AI x Health = High Trust REQUIRED

Me Jonathan Dean talk about all the time how it's freaking sus I can't open up ChatGPT to look at what it knows about me. Right. So it's like even kids just got introduced to an AI in their first five interactions. They're even though it's really cool make them feel they have a creative mind.

But one of the epistemology that we use and the reason that we use predictive processing is so you can actually like model the selves within yourselves.

Yeah. So I wanted just to more set that lens. Like that's one of the ways in which I'm now looking at our backlog. I've implemented a prioritization framework and now I'm also thinking about it from those pillars of like B2B software. Any thoughts on that?

All these examples I just gave are like different context of like where you know, a doctor could draw a circle around there and say like just got.

Introduced to an AI in their first five interactions. They're even though it's really cool, make them feel that have a creative mind.

We start with this kind of like question answer like where I can say, you know, where you can articulate your stated beliefs.

I have diving more into the developer experience. Like what if I were the application developer Building an AI health coach for Brian Johnson. Like how what would that experience look like for me? Foreign welcome to the ABCs for building the Future podcast. Today we have myself, Robert and my co founder Jonathan on epistemic me. We bring on entrepreneurs, builders, researchers, scientists, people in the front lines building the future, people doing big things with big stories to tell. This episode we're going to dive into some of our Building in the Open Building in public updates with some recent exciting developments with our new use case in front of us of an AI health coach and we've synthesized some of our learnings and have validated a few different hypotheses and we have a coherent roadmap that we're building for a partner for a hyper personalized AI health coach. And in today's podcast we're going to be going over that in detail, debriefing some of the user research and insights and talking through another use case that is emerging around private healthcare practice from a doctor who is very kind of cutting edge technology that we both know that is very interested in using epistemic me for revamping their intake. And we'll get into that in a sec and then we'll go over actually some of the latest code based changes that we're making. Actually literally PRs that Jonathan has in the queue and we can take a look at that and dive into some thinking about the changes we made in accordance to our roadmap and the use cases we're building for. And we'll also get into more double click deep dives just towards the whole theme of building in public. It'll be me and Jonathan literally thinking organically in real time about what should the feature prioritization be and just making sure that we're aligned as we're executing on this AI Health Coach. Quick update on the AI Health Coach so I had gone to the Don't Die summit a few weeks ago. I interviewed lots of different hypothesized ideal customer profile ICP type prospects there. Landed a soft partnership with the Don't Die team where we're now building a hyper personalized Brian Johnson AI as a new offering for them in their app to help their users get personalized health plans or blueprint protocols kind of plans just get their users healthier. And one of the biggest things that I wanted to share that I learned in my customer research is that there's uncertainty in the Personas involved. So I am interested in getting healthy. I am maybe a couch potato or I'm not as healthy as Brian Johnson. There's a lot of effort to change my habits, a lot of friction to change my habits and my systems. How do I know what the payoff is? So that that's a hypothesis we've had which is people know that they should sleep more, they should eat better, and they should exercise more. But like what's stopping them? Our hypothesis has been if they're limiting the beliefs and underlying that is uncertainty about well, if I change habits, like how do I know the roi, when will I actually get the benefit? When will I look hotter, feel better, live longer? Right. On the other side of it though, my product offerings and service offerings are typically for from less personal to more personal. So like a recorded webinar that's prerecorded or ebook or something like that about meal plans to like group coaching to one on one very personalized wellness plan based on your lifestyle. So those are all good data points that tell us hyper personalization is the key to solve this. And epistemic me is positioned well solving that from a philosophical lens and the lens of creating your model of self tied to your belief systems and those evolve over time. So with that said, we had kind of gotten a roadmap together and I wanted to ask Jonathan to loosely go over that at a high level there's like four big things we want to build to solve the hypers personalization problem for the AI health coach and the uncertainty problem for both the consumer of the AI health coach and the influencer who the AI health coach represents. And then I'd love to talk a little bit more about overlapping use cases with the healthcare practitioner case. Jonathan, do you want to talk a little bit about the big boulders we got to tackle on our roadmap?

Yeah. So I think I'll just jump over to the Excalidraw that has the roadmap on it. Yeah. So with our epistemic me code base, we have an app repository where we have examples of like how to use the SDK. And our roadmap here shows like four different sections of the roadmap in terms of the example apps that we're building. And the first one we have something called a perspective taker app. And it, this is one that had a dean started building that is meant to like take files from a particular individual and then build a self model of them. So you know, like the first he built one of you, Robert, with some of your, the newsletters that you wrote and he built some of Brian Johnson with some of the documents that he built. And it's just a basic example. If I have, if I have data about my user and I want to like bootstrap or start a self model for the user, how can I start? How can I use that to build a belief system? And so that's one important aspect of, you know, and I think we'll talk about the personalized healthcare use case in a second here. But you know, if you think of a doctor, if they have like a bunch of patient information and they wanted to like map it into a self model, that they would use a tool like this. The second thing we've built is this belief mirror chat. And it is a way to just kind of chat as if you would with chat GPT. But then you can see your beliefs mirrored back to you as you do that. And that's kind of a work in progress. We have a number of sample chat sessions that we've created in the past and the ability for the, using the SDK as, as you sort of like submit an answer, it'll decompose it. It'll decompose like a question that Chatbot asks and an answer that a user gives into question answer pairs that we use to extract the beliefs that are mentioned in the answer. Or using the, you can use the SDK both to create the question Answer pairs and to extract the beliefs. But I think where things are really getting exciting is this idea of a dialectic designer. And this is where if you're building an AI coach, um, this is a tool that you can use to like actually train a dialectic. And the idea here is to introduce this new terminology, this new idea of a learning objective so that when you create the dialectic, the, the system knows that you're, that, that the goal of the dialectic is to learn about the user to a certain extent and have a little bit, little mock up here, you know. So this is, this is what the dialectic designer that we're building is. It's going to look like and you, it's, it's really simple because that falls a start. It's just like what do I want to learn about my user? So I asked you that and then say I want to learn about my user's sleep, diet, exercise habits would be an example. Now if we set that learning objective for the, the question and answer session, the dialectic, then like there's this, how do we know that it's done right? Um, and the, the purpose of epistemic me is so that we ask the next best question at each stance so that we, we can incrementally increase the belief systems of our understanding of the user. And then, you know, so how, how do we explore the topic area of say sleep, diet and exercise so that we have a basic understanding of the, of how we want to understand the user. And so the, the idea is that you can trade dialectics to like, you know, basically get, get acquainted with the, with a user. But then you could have like dozens of these dialectics as you want to learn like deeper and more nuanced aspects and like sort of understand them better. And so what this diagram shows is some user archetypes. And so you could, if you're, if you're, if you're building, if you have an app and you have a user community already, you might know of user archetypes. And these could be like people that have specific kind of philosophies or reasons that they use your product. And you can also use the Perspective Taker app to you know, like populate initial data so that you can have like user archetypes that have philosophies defined and initial belief system defined. And then the, on the other side you can say how do we want to ask questions to this user? And we can, we have this concept of perspective cells. And the idea there is that we can have like other sort of advisors I guess might be a good word in the chat that would actually inform what questions to be asked. And so in this case of sleep, diet, exercise, we have this advisor, Brian Johnson and we would sort of like take from perspective taker all his data, create a belief system, a self model for him and he would be influencing the questions that are asked incrementally to understand the user. So it's kind of a simple setup but then you, you then you run it and it basically kind of goes through a training regimen where it will simulate the user. It'll ask the question, a question to the user archetype, an AI representation of that user will answer on their behalf and then it'll continue to ask questions until the top entire topic space is explored, until we have a completion of our or like a completion percentage of our learning objective complete. So in this way you're, you're training the dialectics to sort of be ready to ask to put out in the wild and ask real users. And that's where you can run an experiment and collect real data and then repeat the cycle. You know, kind of like this.

Could I, could I ask a clarifying question Jonathan?

Yeah, of course.

So thanks for laying this out. I especially love the baseline where it's like hey, we're helping you kind of do cohort analysis. And even before you put data out of the wild, this is how we're helping you gain trust of the AI health coach we'll put out in the wild is let's look at your existing user data, let's create models of them, let's simulate dialectic sessions for a particular learning objective. Like I want to learn more about my users beliefs around sleep. What do they believe around sleep? That's one. And so I think a novel thing that I just wanted to shout out is like Jonathan had this great design idea which is in other platforms the learning objective is not so like natural language processing based where it's like I want to learn more about sleep. It's much more configured, higher lift, friction wise for you to determine a learning objective. So Jonathan came up with this really novel approach where it's like, oh, we have LLMs now to kind of do some sense making of a particular statement. We can convert that into a series of dialectics and allow overrides for it. So like question is the middle part. This is still a simulation. Like we've created simulation model AI models of maybe a hundred users to start and then we've refined a trained dialectic and then that experiment part you're saying that's where we decide. All right, we could put this out in wild, get real user data to back our predictions and our model. Yeah.

So for each user archetype and kind of like output, you know, the questions and answers that were asked and you can see whether they're achieving their job of completing the learning percentage or not. And so like what you, what you end up with is just like, here's a set of recommended questions and expected answers from this user archetype. And I want to output this as like a question tree. So like if you just kind of want to release your AI coach in the wild, you know, it'll, it'll primarily stick to that like question tree and anticipate that the user's gonna answer things in a certain way. And if it doesn't, then you come back here, you collect real data and then you can update it to like evolve the question tree so it's more like a constrained approach where you don't have to worry about the AI coach like hallucinating off in some direction. And I think we're gonna need to fine tune that. But I think of this like, okay, so if you, if you run the training session on an example user and you output this kind of like question tree, I want it to look more like, you know, like evaluations that are used and the AI tools and machine learning where you can say, you know, given this prompt to the user like, and then given the expected user responses, it will increase the efficacy of the learning objective, you know, by, by some sort of like quantified amount. And so we have evaluations that say like, yes, this is a useful question to ask at this point in time. And it'd be like, you know, yes, we're, we're sort of like able to predict the user's answer with some amount of accuracy. And, and so you know, to have those, like to have a kind of an evaluation output set that you can use to then know like, know that your dialectics ready within the, the coaching experience is, is the goal here.

Very cool, thanks for sharing that and leveling back up to the, the underserved needs in market that we see. This is like our rev one that will evolve of how we're thinking about providing certainty to the influencer and their development team to whatever they put out in the wild for their AI health coach offering. They're sure of the results that they can get and they're sure of the protecting their brand credibility and keeping that user trust. So cool. Thanks for walking through this. Do you have any, maybe a funny, like, maybe innovative kind of fun question is because me and you think about these topics a lot. I'm curious if you thought about, you know, five steps ahead, what, what this might look like or like what other things, what other cool things we could do from a user experience perspective to help with those pain points?

Yeah, I mean there's a lot, there's a lot going on here. I think in terms of one, I'd like to have just like an interface like this that just makes it really easy to interact with and I think that's like, you know, that's an easy way to use epistemic me and you can sort of like, you can train it and it'll feel like, you know, you're just, it's a familiar use case with your own user data. So that's what we want to get to now backing out from that, you have a lot of interesting things going on. Like you know, sort of how do I, if I have a learning objective defined, how do I ask the next best question in order to like, you know, increase the, the completion percentage of my learning objective? And I think that there's like a number of heuristics that are used to do that, like breadth versus depth searches for instance. But then we also have these tools like the philosophy of the self model and being able to impart an epistemology. And I think that's where there's a lot of different ways to go about this. And so I'm exploring that through like a training system right now where I've, I've pulled epistemic ME SDK into a Python application and created a reinforcement learning tool that calculate these question answer sort of outputs and that we. And so there's going to be a trade off between like do we delegate this to the machine learning harness or do we delegate this to the epistemology within epistemic me in order to like solve the problem of what, what is the next best question to ask? And our goal's always to been sort of to use philosophy for this. But I'm actively kind of seeing like what can we, what kind of separation of concerns should there be between like three and fourth learning, training, harness and like just to make a representation of the user.

That's a cool thought. What trade offs come top of mind as you've been thinking through that?

Yeah, I mean like, like most things with machine learning there's a lot of like just existing best practices that you can use and you know, like say downloading what used to be the OpenAI gem, which is a reinforcement learning harness. And then, you know, like iterating over a question answer pair. A simulated user is a pretty easy kind of, you know, like test harness to set up. And, you know, then you can assign, you can assign reward functions. And so let's say that I wanted to like, so I can, so I can have an LLM on, on like the, my on the app side of this in the Python code, and I can have an LLM on the back end. Where my epistemology is with epistemic me. And in both cases, we're trying to like, you know, kind of prompt engineer a way that we can generate the next best question for the user.

Okay. So not to derail too much further on the product roadmap, just want to double click into some of that to give the audience kind of a taste about how we think about these things, how we think about problem solving for the problem space of like, oh, there's uncertainty. How do we give them certainty? Okay, cool. We backed into that by looking at these tools. We showed this to the Don't Die team and they were aligned. Do you want to move to that next one AI coach design with that workflow? Like the low code workflow?

Yeah, that's the. So that's the fourth kind of step of the example apps that we're building. And this would be once we have the dialectic designer and you can, you can train a dialectic, then you can insert it in here into kind of like a flow that you, you could design a coach where, you know, we have, we had like one dialectic trained and then we want to like, you know, bring the user through. Maybe, maybe it's going to be dozens of different sort of like dialectic interactions. But, you know, this would be the way that you could use epistemic me to design like portions of your coaching experience and then put them together for a holistic or use this to complement. And this diagram is just showing like, here's my user archetypes over here. Here's my principal respective cells. And then here's the flow of different dialectics that we're using to understand the user along these points. And so then this comes together nicely. It's a nice clean sort of like interaction that reminds me of some of the chatbot tools that I've used over the years with like, what you see, what you get interfaces. But the big difference is just that each one of these, each one of these dialectic sessions is specifically trained to learn something about your user. Um, and and to like sort of fulfill that learning objective.

Yeah. Cool. So the idea being like hey, I'm a developer on this influencers team or this health longevity coach's team. I've gone through the dialectic designer. So now I'm confident about the dialectics and then now I want to put together particular user journeys of chain dialectics for a given archetype or set of archetypes. So I could do that to curate the experience for those archetypes. Cool. Yeah, I think this rounds it out nicely. Could you go back up a level to the four? So it's you know, we start with perspective taker. So how do we actually just bootstrap with existing user data a self model. Model of self of the user as well as their underlying beliefs. Belief Mirror Chat, which is reference application in our code base in our GitHub repo which kind of just shows a exemplified dialectic with data how we extract the belief systems, how we could kind of show you what's under the hood if you talk to a GPT. In this case the GPT is our own thing, dialectic designer where you could design and simulate these dialectics to look at prediction and efficacy. And then AI coach designer where you could now be confident about everything underneath the user data. Have we extracted that? Right. And their belief systems. Right. And then the belief systems look accurate and then now the dialectic designer for different archetypes and dialectics for them and then the coach holistic designer. So yeah, for the audience we basically shopped this as a high level roadmap for the don't die team about how we we're thinking about solving their pain point and now we're in a bi weekly sprint cycle where we're just shipping against this which feels good because it's like solid prioritized use case. With that said, Jonathan, do you want to talk a little bit about how you shared this high level roadmap with Victor, our common friend?

Yeah, so Victor's a doctor and he's in the process of kind of building his own clinic and he's very tech forward in fact has a background in software engineering amongst in addition to like all the knowledge about being a doctor and I think you know even at anybody who's been in doctor's office and you had to fill out like all those forms of your personal history and stuff like that. And then you know you usually go talk to people and they like don't even know how you answer the forms necessarily. But there's a lot of information I think you can see immediate benefit if you say, like, okay, we have this information about user data form like the sentover from the previous doctor or something like that, or the test that we had ordered or in. And then you can answer the questions automatically, you know, sort of like. And then, then whatever is left over to answer. Then you can create a question and answer sort of system for them to interact with and answer in maybe their own natural language or something like that. So what you get is both like everything that we have for the data and all the questions that we want to ask the user are now synthesized into a cumulative self model that is, that has a belief system of the user.

If we had to state it as value propositions in Victor's mind, what are the key value propositions for using this?

Yeah, I think I have. I have his exact words, but, you know, in some of our chat recordings. But just understanding, I think both were like, he, he gave this analogy, like, if he's got a patient who's a plumber, then he'll meet them in that. In that world, right? And he'll like, talk about like the sophisticated, know, pathophysiological interactions of the symptoms that they're experiencing in terms of like plumbing analogies. Right. Um, okay, so if you can get a more even a deeper view of like, you know, where. Where is this person? How is. How is this person trying to manage their health? Like, what's important to them? What kind of terms can I use to meet them where they're at? It helps with communication during the doctor session. It helps with adherence to protocols and like meeting the user where they are, the patient or where they are in their own world, you know, all these data points. I might have genetic records, I might have epigenetic record. I might have functional health information. How am I going to synthesize this all and like, meet the patient where they're at? Creating like this, the holistic model that pulls all that together is something I've talked a number of times with Victor about.

One thing I wanted to zoom out on is like, so you mentioned, hey, I'm Victor, okay? I have like, you know, 10, 20 patients a day. I'm running from thing to thing to thing. They're filling out a bunch of data. The patient intake experience. I think a lot of people listening to this will feel that way. Where in particular in America, I would say my experience, where we fill out these pieces of paper, doctor shows up, they barely read any of it, probably because of, you know, incentives in the healthcare system, insurance. And their job is in that very limited 10, 15 minute interaction is make sense of everything about you. Can then give you a plan that you will follow and will make you healthier. And so then the question is, how can we help to optimize that with those set of constraints? And so at this night, me being really well positioned to make sense of that, extract belief systems about the user, create a model, and then actually predict what is the next best thing we could recommend for this patient. That's how we're kind of thinking about it. And so I think from, from both sides, this, there's that uncertainty story that I keep thinking about, but I think this, that also kind of happens when you walk into the doctor's office, right. It's like you're kind of uncertain about what's gonna happen. You don't know. There's some fear driving that. I think that's an opportunity to change that terrible patient intake form thing, which is like, I fill out this thing, you'll probably be tossed in the void. Like, it sucks to even fill it out. It's different at every clinic. That is an opportunity for delighting. And then on the doctor side, they're just running around from thing to thing and that they don't feel satisfied. That's something that I got. Because they really want to make people healthier, but they're also inundated. Right. With all this sense making they got to do. And then the healthcare incentives in the system they're working in. So they ultimately just want to provide the patient with a treatment plan that they know rock solid. This meets them where they're at, they'll get healthier following it.

Yeah, I think I would, I would describe what I presented, what I just did in like less than 15 minutes. And he's just nodding his head and being like, yeah, okay. Um, oh, just like, yeah, we have, we've had a lot of previous conversations about it. And so, you know, I think he just wanted to see how's the software progressing, you know, we talked about. And I think that the specific like, okay, there's a tool set for data ingestion and that populates the belief systems of self model. And there's a tool set for question and answer through dialectic and that will populate the same self model. And again, he develops his own software. So he's kind of like wanting to make his own software experience for his patients. He has some clinic, right?

Yeah, he's, he's squarely in the innovator category of the technology lifecycle adoption curve. Same, same with Brian's team. Like not every influencer has a dev team, you know, so they fit that profile better of where we're at as a product right now. And then on the, on the flip side, Victor, I kind of see him as an innovator in the healthcare space, but on the healthcare side with the technical bent. So those are seem to be strong profiles emerging. I was just looking through my notes for a quote from a discovery meeting we had with Victor a while back. I have the quote. The right health outcomes come from understanding who and what a person believes in and using that insight in order to help them make positive changes in their life. That's fundamental to medicine. I think that says it all. So like we just make all that better and we'll get to the right health outcomes for the user. Cool. So we walked through kind of an update of user research, landed a soft partnership. We're hot on building AI Brian getting into biweekly sprint cycles. We're still open on discovery of other use cases. Just so as we're building this, our architecture reflects something flexible enough to where we can really go to market in health and longevity. Tech focused influencers as a space, knock that out of the park and then hopefully what we build is flexible enough to expand into other use cases like the private healthcare practice use case. So this was great to go over and get that alignment. Maybe next what we can go into is some of the code based changes you made recently. So we presented this to prospects and now one early adopter partner slash customer. And so then it's like all right, well now we got to take the Excalidraw and implement in code. So but you had kind of a monster coding session over the week weekend so I thought it would be fun. So now just for the pod, I've been really putting on my, you know, sales marketing partnership outreach hat and then now that we have an established feedback loop, the way I think about entrepreneurship is like the founder, especially the outbound facing founder, like that's me in this group. They're solely responsible for just getting validation of the hypotheses and getting to a point where we have a fairly high confident go to market motion and a partner to kind of start with and that will help prioritize the roadmap and our backlog and stuff like that. So now that that's in place, I, I am diving more into the developer experience. Like what if I were the application developer building an AI health coach for Brian Johnson? Like how what would that experience look like for me? So I've been getting ramped up on our SDK and I thought it would just be fun to synergize me learning more about the code base, getting closer to it. So Jonathan, did you want to talk a little bit about that? Or maybe we could even go through these PRs and talk through like how you thought about making those changes from roadmap to code base. So you were talking about the belief systems. Go ahead about that.

Yeah, I think in both like the AI health coach and also like a doctor patient interaction that, uh, if we're talking about like the aspect of personalization and really getting to know the patient, the user, we start with this kind of like question answer, like where I can say, you know, where you can articulate your stated beliefs. But one of the epistemologies that we use and the reason that we use predictive processing is so you can actually like model the, the selves within yourselves. And you know, I think like the movie Inside out is kind of a good example that even children can understand where there's kind of like these competing forces. Inside out has these kind of avatars of emotion that each represent like different needs that you have as a person. They, they're, they're constantly debating each other with every like, you know, life decision or something that happens with some of them are going to take override and you're going to be like, you know, like overly anxious in the situation. And that's like one of these avatars kind of coming to life. So what the. When you think about like the biomarkers that we're taking and we take blood and urine samples and like, those are going to be like foundational aspects of both the health coach and the doctor experience we're talking about right now. Um, you typically think of those being like, oh, this is like part of the user's profile, right? Like, these are maybe characteristics that they have. And one of our goals in, in the project is actually to like map those biomarkers to like in individual cells within, within yourself that kind of like modify the homeostatic functions that are happening in your body. You can start to see like the hierarchical level levels of like, you know, a self model and the cells within the cell. Um, and I think this is one of the ways that we can uniquely model like a personalized self model for patients is to interpret those results in terms of the ways that like, they, they fully interact with like some examples of that that I've thought through for the functional health examples would be, you know, like taking metabolic profiles. I think metabolism Is, is a, is one like if we can trace metabolism from ourselves, ourselves through our tissues, through our organs, through our, you know, energy output that our Apple Watch measures. Like we have biomarkers on all that stuff now. And so like, you know, how do we, how do we put that together in a way that's like most predictive and helps us learn the most. And so at each one of those levels there's like questions to ask, data to collect and like, you know, information to learn. And so like, while we're just building tools right now for this kind of like, you know, human question answer interface, I think what I see when I talk through these use cases is that they very quickly, we very quickly get to points of like how can we make the most meaning out of this data? And it comes, it comes down to like building belief systems, you know that the belief systems that are sort of trained and operating within our body that, that are subconscious and we don't even think about.

Yeah, let me try to reflect that back in my own understanding and, and throughout our conversations about this. But I, I think what you said is really important for people to understand. Uh, we're starting at this layer that are very market understandable which is can I have an AI health coach? Just recommend me some habit changes that I'll get a high roi. I, I could trust that. And I do it, I see better, I eat it, have a better diet, I look hotter. Like you know, some of those maybe high level desires and then lower level, you know, I always love thinking about like, you know, if we're successful 10, 20 years down the road, like what we, what we're actually doing is we are able to create models ourselves for every agent on the planet. That's really what Jonathan's talking about. And then another thing that I mean Jonathan always nerd out about is like, okay, cool. Like well if we could do that. A human is simply a self model, which is a collection of self models, these agents at a lower level. So then there's levels in between. So then a nation is just a collective intelligence which is just a collection of self models of those humans which have self models all the way down to cells. So it's like self models all the way down. Like instead of turtles all the way down, self models all the way down. And along all of those points, every one of those agents is making decisions. That's what Jung been saying. I'm breathing all the time and I've been doing that since I was born. That keeps me alive. But I don't Consciously think about it, unless I'm meditating in a particular mindful state. So there's so many other things that are happening in the body that we just kind of take for granted. With emerging technologies and science, we're getting way better at quantifying and measuring these things. So I can nerd about this all the time, but like, what Jonathan's saying that I'm excited about is like, if we land this use case and it's in health, the next thing is like, okay, cool, we've solved the, you know, superficial. We have this personalized exchange thing now. It's like, shit. We could like, add in contextual biomarker data down to dialectics at every level of agency, right. Of those self models. So then back up to like, Victor's use case. And really the health use case is how can I make you healthier based on all of the models of cells that we've created that make up the agents that make you. Did you have more on the belief systems?

Maybe just give an example of like, what that looks like concretely. You know, let's say, you know, you imagine your heart and your lungs and your, your nervous system, your muscle tissue and your blood like at, you know, so we're talking about like tissue and organ sort of like level and like how your heart is going to respond, you know, like while you're sleeping is one aspect that we measure. And like that and I think like resting heart rate, heart rate variability, for instance, these are all things that we can sort of measure about the heart at any given point in time. But you take it in the, in the context of like, okay, let's, let's now do that while I'm doing a breathing meditation, right? What is my heart doing during that, that time? What is my lungs doing? What kind of like metabolism, you know, metabolism is my blood supporting at this point in time. And what, what's my nervous system doing? You can sort of like, you can take measurements of all of those at the same time and put it in a context of like, know, okay, this is, this is sort of like the context of the belief system that my heart needs at this point in time to respond to the breathing patterns that I'm, I'm enacting. You know, like during ZO2 exercise, your blood is pumping a lot, you know, is helping you facilitate fat oxidation. So you know that you're, you're burning fat and your, your cells are using, your, your cells and your muscles are using that for energy. Um, like we have, we have a model of Interaction for our muscle, we have mil interaction for blood. Our, our lungs are, are pumping at a certain rate, our heart is pumping at a certain rate. And all those things are happening while we as a, as a human are experiencing like a, like a net, like what's it called? Rate of perceived exertion. Right. Like I'm tired, I'm tired on a scale of like 0 to 10. And you can correlate like, okay, now here's, here's the metabolic response of my muscles when I feel like an RPE of 6 or 7. Now I'm going to shift into like you say, like some four by fours and I'm going to like really elevate my heart. I'm going to be using different metabolic profiles, pumping different sort of energy through my blood to my muscles. My heart and lungs are doing different things. And so all these examples I just gave are like different contexts of like where, you know, a doctor could draw a circle around there and say like, you know, we really see trachocardia in your heart once you get to like your four by four exercise. Um, but over here, when you're breathing, your heart looks very normal. So let's isolate that variable. Let's understand, you know, so like looking at the body this way and like kind of modeling it, I think it, it helps give like, you know, metabolic context, environmental context and, and like just a lot of meaning that we can sort of like drill deeper into. And, and of course, we're just drawing Starman Strawman here. Um, yes. You know, I think once you get to the level of like genetic antigenic testing and, and modeling all the homeostatic processes that are happening in the body all at once, we can get pictures of ourselves that are just far more complex than we could even imagine. So that's, that's where I really think the cases will naturally evolve as we, you know, sort of meet patient data. And I just, yeah, I wanted to sync up with that, that picture with you so, you know, something we can see the roadmap together for.

We're doing it live. Cool. Yeah. Actually, as you speak about that. So reflecting my thoughts. Yeah, I think I'm seeing where your head's going, which is like, that's the natural progression. Right? It's like, well, if we can give a doctor or a health coach a tool that personalizes a plan, optimizes patient adherence and health outcomes, then the next question is like, well, how do we do that even better? And the answer obviously is get more context, get more data, make it better and so that's really what we're talking about here. And as you were talking through that from a product perspective, I started visualizing, like, really just to add some words to it. It's like a real time, like, if we accomplish this right, it's like a real time360 health view of you, right? Like, I'm imagining this, like, avatar UI where I'm a doctor, I look at this, and then there's like, the human body, it looks like me. And then it's like a physiological model, human anatomy. And then we have color coding where it's like, certain spots based on the context of certain biomarkers. We can kind of have that as a heat map. And then I'm also imagining we could drag a slider to be like, all right, control these variables and then zoom back to a day before. What did the body look like then? What did the heat map look like then? And, and this is all to say, like, by contextualing all the, contextualizing all this and tracking it and tying it to quantified. Like, things like beliefs can be subjective, but quantifying it gives you some numbers. And then we could provide that sort of experience. And then bigger picture, what we allow a doctor to do would be to isolate variables or control certain variables, and then better experiment or test or make recommendations to experiment or test or in the future enable a patient. How are you imagining that? And also for the pod viewers, this is the first time we're actually syncing on this feature, so we're literally brainstorming it live. So that's fun.

If I got a health coach and it's gonna, and I can ask it, like, what's my optimal day look like? And it, it's trained and it can give me a pretty good answer that we can zoom into, like, these different, like, metabolic contexts. Like, our circadian rhythm, for instance, gives us a, like a daily metabolic context, right? Like, where are we within that? And then, like, if I, if I go to a certain part of the morning or afternoon, like, where are my hormone levels supposed to be at? Maybe I've measured them. You know, maybe I have actual data about where my hormones at, and I can, I can see the fluctuations. Um, and, you know, like, then if I'm going into a, a sleep session or an exercise session, that I can just, like, we have a unique context metabolically that we should be able to look at just from top down, what's going on in our body. And, you know, I, I, I've done enough, like, I think exercise physiology is just like really interesting, you know, Cuz like when by the time you're, you're, you're doing cardiovascular exercise, what you're really doing is stressing all the different metabolic systems in like your muscle interactions. And I think, you know, being able to like be that and understand it, you know, I actually think that, that like more analytical people would enjoy exercise more because they could sort of like go a few levels deeper. But you know, it's like, what if you can go through a session, you can say, like, I stress this percentage of my cells and these muscle tissue systems today. And you know, I use lactic acid and I have this level of lactic acid in my blood of which this percentage of which might be being turned into BDNF in my nervous system, in my brain. You know, you get to like more specificity about how you're managing your body.

Dude, it's like the whole game of don't die, right? Like everybody's playing the game and if we just gave numbers to all these things, I think in particular the people like my brother for is a great, great use case. Uh, his sleep was really bad for a while. He believed that he needed to eat right before bed to feel a cook food coma to have deeper sleep. And Saul's like, oh, why do you believe that? And he has a PhD in biophysics. And he's like, oh, it's just based on feeling. And I'm like, all right, well he's the scientific type. So I bought him a wearable and I showed him in the data. I was like, you could just do science on yourself and see how much restorative sleep you're getting. You know, 30 days later he's changed his whole eating schedule. So like that says a lot. And so you're really speaking to the point of, for the maybe technical data driven numbers driven archetype. If we just gave them an experience where they saw numbers that were relevant to a particular exercise protocol or whatever, then that would incent them to play the game of getting healthier. I mean, that would just even be cool, right? Like I'm now imagining, remember the whole first health boost AI vision was like, oh, what if like we made this super different than Shrava and you got a little avatar, it grew like a Pokemon and it's like the better version of you or whatever. I'm kind of having those thoughts, like we create this super gamified experience backed by all this data that we can now have. I have a really strong vision in my head of like it's like a Pokemon esque experience where you have this really giant avatar as the main screen and lots of stealth health things added to it. It's super interactive. You could turn around and stuff like that and click on different parts of the anatomy and then you can learn more contextualized by the anatomy in terms of biomarkers and stuff like that. That's what I'm picturing.

Yeah, I think that's great. I mean, you know, in our, in our model, you can have different self models of yourself. So if there's like a particular way of. I mean, being a human is just really complex. You know, so much. You have one model self that's just like, it's kind of your beautification model self. You want to, you want to look your best. You're looking at models of like, would be what your body fat percentage is. Maybe you're looking at skin follicles and hair follicles, you know, like, and in, in deep nuance, right? And so there's all these avatars that can kind of represent those things. But then, you know, you switch over and you're like, okay, now I'm professional Sleeper Johnson, right? And here's the, here's all the. Here's the model of self. And like, there's gonna be some overlap between those two models. But, but really they're like two different conceptualizations of self that you need for these purposes. And I also think like we talk about, so these groups and gamification, say within the don't die community, like those, that's, that's where the groups happen is like, you know, you're actually aligning on a particular like self model that you can share with other people so that you can like pure subjective subjectivity, correlations and outcomes together using those things.

When we talk through this and we talk really at this level, like, let's quantify fucking everything. It almost removes the human element. But just to be really clear, we believe the human element is like front and center. I just wanted to share, like, I was giving this talk at Dayforce to our engineering workforce and I had to convince them it's related to this. I promise. I had to convince them basically, like, we need the source of truth of all of our work to be done on our architecture in one place. And the reason for that is like Conway's law and just maximizing information flow and efficient information flow. But the way I explained it back to like you were saying, Victor likes knowing if this person's a plumber, so he'll talk in Plumber analogies, right, about health. And so I think that's a big part of it for the developer Persona. I was like, oh, like, I'll just explain to them how we're all nodes and we just read and write APIs to each other. But like, our APIs are closed and they don't follow a schema or open API spec. It's like closed API. No spec is what it is. So obviously that's gonna be bad. And developers understand, oh, if it's a shitty API that's hard to work with and not gonna be able to integrate a lot of systems with it. So then they groked, oh, okay, so that's why we should have the work in one place. So when I'm too lazy to make the jira, I'm actually messing up everybody else's API in a way or integration in a way. So anything else on belief systems and in this particular use Case Rift that we're going down before we move into, like code base changes.

This is what you just said. You brought the thoughts. You know, we talked about consciousness models of consciousness, models of agency. And that is like, you know, think about agency as the lens that you look at yourself and every part of yourself through that is its own conscious actor. And so, yeah, like, when you say that the human side, this is, it's a type of philosophy that's really trying to quantify, like, what. What is that agency? And, you know, being put in a certain environment, how will this, you know, rational agent actor behave in this environment? And so, you know, like, that's, that's not a model that conventional medicine uses. It is a model that, like functional medicine is sort of approaching. But, you know, there's an underlying theory here that if you can use something like predictive processing to model the consciousness and the agency of all the parts of us that are interacting, you just get more predictive, you get better data, you get more productivity.

Great example of this is what I talked about with my brother. There was a version of him, Ph.D. biophysics, that thought, if I eat food right before bed or in the middle of the night, I'll get that food combo feeling. So obviously that's better for my sleep. From the lens of predictive processing, he lived within this world. His self model was there. There was no data coming in that challenged that limiting belief, right? So then I introduced some data. I was like, here's a wearable, here's a Huberman podcast. And that was contextualized to him. It's scientific based and there's numbers that you could do science on your own. And here's another PhD who talks about sleep and longevity and health. And so then the second version of James, my brother, after getting his limiting beliefs challenged was, oh shit, okay, now that I have this, my self model is updated. New self model, new belief system. Throw away that limiting belief. Cool, I'm gonna change my diet. Right? That, that's a very real, you know when we say predictive processing, consciousness, whatever, like that very real use case of my brother thought food coma ing led to a better restorative sleep. Gave him a wearable, gave him human podcast and now he's a new version of him. Predictive processing as a theory says he is going to make the best decision given what he knows to evolve his self model. And so the previous before injecting that data was I'm just going to do what I know to be the best. Now with new data it's like, oh, I really know this, this is the best. So I'm going to keep doing this. So I think that's just the ground everybody that we're talking about fancier terms, but ultimately it just comes down to those very real human use cases.

It's a great example.

That was awesome. We'll get back to feature sync in a little bit. Let's spend some time diving into chunky PRs that you have gotten together based on the roadmap we've agreed to with our partners on the Don't Die team. Now that we talked through all of this, let's get into some of the recent code base changes that we've made as a result of these confirmed features on our roadmap that we know to be high value to.

So let me start with the proto changes that I made. Because the data model is like really kind of core to everything we do. I created this, this PR for the learning objective in the dialectic. Um, so this is a fairly straightforward change here in the, the, in the dialectic model we now have a learning objective that is associated with it and basically has a description of like, what is it that you want to learn? And then there's some topics that you can specify here. Part of the learning objective is what kind of beliefs do we want to collect? And in our, in our system we have like three defined levels of beliefs. But you know, this could be used more broadly. There's belief statements, which is just like a user stating what their belief is in natural language. There's a falsifiable belief where we take those belief statements and we make them into Something that we can collect evidence for and then there's an evidence belief. And I guess we have a fourth level which is a belief that's thoroughly evidenced is a predictive, predictable to belief. And so, you know, like during a dialectic you're, you're wanting to ask questions. So you can say like let's shape this into falsify belief for instance. And so that would be, you can, you could specify the belief type to do that. And then we have this completion percentage which is like how complete is this dialectic towards learning objective?

Yeah, a pretty simple update.

And we also added these perspective selves or perspective model IDs to the dialectics. And now you can have a dialectic with the user. But you can also sort of include this advisory panel of other perspectives that can help ask questions and help understand the user.

And that's the hey, I want Brian Johnson's perspective. Hey, I want Andrew Huber's perspective. Or hey, I want Robert Todd's perspective. Like whatever other models of self we enable in the system.

Yeah, exactly.

Cool. So I kind of think of, I kind of think, you know, how we're in sidebar. I think about that as like, oh cool, I could have a programmatic board of directors at some point or I could ask them questions about career and get their thoughts.

Well, I think it's, it's part of being, you know, like a person is like we, we carry our values around as a reflection of like all the interactions that we have with other people, the things that we'll read in media, like, you know, and basically keeping mental models of people or groups of people in our mind all the time. You know, just to like make sense of the world.

Yeah, a hundred percent. 100%. That's a very common leadership tactic which is asking yourself what would this leader that I admire do in this situation? Right, yeah, very cool.

Yeah, that's. And I think Dunbar and Selman have some interesting kind of like theories around perspective taking specifically. But we'll get there.

And for everybody, this PR was around just changing our models to incorporate that learning objective as a user experience. So remember that wireframe where it's like, I just want to learn more about my users sleep habits or beliefs. This is what enables that.

Yeah, this is the implementation of this went over earlier. Great, awesome. What I'm looking at now is like the server code that's updated and if you, if you look at our project, we have, you know, we have the proto buff definitions, we have a server and then we have two SDKs. The Python and TypeScript SDK that sit on top of the server. And so this is the actual update to the Go code that know implements this. And yeah, it just. It allows you to start sort of like incorporating this dialectic strategy into the dialect or the learning. So it allows you to start incorporating Learning Detective into the dialectic strategy. But, yeah, these rules.

Jonathan, how did you think about going about translating the feature requirement into code? Like, that would be something interesting for our viewers, like, for this particular.

Yeah. Um, so in this case, I did go. I did initially take an approach that I evolved. The. The initial approach was that I knew I wanted to introduce Learning Objective, and I thought I would have it. But within. Within the. The Update Dialectic API call that you would have the tools to sort of like, use that learning objective. And then with. So within. When you call Update Dialectic, it uses the AI helpers that are different, like prompts that we have stored in our code. And so I thought, like, let me put all. Let me kind of evolve the prompt so that it can work with this learning objective. And so it had to do new things. Like, it had to, you know, say, given a specific topic space, what is the next best question to ask and had to come up with strategies for that. And I put all of those initially in our sort of AI helper, you know, tool code. Here's like, generate question from learning objective for learning objective, for instance.

Okay.

And, you know, so you have toolkits like this. And I thought about just breaking it down so that we were using. We use a, you know, a code pattern where we have this kind of like, orchestration layer with a service model. So, you know, the request hits the server endpoint, and then our service models kind of like do all the heavy work with data models in the middle, and they call the AI helpers from within them. And so I just kind of evolved our update dialectic service methods to do all this. But then that became. I realized that was kind of like it was too heavy of a lift for Update Dialectic to be doing that. So I moved some of that code out into, like, into another training harness, which is one of the example apps that I'm building.

And that's for the give me the next best question functionality.

Right. So I think. Let's see here. Here's Generate answer from Belief System method. Yeah. And so the. This app that I'm building right now, this is. I haven't merged this yet. It's not quite where I want it to be.

Cool.

So this one actually will. It creates A learning harness where it uses the learning objective as like a cue to whether or not it's done learning. And then it will like, ask different questions. And it also has its own sort of like, it has its own AI helper and prompts that it can use to, to like, figure. Figure out like, what next best question should be.

Okay, got it. Cool. Okay. So you went down a path and then you realized, okay, this dialectic service is getting too chunky with this new evolution. We need of it. And so that's how you evolved into the current approach and the committed priority.

Yeah, yeah. And it's still kind of like a work in progress to figure out like, what the right balance of separation that concern is. What I ultimately want to happen is that the next best question can be determined by the epistemology that we're using. And so now I'm kind of hacking through that by using different heuristics to like, figure out what the next best question is.

Okay, what's the, what's the hacky method? Right now there's, yeah, there's, there's different.

Like, kind of metrics that the prompt will use. For instance, it's going to evaluate whether a question is answered well by say, you know, like group it into foundational beliefs, practice beliefs, cause effect beliefs, and experience based beliefs. And then like, you can kind of give a score. And this was something that like, as I was working with the generative tools, it was like looking at our code base and how we defined, you know, and it was like using machine learning best practices and then prescribing a way that we could determine whether or not we're learning or not and whether what, what the completion percentage of the learning environment is. So this is kind of cool. You know, I'm just like.

Yeah.

So my goal right now is just to have this like the training system in the environment where basically like, you know, uses the Python gem, it creates a reinforcement learning loop, and it steps through it incrementally. My goal right now is just to see that we can come up with any strategy at all to increase learning percentage and like, you know, ask better questions as time goes on and complete the learning objective. Then once I have that done, I'll.

Then I'll have this test harness in.

Place and then I can go and like, refactor the methodology for how we calculate the next best question and use the epistemological tools that we've kind of had placeholders for to this point.

Cool. Yeah. Okay, that makes a lot of sense. Cool, man. Very Cool. So where are we at with dialectic harness in terms of, like, maturity?

You would say, yeah, it's been really fun to build. I got an initial version, like, working where I was like, okay, I like the flow of this. Now let me introduce the update dialectic SDK into it. And I'm currently configuring some of, like, the data specificities between those. And then, then, you know, just. You just keep running it every time and you make improvements and you try to, like, get it. Get it to ask better questions so it can increase the learning percentage better. But I also, you know, it got me to think, like, what's the most useful output of this? Like, say we train a dialectic. Like, you know, what is the most useful way to know that it's trained? And I wanted to output, like, a table of evaluations to say, like, here's the questions that are like or most effective for us to help to learn with this user type, and here's the answers we expect from them. And here are the, like, you know, sort of evaluation metrics. And like, you know, in. In this approach, I forget if I committed this code yet or not. You know, there's. There's basically, like, rewards, you know, like, if we come up with a strategy to ask a good question, then the system rewards it. And those. That's, like, the nature of, like, kind of output evaluations that you get. But let's say that you want to manually curate that, right? Okay. You could take. You could take the outputs and then, like, manually curate the rewards so that you kind of like, give more reward to a certain type of question to a certain type of user in a certain context. And in that way, I think you just have like, full control to kind of like train something and curate it, then test it again, and then just know that, like, you know, so you have, like a human loop to help kind of tune the questions that in the way that you want your AI coach to perform.

Yeah, I like how you're bringing that, you know, direct synthesis of research of what we learned about. Hey, like, uncertainty and trust are really key to the solution here. So providing that all the way through as like, a philosophical approach and product and development, I think is great. Could you explain to me what episode means? I. I just got caught up with that. I saw Episode Step to next episode, and I was like, what does that mean for us? I'm guessing this. I'm guessing this is like a dialectic session, training session or something like that.

Yeah. Like, I think each episode is going to be Its own sort of like question and answer session, like interaction with the user.

So like I am a user. I type to Brian AI, it types back or it asks me questions, I type, I typed back. That would be one episode.

Uh, yeah, exactly.

So, okay, I have to ask, is like, is episode the right word to use here? Because I thought, I, I, I think of that question answer back and forth as a dialectic. So. But I also see with our given data model why you would come up with another sort of placeholder noun for this testing use case. So I'm, yeah, I'm curious how you even come up, came up with the term or if you think it'll change or if this is just placeholder or something like that. But yeah, it does, it does confuse me when I hear episode in that context. And I know dialectic is a word.

Yeah, I think episode is like, it's not a, it's not a domain language word. It's just more like here's, here's the test harness. And like, you know, we're, we're iterating over something so in the sense that like, you know, you have, you have this, you have a gym, you're training a system. Each of the episodes is just like a word that is like the episodic nature of like each test that you're running in it. And so it's not really meant to be domain language. I'll say like, you know, I, I generated this tool and then as I'm updating it, I have a lot of changes locally that I haven't committed to what we're looking at right now and, and getting it more integrated with the SDK. Yeah, there's lots of places where I want to update the language and you know, like that, that variable, maybe a variable that I update.

Okay, cool. Yeah, so episode in this context is really, I'm thinking of as a training session, like, or at, at a gym. It's like I go for a training session or something like that. That's how I think about it. But. Okay, cool. Thanks for the clarification. Any other changes that are notable to go over?

I think this is the work in progress stuff. And let me just, you know, when we look at our apps repository here, we have, you know, so the PR will be a third app that's created here and it'll be the dialectic test harness and then that's using the Python SDK.

Yeah.

And then I'll take what I learned there and make it into like a user interface version of the dialectic designer we're using the TypeScript SDK to sort of do the same things or it might even work alongside the Python app. Haven't really decided, but that's. Yeah, we just kind of like demonstrate how to use these SDKs and how to approach these use cases.

Cool. Awesome. And there's four domains that are really important for implementing a customer on your B2B SaaS and then getting them successful and happy and repeat customers. Buy more of your stuff. It's data conversions, integrations, reporting and analytics and then finally change management. And so change management is honestly usually the hardest part just to explain for the audience. In B2B, there are multiple decision makers that you basically have to get to agree to, to buy your thing, to leverage, depending on your software. So for our software, it's like we have an SDK that enables the application developer Persona who works for the company of the health and longevity influencer, and they may or may not be the ultimate decision maker of the technology roadmap. There may be a CTO involved and so for no die, that's the case. There's an application developer who we're working with day to day, who's doing the AI health coach implementation and really empathizing with them in this user journey. And there's the CTO Persona who basically need that person to be happy, but they also need to know, is this flexible enough? Are there other things that I'm getting from the product roadmap where, you know, if we choose this technology, we'd be super locked in, could we change? There's total cost of ownership. They're thinking about that too because they're worried about budget, they're worried about meeting a future product roadmap that the influencer might be. So for our cases, there's three levels of decision makers and for a go to market motion in B2B, you generally start at that very like that. For us, it's the application developer for like a workday or a day force or gusto or something like that. It'd be like the payroll specialist or the recruiter or the H R S administrator. Back it up a little bit. Looking at our roadmap from the lens of integrations, reporting, data conversions and change management, it's, it's funny that the private healthcare use case and this AI health coach use case, one big commonality beyond, hey, give me, you know, personalization. One big commonality is like, hey, we have existing data, can we just use that? And can we be bootstrapped really quickly? Can we get time to value fast Right. So that really tells me, from a product perspective, like, that's a hurdle. If we eliminate it, it basically eliminates data conversion as a problem and then reporting and analytics. We've talked about that throughout, but it's how can we give numbers to show the Persona, like the application developer Persona, at every step of the journey? We are solving that credibility, we're solving that uncertainty problem. Right. So in the dialectic designer, we could show you that with, hey, you could control what cohorts or archetypes you apply to. Hey, you could look at their underlying beliefs. Hey, you could look at making your dialectics more effective. And, and before you release anything in the wild, like, we're giving you a lot of reporting analytics that makes you feel certain about that. And then change management. So I think change management is going to be something that will be heavier once we move into the early adopter, early majority, like up the technology lifecycle adoption curve. Right now, we have landed very luckily with innovator Personas where I think they're more okay with the change management stuff being like, yeah, we'll figure that shit out at some point. Day two operations, we'll figure that out at some point. But, like, can we even do the thing? So in my experience, for the early adopter, early majority laggards, change management is way harder than for the innovators. Yeah. So I wanted just to more set that lens. Like, that's one of the ways in which I'm now looking at our backlog. I've implemented a prioritization framework and now I'm also thinking about it from those pillars of like, B2B software. Any thoughts on that? Those four pillars?

Yeah, it seems like a great lens for roadmap and, you know, like, just how to prioritize the features. The integration is the part I think about the most, you know, just making it familiar and easy. And say, like, say you're kind of, you're a developer coming from the standpoint of having a database full of users, like, how can you very quickly plug in the epistemic SDK and you know, just, just like test out some, some ways to understand your users better. Yeah. And I think that's one of the beauties of building the developer tool is like, if you can get the integration that, like seamless and really elegant, then, you know, you can kind of like hack through those other problems one at a time afterwards.

Yeah. I think the most seamless experience that we can maybe get to one day is here, drop one line in your code, whatever, and then somehow we don't, we don't have a SaaS service right now, but somehow later we do and then we just create your models itself. Like we just auto detect everything and just create your first self models for all your users or a hundred users or something like that. Well how do we. Like this is us brainstorming now. But like one question I have is what's the quickest path to get there?

Yeah, where like a developer could integrate quickly with like a couple lines of code and in the beginning you see some amount of value, right. You know, if you think about a service deployed and like you can create a certain number of self models and then you have a certain amount of sort of dialectical interactions that gives you ways that you can incorporate it in the user experience that, that you're using. And I think like really, really can that make it really easy sort of, sort of like for the developer to get their developer key to get it set up and then to just like start associating users and their, their database with self models and start populating data into those. And like building belief systems. A magical experience might be is if like you know, I have an existing user base and I can like I can create self models for each of the users but then I can import like all my existing data and it'll create belief systems for them and then I can simulate a chat with each of the user archetypes that come out of that sort of process. You know like I could see a developer spending like you know like a four hour session with our SDK and then at the end of it being like I see my users in a whole new way. And I think that would be a big win.

One of the things I think about is like Stripe for example is a good exemplar, very developer focused, API, SDK company, product company and they just have it out there. And so I'm kind of imagining oh if the education part for us like we're not stripe, we're, it's, we're way lower on the maturity curve. So I'm thinking once we get to you know, other innovators, early adopters, early majority, I'm thinking of like maybe even in our website in the future we have example user generated data with very easy, just visual tools to click through and show oh, this is example analytics. If you are building an AI health coach is what it'll look like. And here's The JSON of 10 exemplar archetype users like the couch potato or the Olympic athlete or the amateur hobbyist or whatever. So now I'm thinking the 10 star experience is like they walk to our website, EpistemicMe AI, they see a little Epistemic Me explorer, let's say instead of API Explorer, we show belief system trees, we show some cohort analysis with archetypes and then some other cool things that, that help them really understand. Oh, this is like what you get out of it. And oh, actually to think about it, the Belief Mirror app would be very powerful in that kind of experience where, oh, cool, I see the raw data. Okay, cool, I see this. And then, boom, here's ChatGPT. But I could open up the hood. Oh, yeah, that's. That is cool. Like, that is thought provoking. Like, I'd imagine the developer. Because the developer we were talking to, they were concerned about hallucination, right? And I think actually one of the most powerful things we're doing is saying, hey, a user can actually pop open the hood and modify it. Like they could look at their beliefs, they could correct it, right? They have a mirror of that. Like, we could provide that experience. Or the application developer could provide that experience if they wanted, which is, can the user modify their belief statements directly? Like almost SQL edit their belief statements. So basically like, that becomes like, users are correcting their own data as they go, right? So that's way better model, at least hypothetically. It's a way better model than, oh, it's closed. Users have to keep talking to the thing, we refine our learning of the thing. But it's like, oh, if we could just put the human in the loop, the user themselves, then the efficacy should go up over time and then reduce the risk of like, bad recommendations. Yeah, just wild tangent, but I'm thinking that's the 10 star experience is we show value and really even solve the educational gap without us being in the loop. It's some artifacts or experience. Belief System Explorer is what I'm calling in my head. Or at best, MKD Explorer is what I'm calling in my head. They see that on our website and then they could be like, click get API key. And then it's that global instance of a service that we give them maybe 10 rows for and they could play around with it and then it's like, all right, call us to upgrade or whatever. Haven't thought through the rest. But that's what I'm kind of imagining. If you want to riff off of.

That, I think it's a really healthy place to go. Because the education component of philosophical world models of metaphysics is hard to Put in a sales cycle. And so, yeah, and so then it's like, there's this aspect of how much personalization or subjectivity do I need? Right. And then how do I really, like, how do I really feel what that means? And I think that's like, we just have to distill all of this philosophy down to, like, when you interact with the belief mirror type experience that you can say, like, oh, wow, there's something here, you know, just, just like, you know, like the first time I played with Claude Artifacts, for instance. For instance, I was like, oh, wow. You know, like, this is how I want to build, this is how I want to generate documents, and this is how I want to generate apps. Like this just you. It clicks. And I think if somebody sits down, they have the same feeling when they're just answering questions and then they can see a belief mirror and they can really feel how valuable that is to the user.

Yeah.

Then like, a lot of the education kind of is taken care of. Then you can graduate them to the next step of like, okay, now let's, let's look at what it means to personalize this for all the different user types that you have and really get them thinking about how important personalization is. Like, we just need to get them mature and, like, really hitting emotional aspects of, you know, the UI interaction.

I think the belief bearer 1 is the strongest one for the general audience of anybody. I think the developer would get it, I think the influencer would get it, I think the CTO would get it, because it's a very human thing. Like, when I've used, this is another thing for the audience that in the past two months we've been working hard on the branding, the positioning, what the copy to describe this. Like, and we tested so many different ways of explaining this, and the ones that land really well are how can AI understand us if we don't fully understand ourselves? Or the opposite, how can. If we don't fully understand ourselves, how can AI understand us in the future? I've not met one person where, like, they hear that and they're like, wait, what does that even mean? I will say, being in a bubble in the Silicon Valley, like, that might add some bias there. But at least that innovator category, grokstack. The second thing that people grok immediately is, hey, do you use ChatGPT? And most people nowadays have had an interaction and I'm like, you know how it's learning about you, right? And it tries to personalize recommendations. And they're like, yeah, because it is, they feel that. And then I say, well you can't pop open the hood to really know what it knows about you. And they're like yeah, that does suck. It's a little scary. Like I've got. Yeah. And then. And they're like, well that's what we're building is we're building that in a way where you could pop open the hud, it's your own data, you can modify it so that part people, when I get there, people are like whoa, that's so cool. Like trying to save humanity or whatever. Like it's just, it gets it, it tickles. At a mission and purpose aligned, like almost ethical good force that we have going for us. Those, those are the two strongest examples that I've been able to just say in conversation. So, okay, so we talked a lot about. So we went through roadmap, our soft partnership private healthcare practitioner use case. We went through code based changes based on our proposed roadmap. We even talked about really far out there things about, you know, how do we find success in the future? It's, we've modeled every agent in the world down to the cellular level and every piece of tissue, organism, muscle fiber on the planet. So hopefully people get a taste of like what are we envisioning? But like we gotta land somewhere, dominate a niche and that's really the longevity health influencers that are looking for an AI health coach as a new offering for their communities.

One thing I think you're just talking about the ICP of somebody who's building an AI coach versus the ICP of AI alignment is just kind of interesting to think about the distinctions between those two cases. The latter case, I mean I don't think we're really ready to you know, present a specific case. But I think, I think a lot about just like connecting dots about like what people are talking about in the AI universe to like AI alignment in the way that we're thinking about it. And I don't have like a sort of a clear narrative or demonstrative system but I think like, I think meeting the coaching use case and showing how useful it is, how valuable it is, um, you know, like it's kind of like the way to get there, I'll just say that much.

Which industry or market can we solve underserved needs and pain points that people would be willing to pay for? Health was the first thing that came to mind. Because we're both health nuts, but also because we're both health nuts, we know how emotional health is. So like Tying to people's beliefs and stuff like that would be I think a really hyper personalization essentially by tying to people's beliefs to them. Like that seems like that was our hypothesis is like there's gotta be a lot of pain here. Let's, let's go do more research. And then we found it. And so the way I think about it is health is a perfect use case because the next if we solve this AI health coach thing and solve some of the early bones or foundations of the private healthcare practitioner case, then the next question is like I said earlier, all right, cool. This is better than the existing experience of I have to one on one get on call with everybody in my million plus follower community. It's not happening. So this is better or I'm a doctor and I couldn't sort through all the intakes. This is a little better and people are actually following my recommendations now and I feel like I'm having a better impact on health then the next. Obviously the next thing is like, all right, well like in between the doctor and the patient there's a lot of things about the patient that we could then contextualize, get more data for and inform better predictive processing and recommendations. So I agree with you. That's all to say I agree with you. Where we've, I feel very lucky we landed in the health case because I see it as a way to further that parallel longer term mission of we're solving AI alignment from a philosophical lens, which I think is needed and but you know, we're a startup so we got to propel it forward and pick problems to solve that will propel us to the next milestone. So this is where we kind of organize our work and stuff like that. I'm just sharing with Jonathan like I've cleaned up. This is the epistemic me product space. So I cleaned up. This is the use case for AI Brian. This is some of the branding positioning like I've been kind of just putting in metaphors and stuff like that here. And what's our narrative? I put in ICPs for different ICPs. So like open source contributors. Would it be an ICP? We need to know about CTO of a health and longevity company, Health and longevity influencer seeking a hyper personalized AI health coach. I had one in here for the actual contributor or developer like the Greg. All right, I gotta find it. It's somewhere in here. And then this has some market analysis. And so House was like number 1,613 billion by 2030. And then after that is a Few others. So enterprise productivity, E commerce, financial services, media, entertainment, education. From my perspective I think enterprise productivity is also pressing because like personalization for employee experience is big getting people more productive in enterprise. That's always top lined E commerce like we talk about matchmaking a lot like can Amazon use models of self for their users to increase conversions for cross sell upsell? Very like pertinent use case. Could they even edge out another 0.1% gain on conversions that would amount to millions and millions of dollars for them. Financial services, we talked about the personal finance use case in a previous podcast episode. Media, entertainment, you know and gaming, that's self explanatory. Education actually is where my, my heart is and I actually think this number is. I don't think this is right. I think this is supposed to actually be a much higher number. But education is also somewhere where I see hyper personalization does really well. So for the audience we really think if we're successful in like 10 years, like we have surface area in every industry but obviously as a startup venture you gotta prioritize so like we're landing in health and that's what we think propels us forward some monetization ideas. And I'm still working this through and this is I messaged you about this but I'm like would love to have whoever's leading the Linux Open Source Foundation, I would love to have them as a mentor because I'm like thinking through like how do we actually grow this into something that propels itself. But there's multiple models in the open source community of that and I'm still working it through it and then oh yeah, this was just some random resources about open source community building. And these are some of my notes on like high level what is the go to market kind of sales motion? Make influencer know who you are, engage in their community, build a relationship with the main app developer, move to building relationship with their cto, continue the relationship, make their life easier with the product which will help make the influencer's business grow. So kind of cleaned up a lot of this and then for AI Brian, I went through the backlog yesterday and started putting together just like a high level take on the I'm using the RICE framework so reach impact conference effort and I just scored my high level take of all of these based on my feeling of reach, impact, confidence, effort, effort part is total. Guess this is where I'll need you to go through and fill this out and then ultimately we get this like super nice prioritized backlog. So yeah, we could even, I mean we could do this now if, if you'd like. Yeah, let me share.

In terms of the icps and like, you know, we kind of started the conversation with the distinction of AI alignment, ICP about the health use case and subjectivity. Is that like, I think it would be a really probably good and interesting thing now if you had everybody in the world had sort of like a self identified health religion or religions that are like, so did two of them the best and you only interacted with AI that was going to be like helpful in that way you were aligned with the AI in that in your health religion. If you think about somebody with a defined health religion then going and getting a job at an enterprise and then needing to be personalized, you know, in the enterprise use case if they already have their health religion defined, like it helps constrain, it helps us to make sure that the AI is going to be value aligned with them on that level before then we think about like now what are your, what is your ideal sort of purpose aligned work? The AI would only recommend things that interactions at work that were like along the lines of your health religion and they wouldn't recommend anything unhealthy to you, right?

Yeah, yeah. So, so basically you're saying I am Robert and I believe in this world that epistemic me is pervasive across every application and software, business and in society. Software and business and society. I use an AI health coach AI Brian. And it is aligned with keeping me alive because that's its job, not just that optimizing for not dying. So then when I bring that self model to the enterprise space to do my job, then whatever system that hooks into are you saying is already aligned on yo, like I'm trying to keep you alive, I'm going to keep trying to keep you alive and give you recommendations that layer on top of that foundational philosophy in which I should engage with you. And then the other thing I would add is like I like how Brian's going about it, which is not nine is like we need to align on just that one game. We're all playing this game, we're not gnawing. So we agree with that AI might make us die. So if we're agreed on this, we should figure that out. That's the simplistic kind of bro science level of me explaining how Brian is trying to inculcate this full stack philosophy to get us to solve AI alignment. But I think you provide a different perspective which is like in this world where every health AI health coaches are ubiquitous. I wanted to share. I actually have this video I recorded. I've interviewed my cousins, because I'm just always on, right? So I was playing with my baby cousins, user interview research questions to them. And so after I showed them a bunch of AI tools, I'll actually share this video with you. It was amazing user research in my mind where. Well, first of all, I don't have. I'm not talking to kids all the time. Like, my baby cousins are kind of my, like, the only kids I get to talk to. And I showed them ChatGPT. And I was like, hey, this is an AI. And Jolie, she's eight. She's like, what's an AI? And I was like, it's artificial intelligence. She's like, what does it do? And I was like, I don't know. Just ask it to do something. She said, can you draw me the sky? And I was like, okay, I'll do you one better. Draw the sky as a watercolor painting. And then it, you know, went into the image processing thing. She was kind of intrigued and she's. And then it popped up and it was like, scott, you know, sky is watercolors. And she's like, what? She blew her mind because it was the first time she's ever seen that. And she's eight years old, right? They have a different perspective on life. They're just way more curious and open. So then she brought her little brother Max, six years old. They both looked at it and were like, what? And so now he's like, can you create me a pizza man? And I was like, okay, what do you want the pizza man to do?

They.

They read Captain Underpants. So, like, very cartoony, kind of satirical Sunday morning cartoon sort of style. Art style. And so I was like, okay, you wanted that in art style. And he's like, I want it underwater. And I was like, okay, so something like Atlantis. So I just prompted ChatGPT. I was like, okay, now make me a pizza man fighting enemies underwater. Atlantis in Captain Underman style. Then the whole like, oh, is processing came up again. And they're like, on the edge of their seats. They're like, what is going to happen? This just blew my mind once. What's going to happen? And it. Boom. Pizza man underwater. In the art style. They're like, what? Like, minds blown. And then they were excited. They were like, oh, my God, now could we make there be turtles around? Could we make the pizza man fighting a toilet that has a turd for a head? And I just made all these things. And then, you know, once we got done with like that, just making images, I was like, well, why don't we make videos? And so, so I went to Sora and just showed them. I was like, what video do you want to make? And I made it something similar. And they were, wow. So then I did the day after, I let them kind of absorb that. But they kept coming back to me throughout the next 24, 36 hours. And they kept asking like, could you make me something else? Could we make this now? Could we do this with the pizza man? And then I had an interview with them 24 hours later and I was like, what do you think of AI? And this is really interesting. So they were really excited. I would say 95% was just excitement. And I asked why? And I'm like, you know, because you have to wait 30 seconds to 60 seconds to get an image or get a video from this. And it's not the best thing. The videos are like 10 seconds long or whatever. So like why is this, why is this cool to you? Because they watch other movies, TV shows, cartoons, they play games. I'm like, why, why do you like this? Why do you keep coming back to this? And they're like, it's because it makes me feel like I have a creative mind. And I was like, wow, holy shit. Like that is really the underlying, like what is the inner child? Cuz they're children. So what is the child thinking when they're using these tools? Not the adults using these tools nowadays. I think some of them might be actually feeling that way. The maybe people who are interested in graphic design or art, but maybe don't have invested in the skills and so now they can do stuff with it. But I actually think if you took Jolie and Max, my baby cousins, and we just put a time slider 20 years down the road, these kids are going to grow up and then grow into consumer purchasing power with a expectation that every software system they ever interact with can provide that it knows what's in their mind and can translate that to reality or some deliverable or some output that that was some of my learnings. Just talked to my baby cousins and. And then the other thing. So that was 90% like super excited. I get to take what's in my head and make it reality. It makes me feel like I have a creative mind. Have that data point, second data point. I got my little cousin Max said that he felt that he was sus. And I was like, whoa. Well, first of all, he, he knew that word or that slang to say it's suspicious. Called it sus. Specifically his. His parents are pretty hip and cool so I think they, he gets it from them. But he compared it to a TV show character and I forget what TV show off the top of my head but it was like some character that was like famous in the TV show for backstabbing everybody else that was like a hero or good. And so he said that he got that same kind of feeling in talking to the AI for the first time. He's like, how can he do this? It's a little sus. And I was like, oh, interesting because me, Jonathan Dean talk about all the time how it's freaking sus. I can't open up chatgpt to look at what it knows about me.

Right.

So it's like even kids just got introduced to an AI in their first five interactions. They're even though it's really cool make them feel that have a creative mind.

You remember when you started chatting with them and I sent you that picture of like robot pregnant women?

Yeah, yeah, yeah.

I mean like it's just, it's just crazy how things, faster things are evolving like so we used to grow up the same way our grandparents did. Same rituals, same beliefs as like you know, our grandparents for as long as humans existed until about 200 years ago. And then like now, now we have media and like each generation kind of lives a different life from, from incubation and like and now like you think about a kid who's not even like you think about say an embryo's nervous system interacting with a digital world of LLMs before they're even like fully conscious in the womb. Right. Um, and like that may, there may be parents who are like I need to raise my kid this way so that they come out and they have like, you know, a hyper advanced education by the time, you know, they're in preschool or something. And I don't know what it means to like it just seems like we're, we're increasingly bringing these tools into ourselves as like you know, it's second nature and yeah, your cousins are growing up one way and guess what, probably kids more 10 years after them are going to grow up even more integrated to technology.

So 100%. Yeah, yeah, yeah, I think you're right. Like we don't know what we don't know. There's a lot of uncertainty in the future. And for the audience, Jonathan, when I described it then hidden to this him about this interaction with my baby cousins in Slack, he sent me this like, I don't think tweet or on X of. I don't think it was real, right. I think it was AI generated of like AI embryos. Like an AI robot feminine.

Yeah, generate.

Yeah, that had like a belly that was translucent that you could see a baby embryo being formed. And I was like, oh, this totally makes sense that we would solve for pregnancy this way in the future. Especially with everything that's going on around like gene therapy and stuff like that. Like that, that I'm like, I know this is like futurism right now, but I'm like, oh yeah, like listening to life force, listening to all these other longevity things at the intersection of AI and tech. I'm like, yeah, that shit's happening at some point. I just don't know when. Yeah, it's a crazy world, man. Crazy world.

How does modernist philosophers saw this coming in the sense of like that the fractured narrative of like a shared reality that religion brought, combined with the evolution of technology is going to give everybody these hyper personalized media environments that effectively, if you take that, if you extrapolate that, that we all just end up living in our own worlds and never need to even like interact with one another. And that was the basis of the Matrix. Like the philosopher Boulidard is who wrote the book that the Matrix is based on. That was like the early 1980s take on like where technology is going given, given like flooding of media and hyper personalization all driven by AI.

So interesting. People who are basically visionaries, they write stuff like Dune or Player Ready Player One or Snow Crash or whatever. And then it's like the Engineers is a funny way to think about it. The engineers read the sci fi books and then they grow up and they go fucking implement it. Yeah, Only recently I realized it was written in the 70s. It was just called 1984 or whatever. I had this bias in my head or something like that that I thought it was made in 19. And I was like, wait, I thought it was so accurate when I read it. And then I realized it was written 15, 20 years before or whatever. And it's even. I mean like it really speaks to the world we're in today, right? Like, yeah, a lot of parallels to society we could definitely read about in sci fi books of the past. It's fascinating. So the AI alignment, icp, tell me more about that.

Yeah, I guess there's just this question in my mind of like, do we really need to worry about other ICPs other than health? Because if we, we just like embrace a version of AI alignment that is health and then build everything else on top of that. Then, you know, it, it just is the most sort of foundational thing. When you talk about AI alignment and you talk about like AI that understands us, like if, if we had people that were could evolve identifying different health religions where the AI was aligned with the way that they want to live their life, then, for example, them going in, finding a job at an enterprise company and being productive in that company is just gonna be within the context of their health religion. So it's like there's sort of like, is there even a need to talk about other ICPs until we kind of like, you know, can match the AI coaching use case ICP to NIA alignment icp.

If you're asking me, yo, should we focus? I'm like, yeah, fuck yeah, we should always focus. But if you're asking me, you know, like, Robert, should we lift up our heads or should we just keep going here and it'll take us there? Which I think is what you're asking. I think you're right. The only caveat I'll throw in, it's.

More of a part of what I'm asking.

Yeah, yeah. It's more of a trade off, I think, which is so zooming out a little bit. I had it upon me for this quarter to try to get an advisory council together to help balance our technical implementation with people. It'd be great if we had Michael Levin or Friston on our advisory council and then we could show them code, we could show them what we're trying to do with AI health coach and they could actually rigorously help us make it better. I think that's still good, but we should probably narrow it now, is what I'm thinking. Like it should probably be people that are also like us that are very interested in the health use case and seeing that through, that think about health and longevity a lot. The only one that I see, and if the audience sees other ones, let me know. That is actually trying to sort of full stack tackle the AI alignment problem from a philosophy lens, not from like a, I'm in a lab, like kind of thinking about it from an ML engineering lens and kind of, you know, theorizing about it. But he's like tackling it from like a human consciousness. We all need to align philosophically and then also at the same time, we got to solve this with technology. So my thinking is if we laser focus on solving this stuff for him, the data points we have from him are that he wants to Solve AI alignment for from the lens of philosophy. So we just keep doing this. I think we will get there because he wants to get there. That's my perspective anyway.

So I think Brian's on the leading edge of like, that AI alignment is a health religion. Yeah, totally. And I think probably not too far behind them is Peter Diamandis with Found Life and Tony Robbins. You know, they're very much moving that direction. Full stack.

I gotta look more into that. Um, what are the latest updates you're getting from them? Like, do they philosophy and stuff?

Well, you read Lifeforce.

I was, I'm still in it, but.

Yeah, yeah, I mean, they don't, they're not talking philosophies specifically, but I think that like the, the aspect of evolving the patient provider relationship. They have four different clinics and so they're, they're just going at it from the direction of like, let's have our own clinics. Let's build data models of all of the patients. Let's use AI to like, follow all these data models and help like. And they have more than just like, they have functional health coaches, they have conventional medicine doctors. They have a whole team around every single one of their patients. And their patients initially were paying like $50,000 a year to be part of the, the clinics, and now they have like a lower tier offering.

Brian, I think, has been very like spectacle based. And then he got into Antics. Yeah. And then he got into food and supplements, which is like, you're starting off mass market. They started off, you know, upper market. Like, let's sell it to the rich and then move down market.

They're selling to the person who has money and they want to turn that money into years to live. So these are older people that are wealthy and will pay anything to just get a few more good years. Right. I see Brian's movement is applying appealing to a much younger demographic.

One of the things I'm thinking about is like, if we reverse engineer their go to market motions, what does it mean? Their go to market motions for AI alignment, also for the audience. We are totally armchair speculating right now. None of this should, all of this should be taken with a grain of salt. Here's my thinking. Even with that grain of salt, why I like Brian's approach is that he is creating an app where you could voluntarily put in data and he will be able to accrue a bigger data set faster with the gamification and loops that he's putting into his app, which is like, every day we get your, your markers and Then we create events where you could measure yourself and then put it in the app. And then you progress, you share that, you interact in the app. You have a social side where you interact. And so that's another piece of data that contextualizes some of the biomarkers and progression and health. Whereas so, so he has a, hey, I have one data model. I don't got to worry about anybody else's systems. I have it in my system that's actually from a, from a product entrepreneurial lens. I like his advantage there. Peter and Tony, they've started these longevity clinics. I gotta look into it. Laura. I don't know if they have a common app between them, but I can't imagine with the complexity of teams that are applying to this. If it is like that, where one patient, 50K one year, you have eight people around you with different expertise around your health. What's the software system those people are using? I highly doubt it's like one data model, one software application they've created for every single longevity clinic across all of those practitioners. So I would guess based on what I know, and this is also grain of salt, they probably have a data fragmentation problem that they will have to bring together at some point, whereas Brian doesn't have that problem. There would just be different problems to solve there. And then eventually you talk about the religions thing, that religion or models of religions tied to models of self are going to be stronger for whoever has the most data. That's, you know, maybe this is like my armchair thinking about it, but that, those are some of my thoughts.

Yeah, the, the Fountain Life clinics, by the way, they aggregate all their data into their, I think it's called Zora AI coach. So they, they are, they're definitely like bringing it all together, synthesizing it in one spot. I just been thinking about the power of, you know, the personalized health religions. If you had to select a health religion in order to, to find the AI that aligned with you.

Yeah.

Then like you're going to default opt into something and then you can like, you know, sort of opt into something that's like more specific or more tailored to you after that. And that might be where your, your Brian Johnson's and your Peter diamond have offerings. But I really see like Peter Attia is going to have a health religion, you know, and Rhonda Patrick's gonna have a health religion. And all these people that are influencers, they're really just disciples of their own health religion that is appealing to people in the way that they wanna live. Their life. And so you know, like if, if that's really, that's really all alignment needs to be about in order for safety to be provided. That seems infinitely doable. It seems like a lot more concrete and specific than the AI safety methods that are, that I hear about coming through OpenAI and anthropic and this kind of thing at least when it comes to like individual interactions.

So a hundred percent agree with that. I, I definitely, I mean I'm biased as shit but like I think our approach is better. We're taking a differentiated. Then maybe it's just going to come down to this. Like I had, I had an assumption that they just had fragmented data but maybe they don't.

Yeah, I don't have firsthand like information but I, I think the, the overarching thing I'm just like like wanting to flesh out I think is this idea that it's just health religions are imminent. They're, they are upon us. It is the same as AI alignment. Do we need to worry about any other ICPs other than just like you know, AI health coaches? Those are some of the like the formative thoughts in my, my thought landscape right now.

I think from an entrepreneurial lens. No. And then from like a purpose lens, no. I think, I think we just hyper focus on this and then we'll get there which is nice like which kind of tells me we picked the correct space to solve or the right problem space to solve with right use case. So that feels good man. Because like I would love to just focus. I just like put all my time and energy on like one thing and get to really, really, really well. So I'd love to ask Dean about that next time we chat with him, get his perspective on it. But logically it makes sense to me like if we hold, if we assume that our hypothesis is correct, which is health religions are going to be just they're upon us. It's going to be a thing Then I think extrapolating that, yeah, I think we just keep focus on the AI health coach thing and it plays out the way we envision then I think naturally we will be on the forefront of solving this. At the very least, like if we get integrated with all every health and longevity influencer or big company like Fountain Life that wants to contextualize their user data set with belief systems, et cetera, then like our tool is, is. Is is literally helping with the AI alignment problem at the lowest level of individuals and macro scale too. That's how I see it. Def. We just got to execute Dominated that market of longevity health influencers that are interested in AI health coach, and then do that, get enough reps there and expand to probably private practice use case and other health use cases and just dominate the entire health and longevity market. Get a hundred percent, get 100% there, and then we can move to enterprise productivity.

Right.

Education and stuff. So. So this is the way of thinking about it for a progression of. Of markets is we had to land on a particular battle zone. Like in Crossing the Chasm, they talk about this concept of D Day. You gotta pick the frontier you're on. So our frontier is health coaches and longevity coaches or influencers. And in particular of that people that have an AI technology bend and then of that people that would love to leverage an AI health coach. I think we kill that, dominate that market, expand, and then we just get to. Okay, now it's click button, drop some content, or send me your YouTube link and we've created your AI health coach. Click button, deploy, you have it on an app. Then we capture the laggards of that category, then we expand past that to anybody in health longevity. That's kind of how I'm thinking about it.

Yeah, it's awesome, man.

That makes sense, right? Like in my head, it makes a lot of sense.

It does. I mean, I can see a similar path and I just. Yeah. Just the whole idea of health religions and AI alignment, it seems like sort of obvious on one end of scale. And then like, sort of like, I can't talk about this with anybody on the other end of scale. I think really wanted to flesh out for myself like that. That's the right. That's the most thoughtful way to think about it.

Yeah, I think we should talk about this like another. I think Don't Die is probably the only group on earth that I feel like where we could probably talk about this and get some traction of an idea because people there are thinking about it. I think a good test would be in two weeks. We're doing that Don't Die event and we should totally bring this up. And then also, like, just tell people, hey, we're reading Super Agency if people want to join the book club. And then we could use that as a feeding ground for testing our ideas. Maybe a, you know, slightly biased group because everybody there is already in Don't Die. Give us a little less bias than just talking to each other. So, yeah, so because we're basically like, yo, bro, am I right? And I'm like, yeah, fuck yeah, dude, man. Of course we're right. So it'd be good to have some of our assumptions questioned, even just like different questions to ask of us that we could strengthen our position in thinking.

Right?

Yeah. I've been going through the Mastery Book. Oh my gosh, dude. I'm maybe 75% of the way done. It is one of the best books I've read. Like, it's so good. Really, it's so good. And I think it also appeals to me a lot because I love stories and it's full of stories tied to takeaways, and I think you're going to really like it. I learned so much more about, like, Einstein and Leonardo da Vinci and Mozart, like, all these, like, historical figures that have made a giant impact on humanity. And I'm just like, wow. Like, they all started off from humble beginnings. Like, yeah, and they all had very human struggles and they went and just left a mark on humanity that was outsized. And I'm like, dude, like, we can do it. Like, all right, everybody, thanks for coming to listen to the ABCs for building the Future podcast. Today we kind of talk through our technical implementation of our feature roadmap that we really prioritize based on this soft partnership with the Don't Die team. And we also have scrutinized our feature roadmap based on some other use cases we're looking at in that private healthcare practitioner use case. So hopefully you found this helpful to actually walk through some of the code changes that we've made around our dialectic service. The test harness that Jonathan had put together for the dialectic designer use case to increase efficacy of a dialectic so that an app developer Persona could release a really high quality AI coach that they're confident in. So appreciate you tuning in and hope to see you next time. Take care. Bye bye.