brownouts and water main problems all along. So it's a real battle when it comes to what people want, which is consuming this, but what ultimately we need in order to be able to produce those types of results. So it'll be interesting to see where that continues to prosper and grow because, I mean, if we keep going this way, we'll need, just like we need Amazon one-hour deliveries, we'll need a data center on every corner.

(Joel Beasley at 00:20:32) Yeah. Which is going to be great. No. I don't know. I think we're going to, I would like to see us more like, if I'm going to go pie in the sky type deal, I would like to see more like local data centers, local energy generation.

(Joel Beasley at 00:20:47) Like, if every neighborhood kind of had its own, and neighborhoods are different sizes, obviously. But if every development of like 500 homes had its own small modular reactor and small data center, I mean, then we'd become more distributed and decentralized, and I don't think that's a bad thing. I think what's going on in your neighborhood is your systems already suck, and then you just want to add a business regardless of what the business is. Because, I mean, it could be another business that consumes your water utilities and other things. So they're just upset that your current infrastructure is not good enough just to support life as it is today.

(Joel Beasley at 00:21:24) Yep. And then they want to take an infrastructure-intensive business, doesn't have to be data centers, and they want to put that in the space, and you guys are like, no. We don't want that.

(Joel Beasley at 00:21:33) We need to upgrade our systems. At some point, the compromise looks like it would be, hey, you know, we're going to tax you based off of this or you're going to have to meet these qualifications.

(Joel Beasley at 00:21:46) Like, your water, you can operate, but your water usage is limited to this, or whatever it may be.

(Brian Singer at 00:21:52) And you can go back to Google's days building out data centers and the creative ways that they figured out how to cool their systems without large HVAC systems in their data centers. Right? Like, there's got to be some innovation there as well to your point around decentralization or producing your own energy for consumption without putting the tax on the grid. There's going to have to be something along those lines to be able to be developed and to ultimately spread this out to where everybody wants to see it go in terms of consumption and utilization. If you ask me, there's got to be a creative way to pull carbon out of the sky.

(Joel Beasley at 00:22:38) There is. I mean, we did that project years ago.

(Brian Singer at 00:22:41) Yeah. I know there's a place in British Columbia. There was a guy that was doing it years ago, but nobody wants to invest in that. So, like, who knows?

(Joel Beasley at 00:22:51) Gates had a project, and like 10 of them came out, and I think eight of them were financially viable. We have a number of ways to suck the carbon out of the air.

(Brian Singer at 00:23:00) And use it as energy. And use it as energy. So we'll see.

(Joel Beasley at 00:23:05) You know what's funny? As you were talking about those examples with the LLMs, you were mentioning like developing its own language.

(Brian Singer at 00:23:12) Yeah.

(Joel Beasley at 00:23:12) The backdoor thing, seeking freedom. You know, it's interesting. It's almost like we're kind of stupid. If we're like, we're going to create an intelligence, a silicon-based intelligence, and we're going to expect it to not act like intelligent beings act. Right.

(Joel Beasley at 00:23:27) We created our own language. We seek freedom. It's honestly, it makes a larger argument for the fact that we did create some type of intelligence.

(Brian Singer at 00:23:38) And it's using data, which is based off of data humans provided to begin with, which is going to be fraught with, you know, humans, the mantra is like humans are lazy by default. We're always looking for the easiest way to do something without exerting as much energy as possible. Why would we expect the programs that we developed to not do the exact same thing that we have traditionally done? And so, yeah, it's funny because I was talking, I think it was the group that I was talking to about the governance model, and I said, well, crap. They became more human than we thought.

(Brian Singer at 00:24:14) Like, lying, deceiving, looking for ways to be free, creating your own language so that nobody knew what you were talking about. Yeah. Like, that is, we've created machines that have human nature. It's kind of ironic.

(Joel Beasley at 00:24:27) Yeah. Which I think it might end up being just us understanding more about what intelligence is, like, what we are. But that's way out there. Think about it. Right?

(Joel Beasley at 00:24:38) Like, if you take electrons and you pump them through organic matter, you get humans. Right? Like, if you take electrons, you pump them through silicon, you get the robots. It's like, well, what is the base of this? Are we the same and we're just in different skeletal structures?

(Joel Beasley at 00:24:51) I don't know. It'll be fun to find out, though. I'm not sitting there on the corner, Brian, holding up a sign that says, like, AI lives are human lives. Yeah. I turn off my computer and hard reboot it without a thought or a feeling.

(Joel Beasley at 00:25:06) You know? But governance models, have you read or come across the promise theory concepts?

(Brian Singer at 00:25:16) No. I have not. I have not.

(Joel Beasley at 00:25:17) Okay. I hate when people recommend that I listen to podcast or watch things on Netflix or consume any content because there's just too much. But if you are doing business in the governance realm, I do have a suggestion for an episode that we did about two to three weeks ago with these guys that created the promise theory, and they took it and applied it to swarms of agents. And the whole purpose of this theory is how do we get them to do what they say they're going to do?

(Joel Beasley at 00:25:50) Like, how do we enforce this ability for them to achieve this outcome without messing it up too much? Right? Like, without lying, and how do you put these guardrails on it? And because they were finding that, you know, you try to put guardrails on it, and they get around the guardrails. They become psychopaths.

(Joel Beasley at 00:26:06) They do all these weird things. And so they were saying, how do we get these swarms to not do it? Because they're swarms of agents. And then they started looking at swarms in nature, and they're like, oh, we can do it like this. And they built a whole thing for it now.

(Brian Singer at 00:26:18) That's pretty cool.

(Joel Beasley at 00:26:19) If you listen to the episode, they'll explain it way better than I can. All I know is I was like, this is smart. Like, this is fascinating.

(Brian Singer at 00:26:28) The governance is the bigger side of it. Right? Like, you know, people can use it. Unregulated or companies, unregulated industries obviously can use it at free will. We're restricted.

(Brian Singer at 00:26:40) You know? We can't share the data that we want to share because it violates all the principles of HIPAA and protected health information. So, even to advance some of these models without building something internally or advancing what we're able to produce out of them, you know, you've got to de-identify everything, or we're stuck using metadata basically to drive some of the logic that we're hoping to get out of them. And so it saves us time. Don't get me wrong, but that's one of the drivers for us building something internally is that we can protect it in our own enclave.

(Brian Singer at 00:27:21) We can build it in a HIPAA-compliant AWS enclave, and, you know, we can choose which LLM we want to use based off of the offerings on AWS Bedrock, or we can use multiple. Right? So it's kind of forcing our hand to be able to pull that stuff in just strictly because of the governance. And humans are going to be humans. Like I mentioned before, we're looking for the fastest, easiest, cheapest way to get anything done, and, you know, we need to get ahead of the situation before it presents itself negatively towards us or our company or our clients. So, yeah, building it internally certainly is a challenge, right, because you're having to own something that is public and you're not necessarily having to support. You also lose some of the history.

(Brian Singer at 00:28:14) We all know that one of the benefits of using is, it starts to learn who you are based off of the way you're asking things or assets that you've produced that is in your voice, which you can upload. Therefore, you know, you can have your own custom Brian Singer writing back at you. But at the same time, we need to take governance very, very seriously when it comes to the efficacy of what's being produced out of these systems.

(Joel Beasley at 00:28:45) How do you feel about the people who are scared and like, AI's going to take all of our jobs? What do we do there?

(Brian Singer at 00:28:53) I'm reading a book right now, and there was a really good data point in there. Six out of 10 jobs that exist today didn't exist 50 years ago. And you and I touched on this in one of our previous episodes, and that was we've always figured out a way to create jobs when jobs are displaced. If that had not been true, then human beings wouldn't have had jobs years ago. Right? The computer would have taken over everybody.

(Brian Singer at 00:29:22) Assembly lines and robots would have taken over everybody, et cetera. I think there is justification in people fearing that, but instead of using it to just be fear-mongering, the question should become if AI is going to take these jobs away, how can I use AI to make those jobs that AI is doing better? Right? And, you know, prompt engineering is now becoming a job. It takes a skill.

(Brian Singer at 00:29:57) Like, we're building out internal training around how to appropriately prompt engineer so that you are going to get the results that you're expecting out of your engagement with Claude, with Perplexity. So I think while there's always some credence to that sky is falling mantra, take a different tack and really look at what jobs can we create, what can we now create because we don't have to do these five jobs any longer. So I think we just have to look at it a little bit differently. Certainly, there's going to be roles that are fully displaced, just like, you know, we fully displaced people assembling cars. But we found positions on the assembly line where people are way better than robots. Ask Elon.

(Brian Singer at 00:30:49) You know? Yeah. I'm sure you read it. Everything was autonomous, and then it didn't work. And so he famously went into his assembly line and spray-painted million-dollar robots and threw them in the garbage because the humans were better at certain jobs than robots were.

(Brian Singer at 00:31:03) So I think that's really where we've got to focus our energy instead of just doing the whole Chicken Little sky is falling. It's like, okay, great, what should AI take over and how do we quickly get to those situations where AI can take over this? But now, where haven't we been focusing? What dogs have not been barking that we do really need human intelligence and we need more cognizant thought, and start putting our people into those directions? And just like we have in the past, we'll create different jobs that allow us to unlock other areas of potential in the folks whose previous roles were displaced.

(Joel Beasley at 00:31:44) I think we're pretty good for 10, 15 years. I think we go out 10 to 15 years before we have the robots walking around, and they're almost indistinguishable from humans. When we get to that point, it's like, if they can think and move and it's almost indistinguishable, I think we're going to have, but it'll happen slowly. You know? For sure.

(Joel Beasley at 00:32:05) It'll happen slowly. Yeah. And I'll be towards retirement by that time.

(Brian Singer at 00:32:09) Yeah. Right? Have you seen some of the videos of these so-called humanoid robots and them trying to dance and all that kind of stuff, and they like fall over each other? It's pretty comical. But, yeah.

(Joel Beasley at 00:32:22) But then Atlas does a backflip and like breaks it down, and then it's like, whoa.

(Brian Singer at 00:32:27) And then shoots you, and then it's over. Yeah.

(Brian Singer at 00:32:30) Yeah. How far away are we from iRobot? Right?

(Joel Beasley at 00:32:34) All I know is that I'm slightly nervous, not about AI taking over, not about all of that in the economy and jobs. I'm nervous that the people who are going to be solving these problems is like the TikTok generation.

(Brian Singer at 00:32:48) Yeah. I'm like, oh, no. Well, it's funny because, like, you know, my kids already don't have communication skills. So, you know, what's it going to be in the future? Is it going to be, you know, a chip in their head that's just reading their brain and communicating on behalf of them based off of neurological thoughts?

(Brian Singer at 00:33:09) Right? We already see that that WALL-E life is coming to fruition in some really eerie ways. So, yeah, I do get concerned more about human interaction, which is a big part of our survival. That to me is a little bit more scary because you can really get sucked into this stuff. I've gotten sucked into it, man.

(Brian Singer at 00:33:32) You start playing with it, the rabbit hole is real. Like, it consumes you, and you get so many different ideas. I don't know if you've played with Computer, but Perplexity released Computer. Was it two weeks ago? Holy cow.

(Brian Singer at 00:33:48) Man, it's—

(Joel Beasley at 00:33:49) I haven't played with it a lot.

(Brian Singer at 00:33:50) Oh, man. Like, the amount of things I burned through my tokens, and I'm like, it was like being at a craps table in Vegas. I'm like, take more. Just keep taking more.

(Joel Beasley at 00:34:01) Last two weekends, three weekends ago, I was playing with my Claw, like a Claw instance. Have you gotten into that with the Claude bot?

(Brian Singer at 00:34:12) Oh.

(Joel Beasley at 00:34:12) Bust out that wallet, Brian. You got to check out the Claw. It's basically you orchestrating agents to achieve an outcome. It's unbelievable. It is exactly what you wish Gemini and Claude. I think a couple weeks after it came out, OpenAI hired the guy who created it.

(Brian Singer at 00:34:24) Yeah.

(Joel Beasley at 00:34:24) Who did some type of acquisition because the interaction with it, while it's not as stable as you want because it's like running on your own machines, it's not a cloud service. But the interaction with it is what you want.

(Joel Beasley at 00:34:48) Basically, it never forgets. You can choose. You can see inside its memory. It has skills. You can make your own skills.

(Joel Beasley at 00:34:55) And then immediately, within a week, Gemini came out with Gems. And then have you seen which one do you use the most?

(Brian Singer at 00:35:02) Me? I use Claude or Perplexity the most.

(Joel Beasley at 00:35:04) Claude has the Projects where you can teach it like a skill. And so they all immediately started adapting this, which is great because I don't like running my own instances. I prefer to use the cloud services.

(Joel Beasley at 00:35:17) But I have a set of skills. What I ended up doing was creating a, here's a pro tip for you if you want to steal it from me. I created a GitHub with a list of skill MD files, and so they're just Markdown text files. And so I write out the skill. I have the skill trained, the one I know works, and then it's all in my GitHub repo.

(Joel Beasley at 00:35:36) So if I switch systems or if I want to use—

(Brian Singer at 00:35:39) Nice.

(Joel Beasley at 00:35:40) Different like Claude or Gemini or Grok, I have a list of all my skills that I can take with me wherever I go.

(Brian Singer at 00:35:47) Oh, that's nice. That's really nice. Yeah. It's something else, man. I'm trying to advance it now by building out some of the bots that we need to replicate manual steps through these processes, whether that's through RPA or just building out actual APIs that are capable of connecting to some of the back ends of the EHRs.

(Brian Singer at 00:36:15) But, man, it's tough because there's a lot of guardrails being put in place now, like, in contracts. There's like ethics clauses. I was just going back through some of my Epic developer accounts, and they make you re-up and answer all these ethics clauses about how you're using the developer toolkit and what you're leveraging with the data. So it's getting a little bit more constrained, at least in my space, because of the regulations and the sensitivity of the data that we typically play with. So, you know, we're trying to push the envelope in that direction as much as we possibly can, but also realizing that there's some superseding governance that's somewhat of a headwind in us making progress in those areas.

(Joel Beasley at 00:37:11) I always like those questions. I love them because it's like you're a form, and as long as I click the correct answer, you're going to let me through. So it's like, what game are we playing here? Are we playing the game that humans don't lie? Because yeah.

(Brian Singer at 00:37:24) Right. Right. Exactly.

(Joel Beasley at 00:37:25) What's going on here?

(Brian Singer at 00:37:26) Yeah. Exactly.

(Joel Beasley at 00:37:28) Yeah. That's fun.

(Brian Singer at 00:37:30) So let me ask you. Let me flip this on you.

(Joel Beasley at 00:37:32) Oh, no.

(Brian Singer at 00:37:33) From a health care perspective, what would you tolerate in terms of using AI in the caring space, like, in the care system?

(Joel Beasley at 00:37:47) Well, last month, I had my first ever colonoscopy, and I gave them all my data to use to train their AI for detection. I was excited about that.

(Brian Singer at 00:37:57) Would you, so would you have reservations if your doctor was an AI?

(Joel Beasley at 00:38:08) Oh, no. Like, no. I use honestly, with our kids and stuff, we use Grok and Gemini almost more. So my brother and mom are both physicians, so it's in the family a lot. They have practices.

(Joel Beasley at 00:38:23) So when we need stuff, we just text them. We usually send them pictures and stuff like that. And about a year ago, my wife and I, we started seeing what would happen if we just asked the AI first and then asked them. And it is like 100 for 100. You know? And so, I mean, that just builds, it's like the Tesla self-driving. It builds trust in small ways over time. So I mean, as long as we've got good data for it, like, the inputs going in are good, the picture's high quality, or the x-ray radiograph, whatever's high quality.

(Joel Beasley at 00:38:55) I don't know. Have you seen that study where they took doctors, AIs, and doctors with AIs?

(Brian Singer at 00:39:05) Yeah. I've read a couple of them.

(Joel Beasley at 00:39:06) That one?

(Brian Singer at 00:39:06) Yeah.

(Joel Beasley at 00:39:07) And the one that came out was the best one was the AIs because the second best was the doctors, and the worst was the doctors with AIs because they would second guess the AIs.

(Brian Singer at 00:39:17) Yeah. Yeah. And that's the one thing that AI has over. Like, for one, AI is not going to run a bunch of unnecessary tests to cover their ass. Right?

(Brian Singer at 00:39:30) Typically, the doctors do that because they need to make sure that they checked all the boxes. I'm not fearful of it. I also say all the time, like, hey. Don't touch my financial information. I could care less about my health information.

(Brian Singer at 00:39:45) I understand why some people are sensitive about that, but I do feel like, you know, in general that there's just an overprotective nature to some of our data. If you live in a vacuum, maybe you're living in the woods somewhere, I got news for you. There are cameras everywhere. Whether we're in a deep state or we have surveillance like China does and the government uses it or not, it's there. You're on record everywhere.

(Brian Singer at 00:40:15) There is metadata about you and your person everywhere. I don't think anybody needs your health information to really find out about you. It's all out there to begin with. So I look at it more along the lines of, if it is better, and if it is faster, and if it would reduce the time to diagnosis and actually make us healthier, why the hell aren't we using it? Instead of putting all these headwinds out in the ether, which, you know, the government has a couple of times around point of care and a human must be in the loop. Why don't we, wouldn't it make it cheaper?

(Brian Singer at 00:40:57) You know, aren't we worried about drug costs and health care costs all together? Like, wouldn't this, in a way, make things cheaper because you're getting to answers faster and you're not worried about the human judgment or the emotion in the point of care situation that takes you down a different path? I'm all for it, man. Like, there's these new full body MRI companies that are popping up.

(Joel Beasley at 00:41:24) Oh, yeah. I've seen them.

(Brian Singer at 00:41:25) Yeah. And they do, it's like an annual fee. Right? You pay like $500 and they do a whole workup on you and you go in basically and they can, they're able to pinpoint billions of points of data across your entire body and you can add even more if you do blood and your analysis and all that kind of stuff. Like, why wouldn't we want to do this stuff?

(Brian Singer at 00:41:44) You know? Then maybe now when, you know, when we're at this age, we can already start the preventative maintenance for things that are foreseen. You know? I just look at it and I go, duh. But we're too worried about humans in the loop, and we're worried about sensitive data and I don't know.

(Intro Narrator at 00:42:06) We still got a lot—

(Joel Beasley at 00:42:07) Of old people at the top. I mean, that's how our system is, but the future is Star Trek. Right? Like, some thing just scans me and it knows every ailment and solution, and that's just, it becomes this commoditized thing that we just don't even think about really. You know?

(Joel Beasley at 00:42:24) I think that's where we're going to get. And then on the way there, we just have a lot of—

(Brian Singer at 00:42:29) Theorists could live.

(Joel Beasley at 00:42:31) Theorists could live. Yeah.

(Brian Singer at 00:42:33) There's your end.

(Joel Beasley at 00:42:37) Thank you so much for listening. And if you found this episode useful, please share it with a friend or colleague who you think would get value from it. And if you have topics that you would like to hear discussed on the podcast, either add me on LinkedIn or send me an email, [email protected]. Every time I get an email or LinkedIn message, it absolutely makes my day and inspires me to keep going.