How AI Will Redefine Humanity

A thing Elon's putting a big bet on. Robo workers are going to be a new category. Right. Draw that to 20 years. I could easily see robo worker made in your house or cleaner. Right.

All of his Mathematica software now built to take these simple rules and generate physical laws that we can simulate gravity and quantum mechanics and everything that we sort of have observed in the universe.

There was this thing he said, I wrote this note. It was, I died laughing about the climate change thing. He said, wolfram has this theory about how actually the entire everything is just all down to planetary computation. Being around people can also affect like very toxic people can also affect like your well being, which affects your stress, which affects your emotional well being and stuff. And that could lead to some level of risk too. So maybe start there as a use case.

We say, oh, the next best thing for me to do is go, you know, like eat some turkey in the fridge. Right. We could actuate that and like go grab the turkey and chew it up and digest it and feed it to our bacteria in our stomach. And when epistemic institution is asking that question, everything that we sort of have observed in the universe, he's got a theory for how that evolved.

Whoa, I need to look at this. That sounds next level. Hey Everybody, welcome to ABCs for building the Future. I have Jonathan, my co founder on epistemic me with me here today. And in this podcast we talk about building in the open and we bring on guests, people doing big things with big stories to tell, people interested in building the future and maybe you are too. So join us today we're going to be talking about two things. One is a really cool talk that we'll link in the show notes and we're going to have a subsequent podcast to talk a little bit more in detail at a high level. It goes through the evolution of technology at the intersection of science's perspective. And we'll layer in some of that of human evolution. And then how does that relate to AI? And then how does that relate to epistemic meat? And we'll also go over a live backlog prioritization grooming session and then we'll translate that into a shared align vision that we have for a demo for a deliverable in a week. So a couple things today. Cool. So Jonathan, do you want to talk about the artificialization of intelligence?

Yeah. Regards to the Benjamin Bratton talk. Right. And so this is interesting because I've been thinking about needing a word for this phenomenon in evolution. And I watched the speaker, Benjamin Bratton talk. And was it, let me find his exact words the first time he introduced the concept, there's a slide that said agency preceded subjectivity. Subjectivity discovered its own agency by measuring how much it had artificialized its environment. And that's a lot to think about what he might mean by that. But one thing he means by artificialization, or at least how interpret it is like think about human domestication. And like, you know, we had, we had wild wolves and we had these like, you know, cows and bison who were all wild and we domesticated them and in the process of that went through natural selective breeding to create sort of an artificial version of the animal that suited our homeostatic needs as, as like a domesticated, you know, sapiens. And so you can think of like the cows that we have today that comprise all this biomass on the planet are just artificial cows. I think this, this is kind of like how he's intending to mean it and that these cows can't exist outside of like the, the farms that we have them on and the fields that we, they would just not live, you know, they wouldn't be sort of fit, they're artificial. And this dynamic happens in evolution quite often. In fact, I started to think like maybe this is the way we think about what's happening with AI and AGI going forward. Another example that's like really prescient is the, the eukaryote co op of mitochondria. Like hundreds of millions of years we had these cells that would like basically kind of through endocytosis they would allow other cells in their cell membrane and some of them would have functional utility. And they're these like kind of, you know, ancestors of the mitochondria which eukaryotes had within their cells that would have like these metabolic function of being able to break down, you know, particles and you know, elements that, that had metabolic utility. And at some point the eukaryote artificialized the mitochondria and it embedded its own DNA into its DNA and it produces the mitochondria as its own organelle within the cell. So this happens so just, just like the cow and like our modern domesticated cow is like the pre eukaryote mitochondria to like the post eukaryotic mitochondria. And you know, like the mitochondria has its own DNA sequence, ancestral DNA through the maternal mitochondrial DNA of humans because it's so like distinct as its own. It wasn't invented, it was just like artificialized. So what a great word Right.

Could you talk about a bit more about the AI analogy there? You touched on it a little bit before going to the minor.

Yeah, well, you start to think about like, what. How will AI evolve? Right. And if you just kind of like, if you, if you, I don't know, draw the right boundaries around homeostatic forces and like how, how things kind of evolve over time. If you use something like assembly theory to look at like the befores and afters of. Yeah, I've done this thought experiment. If you think that what AGI is is basically kind of like this replicator of our economic value, like it'll study a human and then it'll like replicate its economic value and then it'll duplicate it like at 1,000,000th of the cost. And so you don't need the human to do that thing anymore. Right. So in a sense, like the language models we built and the companies that are facilitating them are sort of artificializing the. They're using AGI to artificialize the economic value, economic agency that we all have as humans. Where does that leave us? Right, yeah. So I mean, and that's just like a kind of stage one of how this thing evolved. I think Bratton took this like through multiple different stages in this talk. In other words, we could be like the, the pre domesticated cows or the pre domesticated mitochondria, as in the role that we play as knowledge workers and humans. Like, like that. If you think our governments and our, our large institutional organizations will all be like run by AI, then like they will artificialize all the economic value and like there won't be, you know, it can just create machines that provide economic value going forward and, and like humans don't have to do that anymore.

I mean.

Yeah.

That should be a T shirt that actually.

I love that it's, it's the dog too, right? Because there's the companion. Like, but we're not wolves. We're not. And yeah, with artificialization comes down to is like that is that there's some other actor creating an environment that, where we can have complete agency. And it's still like, like the mitochondria can have complete agency and it can do everything. It can from interoception and perception and actuation that it was genetically like or sort of evolutionary like geared to do. It can express that agency, but it lives in a simulated environment. It's a little bit like we talked about the conversation about the Matrix yesterday, you know, like, I think techno feudalism kind of describes like where we're at today. Like we got to get these artificial environments created for us that fulfill these kind of like underlying needs. Like flicking through TikTok, you know, is going to be a dopamine burst that like kind of grabs our attention in such a way that it fulfills like a satiation. And we are independent acting agents, but we live within this bubble of a simulated reality. Right now that simulated reality is going to get like, even more sort of like purposeful.

100%. Somebody just told me they rewatched the Matrix today. And we're like, actually, yeah, it's so. It's so apt, right? It's so apt. Yeah. Did you have more thoughts on that? I liked how you connected the ideas of, you know, the cow, the domestication, the cow, artificialization of the human mitochondria, tying that to evolution in general and then how robo workers are a thing Elon's putting a big bet on. Robo workers are going to be a new category, right? I think we just take the time slider, draw that to 20 years. I could easily see robo worker made in your house or cleaner, right? Robo worker that cleans your floors. I mean, we have Roombas. Like, we're literally. We were trying. We were trying. What do you think life will be like when we get to that? Like, as a human? Like, what's my day to day as a human?

The overall presentation by Bratton made me think that that like what we, the Homo sapien kind of world that we live in today is just a very momentary flash glimpse in the evolutionary history that like, you know, we may become something different than we are today. The Earth will like. You know, I really liked how he talked about, you know, the way that we talk about as climate change, as if we've broken something and something's wrong, is actually just the inevitable progression of computation and technology, you know, blending with, with the Earth. So the Anthropocene is just like, it's on its way to this thing called the new sphere, right? Which is just like a huge collective consciousness of human beings and all the, and all the things that we will become and all the things that we came from all interacting together in this like, computational world.

Dude. Yeah, so. So there was this thing. He said, I wrote this note. It was. I died laughing about the climate change thing. He said, wolfram has this theory about how actually the entire everything is just all down to planetary computation and dark energy is the exhaust of heat. Time is just the refresh rates. And I was thinking Hertz on a monitor, right? Like, I'm like, that's actually beautiful. Like, I could see that. I see what you mean. And the climate change, like, you're saying, is just an inevitable inevitability of that planetary compute. Yeah. Mind blowing stuff. Mind blowing stuff. Yeah.

Have you ever listened to Wolfram talk about his physics before? Yeah, he's next level, man. He's just. He's really just been trying to derive physics from the ground up in terms of these, like, cellular. Yeah.

What? I don't even know that word. I remember I took a lot of, you know, P cam and took some quantum mechanics and stuff like that. That was years ago, though, that I went into software and I stopped thinking about that for a while. So it's kind of cool to think about these things again and see how far has academic thinking gotten. I'll have to add that. Wolfram, Physics lectures.

Yeah.

Cool.

There's a really great talk last year where he compared his framework to Donald Hoffman, who wrote the Case Against Reality. And they basically decided that they're saying the same thing for the most part, except for a couple little nuances, which is incredible if you look at the places the two people came from. Like, Hoffman's a perceptual scientist. He's trying to deconstruct reality from a consciousness first point of view. Wolfram, he released his book A New Kind of Science in, like, the mid-2000s. You know, like, this came from very different places. Both of them are, like, highly generative. Like Toppan saying, like, simple consciousness leads to complex consciousness, and Wolfram saying, like, you know, simple behaviors lead to complex behaviors. And then they have all these kind of, like, theories of evolution in between and why they are what it looks like. And then just recently, Wolfram released Wolfram Physics, where he's got all of his Mathematica software now built to, like, take these. These simple rules and generate physical laws that we can simulate gravity and quantum mechanics and everything that we sort of have observed in the universe. He's got a theory for how that evolved.

Whoa, I need to look at this. That sounds. Next level.

From. This guy is just. He's such a hard worker. He's always filming himself. He does work in public all the time. He's a huge longevity geek.

Oh, yeah?

Yeah. I don't feel jealousy very often, but there's something. Sometimes when I'm watching Wolfram and listen to how to talk about how he works, or I just, like, feel jealousy and I'm like, this guy he talks about. Yeah, like, how he works, his level of energy, his level of cognition. He Just doesn't turn off. Like, you know, just he's, he's just an incredible thinker.

Heck yeah. Cool. Okay. Sounds like somebody else to aspire to. Great. I will definitely listen to some of his talks. At the age of 15, began research and applied quantum field theory and particle physics and published scientific papers and peer reviewed scientific journals. What? Yeah, 15.

This is kind of next level. Yeah.

Oh, it's like that. Okay, so.

But it's funny like Bratton brought him up. He was, he was actually making fun of his description of the universe. Like. Yeah, you know.

Yeah, no it was, I was like, oh, this must be that like academic sort of ribbing right here about that happens. That's kind of funny. Okay, I'm gonna look into him. Yeah, that's really cool. I didn't. Yeah, I gotta get more like inculcated in this. Are there other people you would recommend I look into if I'm interested in this talk?

Well, from he talked with Hoffman. I mean it's the same people we talk about usually. And I, I that that talk was on the Theories of Everything podcast, which is on YouTube. He does a really good job and he gets really good guests and so I think that's good. Like first level filter for big thinkers. Yeah.

Okay, Very cool. There was a really good quote in the talk that I loved that maybe we should talk about a little bit. Sciences are born when philosophy learns to ask the right questions. Philosophies are born when technologies force the birth of new languages. When you, when that came up, how did you feel? What'd you think?

Oh man, I was giddy. Like you know, and then it then a specific phrasing. I like didn't have time to actually understand it fully what he meant in the moment. So I've had it on my list to kind of like pressure test it. I think one, one inflection point I always use as a stress test about like the philosophy and technology is the invention of grammar. In 382 B.C. it was at least labeled as grammar by Plato. And this was like after you had the Plato's Academy. So you had this new technology of written language and you had enough sort of like philosopher free time that there was enough abundance wealth from agricultural surpluses that you could have people like actually like in rooms together writing language for, for like enough time. And then what emerged out of it was grammar. And we didn't have grammar before that. And then, then, then you like after the, after grammar there was about two more generations and you had Alexander the Great where you had These hugely sophisticated supply chains and military tactics that were never seen before in ancient warfare. I don't think that's a coincidence. I think that like philosophy was born out of the technology, out of writing, necessarily so. And did the questions before then lead to the, to the technologies, lead to the sciences that was born. I hadn't thought about that way. So that, that's kind of an interesting thought experiment.

Totally. And then it goes. There's this slide he had about the scaffolding and how intelligence got to create symbolic language to further intelligence. And then that grew artificial intelligence. Right. That we talked about. I think of it that way with, with the evolution of, of language, it's that it's we're growing the symbolic language that allows us to do more sense making. Like we zoom out a little bit that, that allows us to do more sense making. We can get more complicated discourse and debate and discussion and, and obviously they're from their innovations and what you said, like we military tactical strategy or supply chains or whatever. And one thing that was really interesting from his lecture is he basically said in all of human history it's always been a good period of time before you get these cascades back where the artificial intelligence has a feedback loop. We create this, these technologies or artificial intelligence from our sense making, the evolution of our symbolic language. And that feedback feeds back into symbolic language and then the intelligence. Good example of this. I think he said something like we were just monkeys creating fire from a rock or something. And that was the first example of like, oh, fire, heat, warm draw, pictures on wall, symbolic language. Oh, now we could see on the wall because of fire. And that's what I mean by artificial intelligence. Fire creates symbolic language talk and then that leads back into the feedback loop of we get more intelligent and then we keep doing that. And so he said one thing that I want your opinion on, which is for the first time in humanity, we're at this point where AI is accelerating really, really, really fast. For anybody in the industry and watching the progress of day to day progress, even like the whole deep sea thing last week, it's crazy how fast things are. And so humans are using these things to do sense making and to gain more intelligence. And it's a useful tool, artificial intelligence. But now at this point in time, it is a tool that is so good at sense making, it's better than anybody on the planet put together. Maybe an elaboration right now, but in the future definitely not. And it is going to help us create more symbolic language and then feed more intelligence and then all of that is just getting shorter and shorter. So, yeah, when he mentioned that, I'm kind of curious what came up for you there.

Yeah, I mean, a lot of stuff I think even we call AI today was really just, you know, it's just a reflection of the cumulative language spoken by people on the planet. Right. Like, it's an. It's an interpretation and compression of like actual, like, lived experience where, like, we had language and we use these terms in certain contexts and then, you know, so I think, I just think it's a hall of mirrors the whole way through. Right. And like this, in this particular, the reflection of all the spoken written language. Now we have a new technology of. And silicon, where we have transformers that kind of make a reflection of it. And now we're in the process of the whole world. It's reflecting back onto how is it going to impact language now. And so I think we're in a phase now where it's kind of like having language models with human existence, where we're creating more media and language all the time is going to create an evolved set of language patterns that we use. And then those are then recompressed as LLMs and, and so on. And I think we're going through some cycles of that, which is really interesting in itself. But it's just like, it's a pattern that is just, you know, it's ongoing.

You know, once you see it, you can't unsee it. Right. That's the interesting thing about it. It's cows all the way down or. Yeah, on that. When you said the symbolic language thing, I mean, it totally makes sense. Like, AI literally is something that we created that is trained from language and symbolic language to talk, like books and the Internet and content. And that was a symbol or representation of people's lived experiences, perspectives, et cetera. And now it's just like all co. Evolving really fast. So then, yeah, that begs the question that he kind of posed, which is. And I loved when you recommended me this, you were like, this guy said epistemic in the first, like five minutes or whatever. You're going to love it. And I was like, yep, sure enough, yeah, it happened. And then I was like, yep, I like it. Epistemic institutions was his epistemic institutions. So I was actually going to say, he said a very kind of bold stance. He said our epistemic institutions aren't asking the right questions. Like, we're in like a very do now, think later mode in science. And he's saying that we actually, that's a structural problem and we need to at this point, think more, do less collectively overall, was his sort of stance. So I'm kind of curious your take on that.

After you. Yeah, I think it's like, I felt since everybody has like, you hear researchers talk about how they do grants to get money for their labs and then to get real work done, they sort of like find some hack either within a grant or outside of a grant. They know they're not asking the right question when they're writing the grant application. So the funding misalignment with like research is just evident and it's a big problem. And so how do you fix that? And I think a label like epistemic institutions is like a really kind of cool, progressive. I felt this in my own career. Like, I didn't want to go into academia and have to like, sort of just follow on to somebody's research that I didn't really believe on in, in order to like, build up a pedigree and then get frustrated and then leave and go to industry and then work for a similar kind of like, you know, trapped in a specialization where you're, you're getting paid a lot of money, but you're not working on the right problems. So how do any of us ask the right questions every day or work on the right problems every day is a question. It's like one of these things, like, I don't know why we're not asking ourselves that all the time. And I think if we did like that, our political structures could sort of respond to it. But I think a lot of us are sort of just kind of like beholden to the techno feudalism. You know, hey, I have a nice life and I have money for my family. And, you know, those are the things that we typically worry about. And it is a big luxury in an entitlement to be able to say like, I'm alive and I get to ask the right questions. Right. So I think this is a unique period in time where like the epistemic next generation epistemic institutions is like, needed for the first time in history and it's needed more than ever right away.

Yeah, yeah. I really like the overall line of thinking that the symptom of people on the cutting edge not being able to ask the right questions is it's a symptom of the structure not being in place to allow them to do that. And what you kind of touched on is that the intersection of capitalism, we're all chasing paper. We all got to feed families and have Shelter and stuff like that. Unless you're ballin, you've beat capitalism, you could eject. Most of us are still in that, so we can't exactly do that. But I think what he kind of showed is you can go find people that have ejected, that have a lot of wealth, that are interested in funding that kind of thing. And that's actually from an entrepreneurial spirit front. I could see like wow, this guy's had decades experience navigating academia grants. He was frustrated with it, he wasn't satisfied with status quo. So he created a new model where this kind of innovation could happen. And I really respected that about him. Yeah. And yeah, I like that term epistemic institution. I think that embodies not just, you know, academic institutions, but institutions that cultivate an environment where you're asking the right questions. Like that's really it. Right. And that could exist in private industry. And it's funny that you mentioned the whole paradigm of as an academic researcher in a typical academic institution, because of the incentive structure, I must constrain my question asking to hop on grant money where trends are to fund the question I actually want to work on. And so I've heard of all sorts of stories of like PhDs working like two PhDs essentially kind of just like wanting to get their baby done, like the thing they really want to ask, but then having to navigate all these politics to get money to do another thing to like hopefully move the needle on their thing a little bit. And, and if you look at that and you zoom out, it's like wait, we are just constraining literally the smartest people on the planet from asking the right questions. So that's fundamentally wrong. And going back to the philosophy thing, I think right now, tying it to, we're at this point where intelligence, symbolic language, artificial intelligence, because superintelligence is being born, those are co evolving so fast, it's necessary to get the best thinkers thinking about these things. Um, but, but we are constrained. So when I saw that because we, we ourselves, we're thinking about like okay, how can we appeal to the ICP of the philosophically values aligned kind of technical contributor or researcher or scientist or practitioner around the I guess human expertise fields of like philosophy, neuroscience, like metaphysics, all those and like technical contributors. So we even have this problem of like it'd be sick if we could keep exploring this problem. These, these questions if how will AI understand us? We don't fully understand ourselves. Can we keep asking that? And that's going to take a lot. And so One of the things I'd love your opinion on, I've been thinking about that a lot and figuring out what's going to be the model that takes us forward. I think obviously go hard at Brian AI right now and that is our best go to market motion right now. But thinking bigger picture, what does the monetization actually look like when we land the next cloud client and the next one and how does that look like in terms of funneling contributions to the open source project, making sure that grows at a healthy evolutionary track and then making sure we could sustain this? I've been thinking about that a lot and it was so cool that this guy actually just made it happen. He was like, ah, no structure exists. I'm going to go make a structure where we could do this. So I just thought that was so cool. What do you think of, of all that? I know I said a lot.

Yeah. And I guess I forget that the organization that funds his program, I think it's called the, the Berggren Institute. So I was thinking that during his pitch he must have had made a very good impact on this institute, known people that were where they really gained trust so that he got enough like liberty to create the anti anti Kytheria institution group. So if you talk about epistemic institutions and what they should be is I think our institutions should be able to ontologize existential risk in the same way we think about longevity, but also at a societal level and make sure that the right questions being what is the knowledge we need to gain in order to protect our society from existential threat, in order to continue to grow and flourish through evolutionary cycles and be able to sort of not just see what's on the horizon, but see what, what change looks like and like, you know, what kind of questions should we ask today so that you know, and then put financing into that. And I think like one of the ways that you do that, like how do you understand like what matters within a population for instance. I think that's where it comes back to representation systems. And if you have representation systems and, and values that you know, are well understood within those representation systems and you know, we, you think about how like how dumb it is right now or like, I mean we can't take for granted democracy but like just at the pace that it moves relative to existential threat like that we go to the ballot at a certain, like once a year, once every four years, cast a vote on things that everybody knows we're not asking the right question when we cast votes, like everybody Knows it.

Well, I guess I shouldn't say everyone, it's not 100%. There are brainwashed people deep in the matrix that think they know and so they think they're right, you know, but.

In general, yes, everybody that I talk, you talk to probably has like got this feeling that my vote isn't voting even for the right thing. It doesn't matter. Right?

No, it's a sham. Yeah, it's a sham then.

But then if you took it, if imagine that like all of us are asking the right question all the time. All of our needs are represented within like, you know, all the homeostatic processes in our body, our supply chains all have representation. And all of this is being fed back back into epistemic institution in real time basis that can make sense of it and then like prioritize what are the right questions to ask. That's what like a next generation epistemic institution looks like to me. And I, you know, so this is a collective intelligence. And I got so interested in like what examples of collective intelligence look like. I studied the transition from unicellular animals into multicellular through the basal ganglia specifically. And if you look at how the basal ganglia works in our body, it started out and it pretty much still is a representation system for all the homeostatic needs, you know, of our cells and our tissues and our organs. Right. And so it's, I think that there, there's evolutionary precedents for like what an epistemic institution could look like at scale.

Yeah. Like just from that. Because we always take this hypothesis that we're all just agents, we're made up of agents and collections of cells. Right. So just from that example, you're saying there's obviously like kind of parallels at a bigger organization or collective level. Right? Am I getting that right?

And I think a sophisticated representation system sort of starts to look like a superorganism in a lot of ways, where it has its own agency and it's its own like sort of sense making apparatuses. It can ontologize, understand existential risk and then like understand what's the next best question to ask and what resources should I actuate in my environment to do that? And when we answer those questions.

Yeah.

And we say, oh, the next best thing for me to do is go, you know, like eat some turkey in the fridge. Right. We could actuate that and like go grab the turkey and chew it up and digest it and feed it to our bacteria in our stomach. And when epistemic institution is asking that question, it can it can allocate financing and it can, it can create training programs and it can, you know, like it's the same sort of actuation to solve the next best question. It just has, you know, it understands what is universally needed in order to make it happen.

It's a different implementation, but essentially it's the same thing. I need to make sense of this to make a decision. I need to ask the next best question to make sense of it, to make a decision to actually either from a collect, like you know, that organism, the, the ganglia, what was it called?

Ganglia, yeah. The older part of our brain. Yeah.

Basal ganglia. Yes. Versus like, okay, just take a different collective epistemic institution. It's just different. Deploy capital. Makes sense. Actuate versus like yo, I'm in this body getting, you know, sensors or data inside this body. Okay, activate. You know, it's a little different but like essentially overall it is the process. Let's do a little exercise. I think this is fun to dream up. What does the next generation epistemic institution in the future look like? So now we have some anchored thoughts around how this guy did it. And then you just gave us a little nugget which is like envisioning a world where real time, we're getting data tracked at every cell, organism, agent, person, human on the planet. And then we could feed that some way in real time to some data set back to an epistemic institution where they could review that, hopefully have like context of that in different almost verticals, like the different sciences departments, I guess, if you can think about it from an academic institution perspective, departments. So there's obviously AI in that. There's obviously. I'm thinking there's a lot of requirements there. But like without getting into the, you know, implementation too much, what other things could we do to like really shorten that feedback loop? Slash, make it even more accurate. Slash make the sense making of it faster and more accurate.

Yeah. You know, should we come up with an example of like what this institution is tasked with? You know, is it.

Yeah, you know, let's use the longevity case because we're really working on it and I see a path there. Right. So starting with Brian AI, let's say he gets a million members by end of year using IT and collecting health wearable data and non significant population of that sub population of that population compared to being biomarker data, other demographic data like gender, where you're from. And then using epistemic me, we contextualize a lot of belief system data too, against Those things. So let's use that as the use case. And then the epistemic institution. Do you want to pick what their goal might be?

Yeah, I think that's a great example. I think maybe just say protecting a certain population of humans from existential threat through a health lens, you know, I think it's a good thought experiment.

All right, so, okay, so now we got a little bit of darkness. How are you?

Yeah, I think, okay, let's say we have, let's call it like 50,000 humans, you know, and then we have an AI that is meant to like aggregate everybody's needs and sort of like it's the accumulation of the people and all the biology that's within that organization plus the AI. But the AI will automatically weight everybody's input and then come up with the next best question to ask and determine what the next action that should be taken. So I think kind of like in the political structure in that sense. And then I think about like, well, okay, what data do we put in? You mentioned biomarkers, right? I think we're really like taking a self model of, of every human and breaking it down to like all the homeostatic functions that we need. Like we talked about hearts and lungs and blood and muscles and you know, at sort of a higher level. And then below that I think, you know, you have all kinds of cellular interactions and you can imagine that like each individual person could be represented by say like a hundred thousand self agents that are all communicating data about their homeostatic needs to survive, you know, and so you want, you know, if you have, if you have cancer cells in your body, they, they are part of this representation system that is saying like, you know, hey, I'm a threat or these cancer cells are a threat to me or something like that, right? And so then you, you're going to bubble all those things up to, to a set where like what the AI is going to be looking at is kind of like these inter, subjective representations of all the agent beings within its representation network. And remember, if you have a hundred thousand per one person, there's 50,000 people. You have millions and millions, like billions of representatives that you need to account for. And so then you're going to take simulated realities and you're going to have to understand the environment in which each of these agent independent agent actors works in. And if you make changes to the environment, what is going to be the outcome of that? So this AI is going to need to run sophisticated simulations across this population set and then it will understand like do we need to take immediate action such as like cleaning the air that these people live in, or do we need to take longer term actions such as researching interventions that are pharmaceutical? Um, and I think that the common thread with all of them is sort of metabolic. Like if you can, if you can understand that holistically we can keep healthy metabolism amongst this group of like 5 billion, you know, representatives in my, my agent system and then I can weight which metabolism sort of like is most important for survivability of the entire entirety. Then I think you've got like a, an epistemic institution that's working on top of a representation system that is holistically sort of like a consciously alien agent actor works towards group survivability.

Okay, so just to piece together what I'm understanding, to solve that problem, we would need to allow the epistemic institution to be able to ask the right questions and get the right answers 100% of the time really fast in real time. So that means that there's a requirement of allowing them to run simple to highly complicated simulations on target population and have sufficient data for that to have pretty much 100 efficacy of representing that target population in the real world. Did I get that right? Like that's, that's the overall thing that's like required. Like that's the foundational that's required.

Right. I mean it gets complex in the details because just like our individual sensitivity. Yeah.

I was just thinking how do we break this? Like literally like let's make it more complicated. So let's say like, so we only were looking at metabolic health. So now let's say, because the bigger question is the exist existential risk. So expanding scope a little bit set of metabolic health. Maybe it's on a micro scale, smaller scale. Like a bigger scale would be like how do we incorporate maybe the fact that nations are still waging war and some of these populations live in a war zone. So that's, that's one use case I just had in my brain. A further off use case is like, shit, what if there's a meteor coming? Like, or we don't, there's something unknown coming. So that begets the requirement of oh, okay, we've got to have sensors up the wazoo in the universe, like somehow. But micro interaction to break the model and then let's evolve it. This is a fun thought experiment is being around people can also affect like very toxic people can also affect like your well being, which affects your stress, which affects your emotional well being and stuff. And that could Lead to some level of risk too. So maybe start there as a use case. Then we evolve into the, maybe you're in war zone, the meteor use case. How might we take that into account?

I think it starts to look a lot little like the needs of a nation state. Right. Or like, you know, because you need to.

Yeah.

Well, if we're needing to maintain the metabolic health of all these 5 billion agents, then like, let's make sure that we have nutrient supply chains that are or like well stocked not just for today and tomorrow, but for like weeks or months in the future. And you know, like one interesting heuristic is like how, how long could a country survive today if it's like shipping channels were cut off? And I think in like Taiwan you might have like 30 to 60 days worth of food or something like that. So this is something that like, you know, an epistemic institution that's acting for, for health purposes would want to like, extend out pretty far. But then on the other side of that, I think you have, you know, you have actuation of, of knowledge, work and physical labor that like you need your population, the aggregate metabolic sort of energy to be applied to, so to like, you know, act on the actuation of the system. And so you're going to want to be taking measures to protect and make sure that people are in psychological conditions and you know, like working conditions, that they're, they're, they're safe and they're able to perform tasks, you know, according to the benefit of all, um, you know, assuming that they're actually responsible for keeping, you know, like we're taking internal resources to like evolve health needs. Right. Um, so I would start there like, like from the bottom up, like metabolic supply chains and from the top down, like real knowledge, work and, and physical labor and productivity gains economically. Uh, you know, and that starts to look a lot like a nation state, at least as I picture it.

So, so then, so then to solve the war problem, it'd be model a bunch of nation states that reflect our reality.

Yeah, because I think that's what happens is that you have, you have a bunch of these actors and there's like limited resources for metabolic health of their organizations. Right.

But that's literally life right now. Yeah, I mean there's ego and pride and narcissism thrown in the mix. But like, yes, those are just helpful.

Tools to, you know, like to solve the underlying resource dependency or constraint. And that, that's why I think like, you know, having abundant in energy is one of the things that unlocks so now you don't have to optimize these epistemic institutions to like be militaristic, you know, or at least like, I think boundary systems is something that all homeostatic life has. And, you know, if you don't have to allocate all your energy towards like gaining new ground, taking new territory for resources for your survival, then you can put it more into the epistemic knowledge work of your system and do more research to reduce other existential threats.

Yeah, if we don't gotta worry about the lower level, Maslow's hierarchy of needs. Yeah. Then it opens up the mind more. Back to a parallel thing we were saying. It just allows more people to be asking the right questions and try to solve those things collectively and, and the accumulation of that knowledge work over time, I think in your words, will help the system actuate in the quote, next best way or right way. Okay, now what about the meteor thing? Is that in my mind, I'm like, all right, we really can't fuck with that. Like, we probably need somebody to go figure out how do we even sense things that are. We could sense things that are, I think, hundreds of light years away. But then some of the problems are like the image processing, image quality and like speed to get image. So then maybe this epistemic institution has to evolve its scope and be like, we're going after that now to even get the right data to make sense of it.

Yeah.

What do you think? Do you think that's foundational requirement?

By the way, did you see that headline that the risk of an asteroid hitting Earth, and this was particular asteroid, they increased the risk from like 0.2% to like 1.6% yesterday or something like that. Yeah, it was in my news headline.

Wait, what? Wait, wait, that's a. There's a more than one in a hundred chance that we're just going to get hit by a meteor. How many?

Oh, it's like 180 miles wide or something like that. It's huge. Yeah.

What? Oh, shit, that would be like northern, like a fifth of California. Like, hang on, that's a lot. What's the square mileage of California? This is also funny. I just asked AI because in the past I would have had to go figure out the square mileage of California and then do some calculations, conversions. AI just did that for me in real time. We are showing Benjamin's model is correct. I am making sense now from the artificial intelligence and I'm evolving my own intelligence to evolve the language that I use in this podcast, California has a total of 163,000 ish square miles. 180.

180 miles per hour.

I mean that's, you know.

Yeah, I mean that.

Yeah. So then it's 180 by 180. Right.

So let me just look up Thargo.

Yeah, That's a lot. 32,000. 32,000. Oh, that is about a fifth. Wow. Thank you math teachers from the past for giving and engineering teachers from the past giving me some level of good estimation. I'm kind of proud of that. Sick. Wait, that's, that's almost actually off a fifth. Wow, thank you math teachers from the past for giving. An engineering teacher from the past giving me some level of good estimation. I'm kind of proud of that. Sick. Wait, that's, that's almost actually on the dot. One fifth. Exactly. Like almost. Cool. Oh, so it's.

It's only 130 miles, bro.

Like that's. Wait, 100.

Yeah. And it's 1.3 chance of risk and it's 20, 32 where it would have potential impact.

Okay, that makes me feel a little bit better because 130ft. I could better imagine that. That's like.

It's a building.

Yeah, that's like a building. I'm just thinking about my apartment square feet. Okay, cool. I don't think we're gonna die from that one, homies.

You're like a giant.

We got this.

Yeah, Chipping.

Cool. Sorry to scare everybody on the pod. We're gonna be all right.

It's like a flight or. Sorry, flight kind of trade off. Right, like where you know, from a sense making perspective, there's all these things that we need to be doing at a higher level, cognition wise, like finding food mates in our world. It's, it's knowledge work and then all of a sudden there's an asteroid coming at you and like nothing else matters. Right. Or if it's a tiger or whatever. And so it then it just shortcuts epistemic capability, you know, and you have to have sort of like pre programmed routines in your survivability history, things that your ancestors had already solved to figure out this problem that you're, you're relying on. ChatGPT won't help you.

Yeah.

With the tiger in front of your face. Yeah.

No, but maybe.

And that was like what you're talking about the back propagation of like, of evolution.

Right?

Like if we get our AI plugged into the sensory loops that happen at like the reflex, reflex level of our nervous system, you know, say 50 milliseconds and below, you know, like, yeah, that can absolutely be like an, an adaptation that could evolve. Years of evolution have taught us that we've now figured out like how to back propagate even more knowledge into, into that system.

Definitely. And as you say that, I mean there's also the other take that, you know, humans and AI evolve in this hybrid way together. So like one example I always think about what's that boat? Where if we replace every part of the boat, is it still the boat? There's some boat. There's that philosophical problem. I forget. I'll add it to the show notes. And so one of the things as I'm listening to Lifeforce, the book by Tony Robbins featuring Peter Diamandis, is there's all these like therapies going on. And one wonders when, if you replace every part of the human body with some synthetic part, you're something new, right? Like you're not Homo sapien anymore, like I'd imagine. Especially if you tie it into literally like this has instantaneous reflex level feedback loops to your brain to do things with you sense making, actuating, then. Then you've come become something different. Yeah. Well, I'm sure you've thought about this. So do you have any evolved thoughts on, on that or additional thoughts on that?

Yeah, it reminds me of my, the, the Consciousness Complexity Index, you know, like where it. Yeah, like if you, if you hold agency for what it is that that makes you conscious, then and you don't like reduce in complexity from that and in fact you increase in complexity from it, then that seems largely value directed. If you have, you know, sort of like an evolutionary lens on, on life and yeah, you could replace yourself piece by piece, but what, what can't be replaced is the ongoing conscious loop of consciousness. Since making an agency that you have between yourself and your environment and that's where you know, like you could easily just kill somebody and create something different in the process of replacing them piece by piece. Or you could have a measure and I think that measure of the Consciousness Complexity Index, where you really have a mutual information score of all the interoceptive and extraoceptive processes that are going on. What the environment is, how we control it, what the complexity of the environment, how many objects we can manage at a time in our interception, how complex internally can we respond to our environment. Like if you can maintain that, then you can evolve it.

It's like the complexity, that factor is essentially like a weighing factor to balance right in this kind of complicated web of systems for the audience that hasn't heard about Jonathan's cci, could you talk about that? And you touched on a little bit about interoceptive extroverts. So it's sense making feedback loops. But can you just talk about a little bit of the origin of why you came up with it and then what problem does it solve?

It's interesting because I was mentioning, I was watching this Bratton presentation you're starting to get on topics of evolution. And I've been working on this idea about assembly theory of consciousness. And then he introduced assembly theory into this presentation, which. So it's just like I was there, I was like, this is awesome. And the, okay, so the, the assembly theory of consciousness, it was a, a problem solved. Like, imagine you have data from a distant solar system through a telescope and you have like, you know, pixels with colors on them and stuff like this. And you're trying to understand like what molecules might be on a planet in this, in this distant solar system. And so there, there was a guy who's, that was. His job was like, figure out questions like that for NASA. And he like, realizes there's this inherent relationship with chemical particles that like, the more complex they are, the more interactions they have gone through. And so he likened that to the process of evolution. Like to say like, that each individual interaction was sort of like an evolutionary step or turn in its evolutionary cycle. And that the more complex the molecule, the more ancient it must be, the more complex it must be, the more evolutionary cycles it must have gone through. And so it. He came up with a formal, like, he calls it the assembly theory. And you know, it's about how particles are assembled and a chemical from a chemical, you know, scientist. And. But it applies to then how biology happens, right? Like, because it's just really the same kind of mechanisms. And this is something that I like, you know, I just, I just love to study evolution because in the sense that it's our history. And I really like kind of, you know, spend a lot of time reading historical narratives of how these especially breakthrough evolutions have happened. And I saw the same thing. It's just like, there's always like, it's not that evolution goes linearly through these more complex adaptations, it's just that you tend to get additive adaptations that find fitness with more complexity. And sometimes you get more complexity that doesn't find fitness, but the more complexity gets passed on, gets integrated and so on. So anyway, I like that theory and I just thought like, I was, I was trying to talk about Michael Levin's Work and like how to explain. Because Levin's work has to do with like cellular biology and how cells like have ion channels that form bioelectric fields that can like build our tissue systems and our embryonic process. And I was, I was in a forum where somebody was talking about, well, his theories really don't approach how stem cell drift happens. As in like the stem cells in our body have actual epistle, epigenetic and genetic drift over time, which is sub intracellular. And I thought like, I think Levin's philosophy actually talks about this. His science doesn't as much, but his philosophy does because it's like about how things assemble. And so I thought how can I articulate this? And when I was, I was in a forum where somebody was talking about, well, his theories really don't approach how stem cell drift that like, you know, we can see this. And, and I kind of, you know, went on chat GPT and created this theory about the assembly theories of consciousness. And it was fun because I just like you wouldn't ever ask a human being to do this, but I was like, I need a theory that explains all of conscious life through history and how everything evolved and how everything's ancestors are defined. And I just listed out these absolutely crazy. And I said, these are acceptance criteria.

Here you go, intern, figure it out.

And it says, okay, here's the assembly theory of consciousness. And it like basically said, I've solved all your acceptance criteria and here's a Python script that demonstrates like this how things evolve. And it noted like, this is not an empirical model, this is just a thought experiment. But like for like, you know, 30 minutes of work on my half, like that was pretty good. And out of it came this idea of the consciousness complexity index which is like, so assembly theory has a complexity index. And so I just applied that to like agency to say like, you know, um, and as you know, I just have enough background in thinking about agency, subjectivity and consciousness to have a mathematical model to say like, I think this is calculatable.

Well, what's that math model look like? Well like in, in assembly theory, I might, I'm not sure.

Oh, the assembly theory. Um, oh, there's, yeah, basically just like, you know, like it's an information theoretic number. I forget exactly the math that's used. But it just says that like here's, here's the complexity of the molecule of water. Here's the complexity of the molecule, of the, of the element of oxygen, you know, and like you can get a number one's higher than the other.

And so he's coming at it from. So I'm really intrigued because I have a chemical engineering background, so I used to love thinking about chemistry all the time. So it made me think about the analogy of a tree. We cut it, we see the middle, we could read its history. Right. And. But. But this is like the lowest level, tiny tree. You have inside you the molecule, molecular level. Right. I mean, there's an even lower level. But, like, at this level is quite sufficient, I think, for this. This theory. So just to get my thinking straight, is that every element and molecular compound has a different number, or is it also the actual matter, like, not just the compound in terms of theory, like on the periodic table or in organic chemistry problems and equations. But is it actually like, let's say, for example, I don't know, even this piece of paper has some carbon, Right. I've written on it. And now that carbon's changed in some way, it's been interacted, there's been some abrasion. Right. Its molecular structure has changed a little bit. So does this theory account for that change?

Assembly theory?

Yeah.

Yeah, I think. I think so. I don't know exactly how it would sort of like categorize the ink on the piece of paper, if that would be its own sort of like metamolecule or whatever. But the idea.

But it's something in that direction, though, of, like, matter is changed or aged or something. And at a molecular level, we can actually attribute a level of change. And that's kind of assembly theory or.

The complexity of that. And even an example you gave, like, where this. Where'd that paper come from? You know, it, like, it grew from a tree, you know, and how did the tree get there? And like, it's got its own history. But then. So every single sort of step where the molecule gets more complex is. Is just, you know, part of the assembly process of. And then the last example you gave was, like, putting ink onto the piece of paper. It's just like, that's the end of the process for right now, right?

Oh, well, is it? Because now it's in my mind what I've written down. Right. Even lower level, maybe not. Right. Because. Because, like, there was something that happened in my brain in terms of molecules, and when I wrote on the thing and made sense of the lecture. Right. So you. One could argue, like, there's some transitive property there.

Yeah. And that's like. I don't think the assembly theory directly would approach that, but, like, that's why I think you need assembly. Theory of consciousness is just like, you know, because there is this world, you know, world consciousness kind of relationship that is like I say, it's a hall of mirrors. Like, the more you look at it, the more it's just like continual reflection back and forth of like world and.

Self, you know, I always think about that first picture you showed me when we met. It's just that human mind and the earth and the perception loop with the disruption and the beliefs. I literally think about that shit all the time now. Yeah, all the time. It just pops up in my head. Yeah, it's really good. It's on our GitHub. For people that are wondering what it is, just go to our docs. It's like one of the first images.

So thanks for categorized the ink on the piece of paper, if that would be its own sort of metamolecule or whatever. But the idea.

But it's something in that direction though, of like matter is changed or aged or something. And at a molecular level we can actually attribute a level of change. And that's kind of assembly theory or.

The complexity of that. And even an example you gave, like.

Where they're explaining the consciousness complexity index is.

Was that it?

Consciousness complexity?

Yeah. Yeah. You know, again, not a formal theory, just a pseudo theory.

No.

Yeah, it sure does make a lot of. Sure does make a lot of sense.

Yeah. You know what's funny is in you doing that, you are doing that thing that he talked about, which is sense making or symbolic language is co. Evolving. Literally this ChatGPT gave you the CCI thing and you're like, oh, okay, this new symbolic language. And that fed back into you to improve your intelligence, to then interrogate it more, the artificial intelligence with that new language. Literally, that's exemplified. Yeah, right there. It's. That's bad. It's so meta. It's so meta.

It's. Where'd that paper come from? You know, it like it grew from a tree, you know, and how did the tree get there? And like it's got its own history. But then. So every single sort of step where the molecule gets more complex is. Is just, you know, part of the assembly process of. And then the last six example you gave was like putting ink onto the piece of paper. It's just like that's the end of the process for right now.

Mirrors all the way down. Or now I'm thinking it's cows. We're all cows. Yeah.

And I think exactly what you described is like the why now? Of epistemic me. Like, I Think consciousness was too complex to like really quantify before AI, before large language models. And now we have a chance to kind of like look at all this data of agency and consciousness through this lens. And that's, that's why we need self models, that's why we need subjectivity in ways that can be expressed through language. So that.

Right. Oh well, is it. Because now it's in my mind what I've written down, right.

I can help us, you know, like really understand the theories of consciousness subjectivity.

You know, how at a, at an even lower level, maybe not. Right, because, because like there was something that happened in my brain in terms of molecules and shit when I wrote all the thinking about this. What, what pops in my head is like obviously as me and you talk about this, we're deep in the thing and made sense of the lecture, right? So you. One could argue like there's some transitive domains and we love nerding out about them and so we get each other. There's not many people that I could talk at this level with about this. So basically we've got to go find more. That's one of our prerogatives. But one of the other things is we as a team have an opportunity to educate the masses on this problem, on the opportunity. And so so far one of our biggest things that we've learned is validation that our copy around positioning of how will AI understand us if we don't fully understand ourselves? People latch onto that question really well. That's been a great thing. We validated in market and the reverse actually hits too, which is if we don't fully understand ourselves then how can AI possibly.

Yeah, and that's like I don't think the assembly theory directly would approach that but like that's why I think you need assembly. Theory of consciousness is just like, you know, because there is this world, you know, world consciousness kind of relationship that is like I say, it's a hall of mirrors. Like the more you look at it, the more it's just like ever understand us.

We have a variation of this on our website and I'd like testing, tested out variations of that.

But continual reflection back and forth of world and self.

You know, I always think about that. Basically those two questions are how we've been explaining this in market without like a demo to show. But a fun thing I just thought about in my head was I always think about Brian Chesky's like Pixar, imagine the 10 star experience and then work backwards to like what you can actually do. But like you should imagine that. And so I'm actually imagining, you know, in yesterday's when we did a pod yesterday, we were talking about, oh, what if we had a gamified. Here's the human anatomy, here's a heat map. We modeled everything. First picture you showed me when we met. It's just the human mind and the earth and the perception loop with the disruption and the beliefs. I literally think about that shit all the time now. Yeah, all the time. It just pop your body, all the biomarkers, we get all this data in real time. And then you could see this heat map as a patient with your doctor or with your fitness coach or longevity coach or doctor whatever. And now I'm thinking of this. Yeah, it's a. It's really good. It's on our GitHub. For people that are wondering what it is, just go to our docs. It's like one of the first images. So thanks for to explain it to the science community because there's in. In this lecture, right. There's a. It showed me there's a large amount of people that are at our level of thinking. Like this is the thing to talk about. And so I think if we just posed our simple questions, people would get it just watching that lecture. But then we go outside to like my mom, like she's like, you know, Vietnamese immigrant lady, she's just worried about putting food on the table or you know, like my brother and just explaining the consciousness complexity index.

Is. Was that it?

Consciousness complexity?

Yeah. Yeah. You know, again not a formal theory, just a pseudo theory like oh no. Yeah, it sure does make a lot of. Sure does make a lot of sense.

Yeah. You know what's funny is, is in oh and being a caregiver for her sister. And so she's like, she's like got immediate thing you doing that. You are doing that thing that he talked about which is sense making or symbolic language things to think about. And so to crack that is like really leaping from innovator to early adopter to early majority to squarely like late laggards. I would say like the last one. So I had this thought in my head which is maybe we have a video that shows like. Like once we do this in. In 10 years or something, once we actually can for a is co evolving literally this chat GPT gave you the CCI thing and you're like oh, okay. This new symbolic language and that fed back into you to improve your intelligence population for health and longevity. We've mapped out everything to maximize metabolic Health for the epistemic institution use case where their whole goal is for a subpopulation to keep them alive. Let's say we do that and something I think that would be so cool is we show. And I got this idea from. From interrogate it more the artificial intelligence with that new language. Literally that's exemplified right there. It's. That's man, it's so meta. It's so meta. It's Benjamin's presentation where in his institute they combine design with research. So the research is presented in this super sexy cool way that actually provokes awareness and mirrors all the way down. Or now I'm thinking it's cows. We're all cows. Yeah.

And I think exactly what you described is like.

And then people are hooked in. They read the research and I thought that was wonderful. And that makes a lot of sense for our thing because our thing is very much on the abstract thinking side. Let's say we ground it very easily with AI Health coach. I think that's why it's a good use case for us. But like abstractly thinking we're in this world.

The why now of epistemic me. Like I think consciousness was too complex to like really quantify before AI, before large language models. And now we have a chance to kind of like look at all this data of agency and consciousness through this lens. And that's, that's why we need self.

We've made sense and have created models of selves, have billions of them for this 500k population or whatever. And what we could do. I'm imagining we go to the World Health Organization or you know, know the un. We're like, here's a sick video that shows I'm imagining in my head pictures, images of like PE models.

That's why we need subjectivity in ways that it can be expressed through language. So that people.

And then you know, zoomed in on one that anatomy of a person, maybe it's a woman, a man.

AI can help us, you know, like really understand the theories of consciousness, subjectivity.

You know, how maybe two. And then it has all these like things of like, oh, we have this data, it's version. It's in real time thinking about this. What, what pops in my head is like obviously as me and you talk about this, we're deep in the resting heart rate, heart rate variability, things around your cortisol levels, just everything about you, everything we can measure. And then we zoom out a little bit and then we see adjust domains and we love nerding out about them and so we get each other. There's not many people that I could tie a population of that on a globe. And it's like here in a continent or something and it's like cool, we've lit that up. It's now lit up. We have all this data and then it shows some sort of at this level with about this. So basically we've got to go find more. That's one of our prerogatives, existential risk score that says oh, the risk of this is this. What's the next question to answer to drive that down? And then it kind of ends with this fog of war across everywhere else that we have. But one of the other things is we as a team have an opportunity to educate the masses on this problem, on this opportunity. And so so far one of our biggest things that we've learned is I haven't mapped this out. And there's this promise of if we do map it all out, if we clear out the fog of war, it's not just this like little blue continent or zone that we've lit up with data. It's everywhere. And validation that our copy around positioning of how will AI understand us? If we don't, then I'm imagining the video expands to the universe, the galaxies like the stars and shit. And it just keeps lighting up in this visual way with the same color scheme in my head to demonstrate fully understand ourselves. People latch onto that question really well. That's been a great thing. We've demonstrate the idea that the story of this is we're trying to validate it in market and the reverse actually hits too. Which is if we don't fully understand ourselves then how can AI possibly ever understand us? We have a variation of this on our website and I've like tested, tested out variations of that. But to measure in every way possible, anything we can possibly measure to make sense of, to protect yourself from not dying. Anyway, so that was in my head of like what can I show my mom? That would be like this is what we're doing. And then she sees it, she gets it and she's like oh. Basically those two questions are how we've been explaining this in market without like a this is really important. This is like humanity saving level important. That's the sense that we want to give people mo to show. But a fun thing I just thought about in my head was I always think about Brian Chesky's like so thanks for sticking me with me through that. I had this whole thing in my head I wanted to get out from what you were saying, do you have anything to add to that? I was sort of imagining that as the ten star education experience.

This is really cool. It's a similar kind of goal in mind when I've made that star.

Imagine the 10 star experience and then work backwards to what you can actually do. But you should imagine that. And so I'm actually imagining, you know, in yesterday's. When we did a pod yesterday, we.

Were talking about longevity radar, risk radar, and I just thought, yeah, like, you know, but you're just taking. You're taking it to like the Nth level, right. Where you can see.

Yeah, yeah.

Because you want to.

Oh, what if we had a gamified. Here's the human anatomy, here's a heat map. We modeled everything shows some correlation between.

Like, here's the thing we can measure in the body, and here's how it relates to your risk, you know, of. Of cancer or heart disease or, or, you know, anything that's like, exist.

Your body, all the biomarkers, we get all this data in real time. And then you could see this heat map as a patient with your potential threat.

And then they kind of go, like, there's different. They happen. There's their cognition and doctor or with.

Your fitness coach or longevity coach or doctor Whatever. And now I'm thinking of this, like.

Say, getting something like Alzheimer's disease, like, it's kind of like game over. But, you know, you got. But then you have things happening, like epigenetic cell drift in your stem cells, which takes. We're not.

We don't explain it to the science community because there's in. In this lecture, right. There's. It showed.

We don't really have to worry about dying from that until we're 150 years old. So, like. Yeah, you know, like, there's a large.

Amount of people that are at our level of thinking, like, this is the thing to talk about.

And which one are you going to pay attention to first? But then if we're looking at this and we're trying to like, really get people to phenomenologically feel that, like they're in a population of people that are all experiencing these things at once and you're. You're going to like, visualize that and kind of like spotlight it like you talked about. Yeah, yeah, that's. It's really powerful vision.

Yeah. So I think if we just posed our simple questions, people would get it just watching that lecture. But then if we go outside, I'm going to try to make this shit.

Soda like.

What we can make this side to like my mom, like, she's like, you know, Vietnamese immigrant lady, she's just worried about putting food on the table or you know, like my brother and just. Oh, and being a caregiver for her sister. And so she's like, she's like got immediate things now. Yeah, I'm going to try to get AI to make this and then you know what it's going to do? It's going to give me new language in a video that I will increase my intelligence on and then make more sense of it and then, then make the AI make bullshit. It smears all the way down, Jonathan. Smears all the way down. It's a quick time check. So we've been going for a while on this. I actually totally think we could have Dean watch that video. He would love us to think about and so to crack that is like really leaping from innovator to early adopter to early majority to squarely late. Oh, laggards. I would say like the last one. So I had this thought in my head which is maybe we have it and then we do another one just on this and go a little bit deeper with more prepared questions next time. This is thoroughly enjoyable to talk about and. And also we just came up with a deliverable on my plate which is good or bad but like I'd say it's good overall. It's more work but I, I definitely see something like that would bridge the education gap with the laggard late majority Persona for us. Like it's. It's harder to explain to somebody like my mom a video that shows like, like once we do this in, in 10 years or something, once we actually can for a population for health and longevity, we've mapped out everything to, to maximize meta. But if she watched something and it was told in this very narrative way that provoked human emotional connection, then I think it would draw her into the story more. More like everybody can relate to health for the epistemic institution use case where their whole goal is for a subpopulation to keep them alive. Let's say we do that right. Like she's very interested in health at this point in her life too. Like she's having some neck issues, her sister has breast cancer. So she's coming to a reckoning with that. And so if we start at the health level, everybody and something, I think that would be so cool is we show. And I got this idea from Benjamin's presentation where in his institute they combine design with research, empathize with that nobody wants to die, be unhealthy. And then we zoom out to the nation continent level and show that fog of war part. The fog of war part. I'm really thinking about now. I'm imagining kind of figures of people, but you can't really see them. They're under this fog and there's question marks about them. And I think actually right now at this time, even my mom, like, she's not really. So the research is presented in this super sexy, cool way that actually provokes awareness. And then people are hooked in. They read the research and I thought that was wonderful. And then that makes a lot of sense for our questioning what's going to happen to all of us. But I think it's an undercurrent even for the laggards late majority. This also reminds me from the video. He talks about the five stages of grief with AI and when he put it on there, right, he's like, what? Anger, bargaining, acceptance, hope, like all those. And I was like, oh, yeah. I see people in my circle in each of these stages of grief, of getting to hope our thing. Because our thing is very much on the abstract thinking side. Let's say we ground it very easily with AI Health coach. I think that's why it's a good use case for us. But like abstractly thinking we're in this world, we've made sense and have created models ourselves, have billions of them for this 500k population or whatever of oh, AI. But like, there's people that are totally denial, totally, like bargaining well, like, maybe it won't take these skills. And then there's like the me and you where it's like, well, replit just came out. It's taking every skill, bro. And so we were squarely in the. All right, we accept it. We got to hope about. And what we could do. I'm imagining we go to the World Health Organization or. Or, you know, the U.N. we're like, here's a sick video. Near future, but so sorry, sorry. Just to finish the thought, I think appealing to that. That shows I'm imagining in my head pictures, images of like, people, emotion of. Of confusion and I don't know what's going on. Like, the undercurrent there, I think that could get at the late majority laggard. I'll. I'll play around with that to. Do you want to go into. Actually, do you have any additions to what I just said? And then we could talk about roadmap prior and then, you know, zoomed in on one that Anatomy of a person. Maybe it's a woman, a man.

Drop that, that fog of war idea into like the, the gamification of health. Like we talk about with the don't die community, maybe two.

And then it has all these like things of like, oh, we have this data, it's version. It's in real time resting heart rate, heart rate variability, things around your cortisol.

And you know, like. Because I think what you're picturing drawing out is actually where like a heavily gamified experience can bring you into these little nuances of your being. Because I think about all the time, like, how. How can you get people to like really just phenomenologically relate to consciousness and how it impacts their life and understand one another and investing in cognitive energy, understanding. And I think, have you ever played these games levels?

Just everything about you, everything we can measure. And then we zoom out a little bit and then we see a giant population of that on a globe. And it's like here in a continent or something.

And it's like cool games where you're just, you're in a. You're in a habit loop, right? And then all of a sudden they like, you know, there's like some kind of like paywall gate and you're like, why am I about to pay $5 to like do this stupid little thing in this game that's not even.

We've lit that up, it's now lit up. We have all this data and then it shows some sort of even part of the core.

They've got me like, they've got me like three loops deep into the game, right? Where. Yeah, and, and that's where I think, you know, like when I go to the gym, like the, the exercise that I just did.

I know existential risk score that says, oh, the risk of this is this. What's the next question to answer to.

Drive that I'm producing BDNF for to like, to nourish my neurons when I.

Sleep tonight that down. And then it kind of ends with this fog of war across everywhere else that we have, right?

And that's me like four or five levels down in the game of exercise where I can. I'm just like thinking about things that I had no idea had to do with exercise when I started. And so I think that I haven't mapped this out.

And there's this promise of if we do map it all out, if we clear out the fog of war, it's.

Not that territory you described where there's a fog of War. And we want people. I think that that's also the space they're exploring.

Not just this, like, little blue continent or zone that we've lit up with data. It's everywhere.

And in a heavily gamified game of Don't Die.

Yeah, totally. Right? Yeah. Then I'm imagining the video expands to the universe, the galaxies, like the stars and shit. And it just keeps. I mean, you're totally right. Especially in America right now with the whole Make America healthy again. It's like nobody knows how. Like, everybody knows conceptually at a high level. And this goes into all of our hypotheses that are lighting up in this visual way with the same color scheme in my head to demon becoming more strong over time. It's the uncertainty. Right? Like, we're using the word confusion right now. But I was using illustrate the idea that the story of this is. We're trying to. When I shipped my newsletter, I had the synthesis of, like, uncertainty across the Personas we're tackling. And I really. It's the same thing. Like with my mom, there's uncertainty. It's just not as apparent. So it's almost education to get you to even think, oh, there's more, and I want to learn more. And then you're on a propelled loop to measure in every way possible, anything we can possibly measure, to make sense of, to go levels deeper on your own. So I think you're totally right that this could be gamified to protect itself from not dying. Like, anyway, so that. That was in my head of, like, look. And yeah, now we're participating in it to make it even more gamified with don't die, which is really fun. So it's. It's so fun. Like, we kind of manifested this. Yeah. I was. I was editing the. The pod where I kind of told that story of. Of like the pitch and stuff. Like that kind of show. My mom, that would be like, the. This is what we're doing. And then she sees it, she gets it, and she's like, oh. And just watching again. And I was like, wow. Like, we really did manifest this. Really cool. Now we get to. This is really important. This is like humanity saving level important. That's the sense that we want to give people. So thanks for sticking me with me through that. I had this whole thing in my head I wanted to get out for work on these problems we really want to want to solve and be thinking about anything else on that. With the fog of war, before wrapping up and moving into the roadmap, and what we want to. What you were saying, do you have anything to add to that? I was sort of imagining that as the 10 star education prioritized for next week in particular.

No. All good here. The general problem of like who getting people interested in this idea of consciousness subjectivity behind epistemic me, you know, I think it's an ongoing. This is really cool. It's a similar kind of like goal in mind when I made that conversation. There's like a lot of stuff there.

Yeah. And one thing I ground on is like we should be thinking about people like my mom. Like it's good to put thought cycles but it's also even more important to.

Just focus on the longevity radar, risk radar. And I just thought, yeah, like, you know, but you're just 100 true fans or whatever.

Right. Because those are the people that are going to have outside.

You're taking it to like the nth level. Right. Where you can see.

Yeah, yeah. Because you want to network effects built high trust in their networks could go champion the cause for us shows some.

Correlation between like here's the thing we can measure in the body and here's how it relates to your risk. You know, if we make them happy.

So I think it's good to think about the future which it's good to get ahead of things. Like I think playing around with that video will be good. But if we just knock out of the park with 100 true fans, I think we'll eventually get there.

Cool. I like that.

Cool. So let me share my screen. Let me know when you of of.

Cancer or heart disease or or you know, anything that's like ex.

So for other people, this is the epistemic me team hub. You're kind of seeing behind the initial.

Threat and then they kind of go like there's different they happen. There's their cognition and like say getting something like Alzheimer's disease, like it's kind of like game over. But you know, you got scenes here.

And typically this is a little bit of a mess. I think everybody's product space is a mess. Here is kind of our product space. Here's where we have like use cases. This AI Brian here is me testing out some things. Here's our ideal customer profile stuff. And then in particular here is where we're keeping a backlog for soft partnership. So epistemic ME SDK requirements to enable Brian AI.

But then you have things happening like epigenetic cell drift in your stem cells which takes. We don't really have to worry about dying from that until we're 150 years old. So, like. Yeah, you know, like, which one are you going to pay attention to first? But then if we're looking at this and we're trying to, like, really get people to phenomenologically feel that, like they're in a population of people that are all experiencing these things at once, and you're. You're going to, like, visualize.

So I don't know if you got a chance to go through this. I put in everything, everything in our backlog that we had gotten together. I think that we had gotten together about a week and a half ago. I haven't added some of these new items, so maybe we could. How about we do this? Let me link you this. And then could you just scan it as I'm scanning it and spotlight it.

Like you talked about? Yeah. Yeah. That's really powerful vision.

Yeah. On your own right now and just be like, is there anything missing? And then let's add that. I'm gonna send this to you on slack. Hold up. Yeah, I'm gonna try to make this. What? We can make this. Cool. So how we do that first we'll just scan through. Is anything missing from. So there's coach designer, there's dialectic designer now. Yeah. I'm gonna try to get AI to make this and then, you know, what's it gonna do? It's gonna give me new language in a video that I will increase my intelligence on and then make more sense of it and then. And then make the AI make bullshit. Actually, could you get me that Excalidraw or slash? You should have edit access. So how about we add the excalidraw to this? I'll make a little.

Cool. Let me do that.

It smears all the way down. Jonathan Smears all the way down. It's a quick time. Check.

This one to be H2. So do you want the roadmap on one and then the dialectic design?

We've been going for a while on this. I actually totally think we could have Dean watch that video. He would love it. And then we do another one just on this and go a little bit deeper with more prepared questions next time. This is thoroughly enjoyable to talk about. And also we just came up with a deliverable on my plate, which is good or bad, but, like, I'd say it's good. Overall, it's more work, but I definitely see something like that would bridge the education gap with the laggard late majority Persona for us. Like, it's. It's harder to explain to somebody like my mom. But if she watched something and it was told in this very narrative way that provoked human emotional connection, then I think it would draw her into the story more. Like, everybody can relate to health on.

Another.

Let'S put everything under this guy. This header epistemic me partnership roadmap for now, just so we have it in one place, so I could see between the excalidraw that we demoed to them along with the Belief Mirror app. And I want to just make sure that it is tied to our backup.

Does it look good?

It is loading on my end. Yeah. Cool. Perfect. Cool. So dialectic designer this experimentic loop, experimental loop, perspective solves, I guess. AI coach.

I'm just trying to picture the best demo and getting from where I'm at now to something I think I can demo about a week from now.

Right. Like, she's very interested in health at this point in her life too. Like she's having some neck issues. Her sister has breast cancer. So she's coming to a reckoning with that. And so if we start at the health level, everybody can empathize with that. Nobody wants to die, be unhealthy. And then we zoom out to the nation continent level and show that fog of war part. The fog of war part I'm really thinking about. Yeah.

And I could start on like, I could start just building the UI for this. It's like a next JS application. Or I could focus more on the trainer that I'm working on right now and show the outputs that it gives so that we can have a conversation about like how the outputs can be used in a coaching application, like for question and answer.

So let's talk about that. So how much effort would it be for focusing on the UI for a V1 versus like getting this up and then maybe it doesn't even work. Like we just mock or stub out data. How much effort do you think that would be like, if we like now I'm imagining kind of like figures of people, but you can't really see them. They're under this fog and there's question marks about them. And I think actually right now at this time, even my mom, like, she's not really questioning what's going to happen to all of us. But I think it's an undercurrent even for the laggards late majority. This also reminds me from the video. He talks about the five stages of grief with AI and when he put it on there, right, ranked it on our little backlog let's just type that in actually.

Yeah. The tools we're using now, taking that mockup flushing out a little bit put into V0 for a NextJS scaffold. You know, I could have something really nice in like four hours and. But the thing is that it would.

Be demonstrate this is like what anger, bargaining, acceptance, hope, like all those. And I was like, oh yeah. I see people in my circle in each of these stages of grief of getting to hope of oh AI. But like there's people that are totally denial, totally like bargaining. Well, like maybe it won't take these skills ui.

But it doesn't quite answer the question of like, how is my AI coach going to work and how's it going to like respond and interact with users? And that's the thing. I'm thinking more like the critical path.

Yeah. If we level up, like the biggest hypothesis that we have a lot of confidence on is that we want to resolve uncertainty. If they release us into wild, they. They need to know. And then there's like the me and you where it's like, well, replit just came out. It's taking every skill, bro. That it will be a hundred percent Brian. That every single interaction they have control over and there's certainty in that it will personalize value and it will not undermine their brand credibility. So I'd say like just from that product lens, we should index towards any feature that really just maps to alleviating uncertainty is probably the highest value. And so we were squarely in the. All right, we accept it. We gotta hope about a better future. But so sorry, sorry. Just to finish the thought, I think appealing to that emotion of confusion and I don't know what's going on, like the undercurrent there, I think that could get at the late majority laggard. I'll play around with that. Do you want to go into feature right now? If we think about what Greg asked us and we have the otter but the singular thing. And I'm thinking about Greg a lot right now because he's got to implement this guy. Right. Everybody else is like, you know, there are other decision makers. But if he sees that it's easy, he could produce value very quickly, which we want to get to. So Greg asked how do we make sure it's not hallucinating? So that totally told me that's where his thoughts. Who actually. Do you have any additions to what I just said? And then we could talk about roadmap priorities where his mind is at and we should meet him where he's at. So give him a certainty. Let me just write this out and then we'll map it in the RICE framework so there's two that we're going to add. But as I do this, does that make sense? Kind of where my head's at with overall prioritization. But we can run it through the.

Framework with that fog of war idea into like the gamification of health like we talk about with the don't die community. And you know, like. Because I think what you're picturing drawing out is actually where like a heavily. Yeah, I think I'm just, I'm wanting to picture that the specifics of what it means to reduce that risk. Gamified experience can bring you into these little nuances of your being. Because I think about all the time like how.

Yeah, take mock up of dialect. Okay. As a developer, I can web application with a nice UI to easily define archetypes.

How can you get people to like really just phenomenologically relate to consciousness and how it impacts their life and understanding one another and investing in cognitive energy understanding. And I think, have you ever played. Played these games where you're just, you're in a. You're in a habit loop, right. And then all of a sudden they like, you know, there's like some kind of like paywall gate and you're like, why am I about to pay $5 to like do this stupid little thing in this game that's not even my.

User base perspective selves to use in the dialectic and a learning objective to ultimately feel confident about a given dialectic I've simulated so that I can then release it into the wild with my.

As an experiment.

Okay.

I've been part of the court. They've got me like, they've got me like three loops deep into the game, right? Where.

Yeah.

Because you get to collect more data and feed it back in to the trainer.

Yeah, yeah. To real users and form a great feedback loop to my.

And. And that's where I think, you know, like when I go to the gym, like the exercise that I just did, I know that I'm producing BDNF for to like to nourish my neurons when I sleep tonight. Right. And that's me like four or five levels down in the game of exercise where I can type meaning data.

Okay, that's decent. And then we basically said this is about four hours. So I have five dollar signs here and I actually haven't normalized this. So I want to talk to you about that. And Then we can just write a rule for that. Anytime we prioritize, we think about that. What's your perspective on like, estimates since we're at a by?

I'm just like thinking about things that I had no idea had to do with exercise when I started. And so I think that the territory you described, where there's a fog of war and we want people, I think that's also the space they're exploring.

A bi weekly sprint cycle. I think we should maybe say, like, I don't know, five is one sprint or maybe three is one sprint. And that could be the benchmark. Yeah, I mean, maybe three and then five.

A lot of the stuff I've delivered historically has taken me more in a heavily gamified game of don't die.

Yeah, totally right. Yeah, I mean, you're totally right. Especially in America right now with the whole make America healthy again.

It's like nobody in one sprint cycle just because, you know.

Oh, okay. So then maybe what would be easy here?

So two.

I mean, we could just like take two. Two weeks divided by five. We could put something else here. That isn't a dollar sign either, to get more granular. He knows how like everybody knows conceptually at a high level. And this goes into all of our hypotheses that are.

Yeah, it's. It's fine.

Becoming more strong over time. It's the uncertainty. Right. Like, we're using the word confusion right now. But I was using. When I shipped my newsletter, I had the synthesis of like uncertainty across the Personas we're tackling. And I really. It's the same thing. Like with my mom, there's uncertainty. It's just not as apparent. So it's almost education to get you to.

Even if, like. How about.

Okay, I got one. Okay. Let's say, like, because two weeks, let's say, you know, in a normal working cycle, that's like 10 days or whatever a business days. So this is nice. Like one of these is like two days. Okay. I mean, more accurately would be at the hourly level, but we're bootstrapped. So, like our hours kind of change. So I don't want to get too anal and think, oh, there's more and I want to learn more. And then you're on a propelled loop.

Dude. I've literally spent like two hours in meetings just having this conversation about how effort is ranked with sprinting.

Go levels deeper on your own.

Yeah.

So I think you're totally right that this could be gamified. And yeah, now we're participating in it to make it even more gamified with don't die which is really fun. So it's so fun. Like we kind of manifested this. Yeah, I was editing. What'd you all come to?

Well, do you ever use pivotal tracker? I think the exponential effort is like more representative of how things scale with complexity. But how about we just say five dollar signs is more than one sprint and then like less than four dollars is linear within the sprint. So four dollar signs is like one sprint cycle.

Effort. Five dollar signs. Editing the pod where I kind of told that story of the pitch and stuff like that and just watching again and I was like wow, we really did manifest this. Really cool. Now we get to. Yeah. Signs work on these problems we really want to solve and be thinking about anything else on that. With the fog of war before wrapping up and moving into the roadmap and what we want to. Oh okay. Is more than once rent lesson four is linear. Cool. Okay, great. So a dialectic designer sounded like Maybe that was $1 sign or $2 signs then just to get the interactive UI up in a next JS with some stuff.

Yeah, I think so. It's the trainer right now.

Prioritize for next week in particular.

No, all good here. The general problem of like who getting people interested in this idea of consciousness subjectivity behind epistemic me. You know I think it's an ongoing. Once that's built the UI for it.

Is that the training harness.

The training harness, the dialectic training harness conversation. There's like a lot of stuff there.

Yeah. And. And one thing I ground on is like we should be thinking about people like my mom. Like it's good to put thought cycles but it's also even more important to just focus on the first Will the user story for this be. Would it be like as a developer I could leverage an out of the box training harness to speed up my simulations.

As another I can. How. How about. Because the output of it is a set of like a question answer tree. Yeah.

Okay.

That might not be the best word for it but 100 true fans or whatever.

Right. Because those are the people that are going to have outsized network effects built high trust in their networks could go champion the cause for us.

But like. Like something concrete like that is that you know you can. You can put example users in and.

You get if we make them happy. So I think it's good to think about the future which it's good to get ahead of things. Like I think playing around with that video will be good. But if we just knock out of the park with 102 fans, I think.

Out, like list of questions to ask them that you can. You can feed into your AI coach.

Or question answer trees on a particular. Is it accurate to say on a particular learning objective? Specified. Is there a specified learning objective? So that we'll eventually get there.

Cool. I like that.

Cool. So let me share my screen. Let me know when you could see it. So for other people, this is the epistemic ME team hub. You're kind of seeing behind the scenes. I can save time and be confident in the dialectics that I experiment with and promote to production scenes here. And typically, this is a little bit of a mess. I think everybody's product space is a mess. Here is kind of our product space. Here's where we have like, use cases. This AI Brian, to real users. Yeah, this sounds good.

Yeah.

How does it sound? Okay, cool. Acceptance criteria. You want to run through that real fast? Do we have that somewhere? We might have that in Slack.

Yeah, I think just that, you know, you can input a learning objective, use reinforcement learning.

And here is me testing out some things. Here's our ideal customer profile stuff to.

Reward the most valuable questions that progress towards.

And then in particular, here is where we're keeping a backlog for soft partnership. So epistemic ME SDK requirements to enable.

Brian AI learning objective. And then.

So I don't know if you got a chance to go through this. I put in everything. Everything.

Curate the rewards manually.

Output of optimized. Optimized question and answer trees for the specified. Great. Okay, I have another MVP column. This is clearly mvp. Let me just put that. And MVP to me is like, Greg has used it and he has 100 private beta users in our backlog that we had gotten together. I think that we had gotten together about a week and a half ago. I haven't added some of these new items, so maybe we could. How about we do this? Let me link you this. And then could you just scan it as I'm scanning it on your own right now, and just be like, is there anything missing? And then let's add that on it. Like, that's like pretty. That's my working mental model of mvp.

Yeah. So this will be an example app. And I don't know if it's just like. It'll be a Python implementation. And so, yeah, like, people could download it and use it, or they could just like, use it as an example and build their own thing, you know?

Yeah.

But I think it should be, like, baked enough. You know that you could do it and.

Yeah. Yeah. I think this is a must have a performance benefit. I'm gonna send this to you on Slack. Hold up. Cool. So how we do that first we'll just scan through is anything missing from. So there's coach Designer, there's dialectic designer. Actually could you get me that Excalidraw or slash. You should have edit access. So how about we add the ext. Okay. So I've been. Reach is like how much reach do we get if we implement this AKA how many users would use it if we implement it for our use case that that's how I'm thinking about Reach. So of our hypothesized ICP market do we believe what with what confidence this would reach everybody versus nobody and nobody or very like one user niche? I'm just putting 1%.

Yeah. I think it's interesting anybody building a coach, the A health coach or scaladraw to this.

I'll make a little.

Cool, let me do that.

Change this one to be H2.

See do you want the road map on one and then not the dialectic design. Any. Any AI coach.

And I think this would be high impact on another. Pretty confident about it. And how much more effort to take this through the acceptance criteria.

Let's make it a three. So like a little bit more than half a sprint.

Okay.

If I get lucky it'll be less but you know.

Yeah.

Okay. So here's one of the key questions on my mind is like what. What's most empowering to the AI coach developer from the. From the output of that training? Is it, you know, like for instance if you're building a knit coach, one way you could have just like constrained questions.

Let's put everything under this guy. This header epistemic me partnership roadmap for now just so we have it in one place. So I could see between the Excalidraw that we demoed to them along with the Belief Mirror app and like a.

Multiple choice questions for instance that's like the most constrained thing is like you.

Know, here's a. I want to just make sure that it is tied to our backlog.

Does that look good? It is question. Here's some multiple choice answers. And then like unconstrained is like here's. Here's some questions that will lightly like fine tune an LLM and then it can just ask whatever question is it wants based on those inputs. So I guess is it most useful to have. Here's a Monte Carlo question answer tree that is heavily Constrained or is it more like here's just like a fine.

Tuning technique loading on my end. Yeah, cool. Perfect. Cool. So dialectic designer this experiment that you.

Can use to then test out like a very broadly answering LLM agent.

And then part of this just for me to envision the user experience. So Greg sees this training hard loop, experimental loop, perspective solves.

So I guess AI coach the and I'm just trying to picture the best.

Demo and he makes a copy of it on local and now there's this in front of him. He could either just give the learning objective and then it is this broad LLM agent model where it makes sense of that and then it creates the experiment, the dialectics, the questions, answers.

Getting from where I'm at now to like something I think I can demo about a week from now.

Yeah.

And I could start on like I.

Could start just building simulates it and then provides. Here's a bunch of things we tested. Cool. Greg, take a look, let me know what you want to do versus the other one would be Greg goes in and he's already thinking, well there's certain.

In the UI for this, it's like a next JS application, questions we want.

To ask or a certain type of way we want to ask things and I want to input that, then I want to run.

Or I could focus more on the trainer that I'm working on right now and show the outputs that it gives.

So that the experiment is that are those the two paths in front of us?

I think of it, it's like more like what is the actual AI coaching experience to be like? You know, is it just something that you can, you can ask Brian anything. In the more constrained example I'm thinking like he's going to, he's going to identify that a user needs to answer. We can have a conversation about like how the outputs can be used in a coaching application. Like certain set of questions and like they, they have to like sort of provide answers for all those before they can get to the next set of questions or something like that. And then I think like there's a lot of precedent like the product you're talking about. Rasa last time for question answer.

Yeah, so let's talk about that. So how much effort would it be for focusing on the UI for a V1 versus like like getting this up and then maybe it doesn't even work like.

Yeah, and I used one called Landbot in the past and so on Landbot like I would just bring together like multiple choice questions like one after Another and you know like that was fine to collect the information from the user. So you know, just what's most useful right now is one thing I'm trying to. I'm trying to hit like how can we make this for people that really know how to build and just mock.

Or stub out data? How much effort do you think that would be? Like if we ranked it on our little backlog? I mean let's just type that in actually.

Yeah, like the, the tools we're using now, um, you know like taking that mock up, flushing out a little bit, putting into V0 for. For a next JS scaffold. You know, I could have something really nice in like four hours and but the thing is that it would be demonstrate this ui but it doesn't quite answer the question of like how is my AI coach going to work and how's it going to like respond and interact with users. And that's the thing. I'm thinking more like the critical path.

Yeah if, if we level up like the biggest hypothesis that we have a.

Lot of confidence on AI coach, you know, how can we make it useful? Maybe provide a set of evaluations across a question as tree.

I think at the heart of this are two questions from my perspective. One is do people building AI agents or chatbots today with the existing models or workflows of Here's a very constrained experience.

I think our institutions should be able to ontologize existential risk in the same way we think about longevity but also at a societal level and make sure that the right questions being like what.

Is the is that we want to resolve uncertainty. If they release this into the wild they need to know that it will be 100% Brian that every single interaction they have control over and there's certainty in that it will personalize value and.

It will not knowledge we need to gain in order to like protect our society from existential threat in order to like continue to grow and flourish through evolutionary cycles and be able to sort of like not just see what's on.

The horizon but undermine their brand credibility. So I'd say like just from that product lens we should index towards any feature that really just maps to alleviating uncertainty is probably the highest value feature right now. If we think about what Greg asked us and we have the otter you have to map together the multiple choice questions and stuff like that. Do they like that experience? Like did you like that experience?

It was like better than a form. So we had like one experience. It was a type form and then one experience that was a land bot. And people, just some people would prefer one over the other. And so he just had both available.

Okay, so, so then the other question is Greg's immediate experience would inform that. But the singular thing, and I'm thinking about Greg a lot right now because he's got to implement this guy, right. Everybody else is like, you know, there are other decision makers, but if we. He sees that it's easy, he could produce value, very see what.

What change looks like. And like, you know, what kind of questions should we ask today so that you know, and then put financing into that. And I think like one of the ways that you do that, like, how do you understand, like, what matters within a population, for instance? I think that's where it comes back to representation systems. And if you have representation systems and values that you know are well understood within those representation systems, question like, how.

From a development experience do I like this to implement it to be useful quickly? Which we want to get to. So Greg asks how do we make sure it's not hallucinating? So that totally told me that's where his mind is at and we should meet him where he's at. So give him a certainty. Let me just write this out. And then, and then there's the bigger question that I think you're asking, which is which path technically gets us the best value for the end user. Keeping in mind, Brian would. Would kind of map it in the RICE framework. So there's two that we're going to add. But as I do this, does that make sense? Kind of where my head's at with overall prioritization. But we could run it through the framework with that.

And you know, we, you think about how like, how dumb it is right now or like, I mean, we can't take for granted democracy, but like just at the pace that it moves relative to existential threat, like that we go to the ballot at a certain, like once a year, once every four years, cast a vote on things that everybody knows we're not asking the right question when we cast votes, like everybody knows it.

Well, I guess I shouldn't say everyone. It's not 100%. There are brainwashed people deep in the matrix that think they know.

And yeah, I think I just, I'm wanting to picture that the specifics of what it means to reduce that risk.

Yeah, take mock up of dialect. Be like his thoughts on this would matter a lot too. Like one use case I'm using to stretch this is what if somebody said, hey, Brian, I want to commit Suicide. What do I do? Like what does the agent say? What do we allow them to say? And that's a question like I think the don't die team would need to help us understand which I think would drive a little bit of the philosophy.

Here in my mind this new paradigm of ChatGPT.

Okay. As a developer I can web application with a nice UI to easily define but they think they're right, you know.

But in general, yes. Everybody that I take theirs that you talk to probably has like got this archetypes. I'm feeling that my vote isn't voting even for the right thing. It doesn't matter. Right?

No, it's a sham. Yeah it's a sham then but then.

If you took it if imagine that like all of us are asking the right question all the time.

My user base perspective selves to use in the dialectic time all of our.

Needs are represented within like you know all the homeostatic processes in of our body in our body is just like it like Brian's probably looking for something like that but it's just like him. Um and then it comes with all the rails of chat GPT also and like you know that kind of thing.

Yeah. So then I guess because that's where my head's going is you want to and a learning objective to ultimately feel confident to be a free form experience. If we index on I want to gather as much belief data from you as possible about a given dialectic.

I've.

Simulated so that I can then release it into the wild with my as an experiment. Okay. Right. And so like I was thinking I was kind of imagining the chat GPT experience but with like the belief mirror.

Our supply chains I'll have representation and all of this is being fed back into epistemic sort of integrated and then.

You know obviously health focus and it's Brian. So then I guess that begs the question how does ChatGPT do it its rules. Is it like a pretty basic rules engine where it's just data labeling.

Because you get to collect more data and.

Feed it back tied to a rules engine that they maintain centrally.

Well I think in our case we're. We're kind of institution in real time basis that can make sense of it and then like prioritize what are the right questions to ask. That's what like a next generation epistemic institution looks like to me. And again to the trainer.

Yeah yeah I think to real users and form a great feedback loop to my.

You know so this is a collective intelligence and I got so interested in like what examples of collective intelligence look like. I studied the transition meaning data.

Okay, that's decent.

Like at the GPT layer we're like, you know. Yeah, like we'd fine tune it by creating a, a GPT like as a set of questions that. Here's what AI Brian Coach would ask. Like please use these questions first when trying to understand the user and then extrapolate to ask the next.

And then we basically said this is about four hours. So I have five dollar signs here.

And I ask question if it's not available.

Okay, so let's take that suicide use case. Let's say Brian's team is like we.

Want to get from unicellular animals into multicellular through the, the basal ganglia specifically. And if you look at how the basal ganglia works in our body, it would. It started out and it pretty much still is a representation system for a.

Very typical kind of response for that use case. And the reason I this is top of mind is I actually haven't normalized this. So I want to talk to you about that and then we can just write a rule for that. And anytime we prioritize we think about that. What's your perspective on like estimates?

Uh, since we're by all the homeostatic needs, you know, of, of our cells and our tissues and our organs. Right.

Um, and it's actually there was a community member that posted that they're thinking about suicide and then somebody else pinged me and was like a bi weekly sprint cycle. I think we should maybe say like I don't know, five is.

And so it's. I think that there, there's evolutionary precedents for like what an epistemic institution could look like. That scale.

Hey, what should we do about this? And I was like oh shit, I should screenshot this. Send it to Greg. And now he's kind of thinking about like how do we handle this use case. I suggested good stopgap is have something that detects that and then one sprint or maybe three is one sprint. And that could be the benchmark. Yeah, like just from that. Because we always take this hypothesis that we're all just agents. We're made up of agents and yeah, I mean maybe three and then five.

A lot of the stuff I've delivered historically has taken me more than one sprint cycle just because, you know.

Oh, okay. So then make collections of cells. Right. So just from that example you're saying there's obviously like kind of parallels maybe what would be easy here? So to an auto post Suicide hotline resources nationwide resources like for that problem, for that Persona to meet them where they're at. But also it would be good to vet with you know, a bigger organization or collective level. Right? Am I getting that right?

And I think a sophisticated representation system sort of starts to look like a super organism in a lot of ways where it has its own agency and.

I mean we could just like take two two weeks divided by five independent organization that is focused on this problem. We're sort of brainstorming here. But that basically tells me okay, if we're living at the GPT layer for that use case which stretches us a lot like this is a very extreme niche use case. But we could put something else here that isn't a dollar sign either to.

Get more granular it's own like sort of sense making apparatuses. It can ontologize, understand existential risk and then like yeah, it's, it's fine.

Find it useful as a way to kind of min max or stress test are are thinking to drive direction of this is. I would imagine that Brian wouldn't if.

Like how about.

Okay, I got one. Okay. Probably want some level of control over that. And so his developer would probably want let's say like because two weeks, let's say, you know, in a normal working cycle that's like 10 days or whatever some way to easily put in hey, if it's topics of this type we have a canned resistance days. So this is nice like one of these is like two days response and they want to know that that gets to 100 so. So backing up from there. Let's assume that is the case. I mean more accurately would be at the hourly level. But we're bootstrapped. So like our hours kind of change. So I don't want to get too anal. That seems reasonable to me based on the interaction I had with Greg about that particular use case.

I've literally spent like two hours and meetings just having this conversation about how effort is ranked with whispering in the wild.

Not them asking a coach, but them just posing it to the community. Then does any director understand what's the.

Next best question to ask and what resources should I actuate in my environment to do that? And when we answer those questions and we say oh, the next best thing for me to do is go, you know like eat some turkey in the fridge. Right? We could actuate that and like go grab the turkey and chew it up and digest it and feed it to our bacteria in our stomach. And when epistemic institution is asking that Question. It can, it can allocate financing and it. It can. It can create training programs and it can. You know like it's the same sort of actual, you know.

What'D you all come to?

Well, do you ever use pivotal tracker to solve the next best question? It just has, you know, it understands what is universally direction.

We're going here with user constrained versus broad ELM agent. Does any direction constrain that like requirement?

No. I'm thinking about, you know how the CCP did this with the deep seeker. I think the exponential effort is. Is like more representative of how things scale with service.

Yeah.

Like they have. They have a parallel LLM like monitoring all the results in order to make it happen.

It's a different implementation but essentially it's the same thing.

I need to make complexity. But how about we just say five dollar signs is more than one sprint. You ask about tianm and square and like the secondary agent comes in is like so you could have like a secondary die agent and then like less than four dollar signs is linear within the sprint. So four dollar signs is like one sprint sign. And if like if. If like something is higher on the dye scale it could interact intervene sense.

Of this to make a decision. I need to ask the next best question to make sense of it to make a decision to actually.

And be like hey, we're hearing a lot of DAI from you right now.

Either from a collect like us. You know that organism, the ganglia. What was it called? Ganglia. Yeah.

It's the oldest part of our brain. Yeah.

Yes. Versus like kind of like that direction. Honestly that as we noodle through it. That seems the best to me. Effort. Five dollar signs. Yeah Signs. I had a gut level because that. That allows for control but it's also flexible.

Yeah.

Oh, okay. Is more than once rent. Okay. Just take a different collective epistemic institution. It's just different deploy capital, you know, so. But and then from a user experience perspective the developer wouldn't have to make lesson four is linear instruments maintain all these like rules. And like a typical rules engine it would just be an LLM with kind of makes sense actuate versus like yo, I'm in this body getting you know, sensors of data inside this body. Okay. Activate. You know, it's a little different but like essentially overall it is the process. Let. Let's do a little exercise. I think this is cool. Okay, great. So a dialectic designer sounded like Maybe that was $1 sign tune prompt set and some content data.

Right. Like and honestly I or $2 signs.

Then just to get the interactive UI up in a next JS with I.

Don'T know how additive we can be to that. Unless it's really like. Unless we can articulate it as a philosophical.

Some studs.

Yeah, I think so. It's the trainer right now divide the philosophy described is one of don't die and then we can actually auto identify things that are anti the philosophy on.

A scale to dream up. What does the next generation epistemic institution in the future look like?

And once that's built, UI for it is that.

So now we have like some, some anchored thoughts around how this guy did it. And then you just gave us a little nugget which is like envisioning a world where real time we're getting data tracked at every cell, organism, agent, person, human on the planet and then we could feed.

But I'm just more thinking about like the utility that we can provide. I'd rather think about these edge cases later because I know we can just overlay a layer, a safety layer on top of whatever we do. Yeah, guardrails.

So I think actually what's going to help us the most is if we just get this down in training harness.

The training harness, the dialectic training harness.

Do a slack message and I just shoot off to Greg and then he just tells us because what would the user story for this be? Would it be like as a developer I could leverage an out of the box training harness to speed up my simulations that some way in real time to some data set back to an epistemic institution where they could review that.

As a villain. I can.

How, how about hopefully have like context of that in different almost verticals like the different sciences departments. I guess if you can think about it from an academic institution perspective departments. Because he's been very open in dialogue with us. So I think he's really open to just literally just giving us the priority at this point in the relationship building. So how about we do that with this particular one. Give you direction I guess.

Can we propose like this is what we were thinking, you know we'll be working on this and you know, let us know if you have any thoughts because I, I think it's something because of the state.

So there's obviously AI in that. There's obviously. I'm thinking there's a lot of requirements.

There because the output of it is a set of like a question answer tree. Yeah.

Okay.

That might not be the best word for it but like, like something concrete like that is that you know you can, you can put Example users in and you get.

But like without getting into the, you know, implementation too much, what other thing.

Out like list of questions to ask them that you can, you can feed into your AI.

Could we do to like really shorten that feedback loop slash make it even more accurate.

The world today it's like sort of obvious like okay, ChatGPT found product market fit now. Now Claude is copying it. Now Deepseek is copying it and that's the form of AI coach that everybody wants. Like that's like Greg shouldn't need to tell us that. We should know that. I guess so then when it. I think what we're describing is here's how we would we're thinking make the.

Sense making of it faster and more accurate.

Yeah. You know, should we come up with an example of like what this coach.

Or question answer trees on a particular.

Is it accurate to say institutions tasked with, you know, is it.

Yeah, you know, let's use the longevity case on a particular learning objective. Specified user specified learning objective. So that. Because we're really working on it and I see a path there. Right. So starting with Brian AI about enabling.

You to program a chat GPT like AI Brian coach. Let us know if this sounds about right. And that's where I would describe like, you know, here's one option where we could provide a prompt of the. The question exploration space. Here's another option. We can provide the embeddings in evaluations for the question answer evaluation space.

Let's say he gets a million members by end of year using IT and collecting health. I can save time and be confident in the dialectics that I experiment with and promote to production to real users. I think it sounds good.

Yeah.

How does it sound when you say embeddings? Explain. Could you explain that a little bit more? What does that actually look like? Well, just wearable data and non significant population of that populate sub population of that.

It's like, you know, if we're, if we're going to encode a question answer pair as in like a vector database, then it's like specifically encoded with the embeddings of the LLM or okay, cool acceptance criteria.

You want to do that contributing biomarker data, other demographic data like gender, where you're from and then using epistemic me we contextualize a lot of belief system data too against where it's going to.

Be trained on and then you can sort of like. So you can have that question answer pair as an embedding.

So let's use that as the use case and then the epistemic institution. Do you want to pick what their.

Goal might be with an evaluation score and you can like, you know, use that as a, as a priority queue for how the bot would ask like, you know, query and ask questions and.

Fast. Do we have that somewhere? We might have that in Slack.

Yeah, I think just that, you know, in a lot of this I'm just, I'm just asking for time to like, let's, let's think this through and I want, you know. Yeah, I think that's a great example. I think maybe just say protecting a certain. You can input a learning objective, use reinforcement learning to reward the most valuable questions that progress towards the population of humans from existential threat through a health lens. You know, I think it's a good thought experiment the learning objective and then bounce some ideas around because just. Yeah, you know.

Yeah. All right, so okay, so now we got a little bit of something else. How you think about it?

Yeah, I think.

Okay, looking at this, does this change the scope any meaningful way going either way or is it just more of a. I just need to know which one would be the most valuable. Doesn't change scope.

Just like let's say we have, let's call it like 50,000 humans. And the specific output that I have, I'll tell you my tendency right now is to go and learn the machine learning best practices for building question answer bots and then follow those heuristics. And I think that's going to guide me towards formal question answer evaluations. And there's probably some things I need to learn that I don't know the rewards manually.

Output of optimize.

Oh yeah. So that's like the feeling that I have. And if, if you were to say like no, like let's just generate a prompt with like some, some questions and like see how it performs. Then we have an AI that is meant to like aggregate everybody's needs and sort of like.

Optimized question in the.

It's the cumulation of the people and you know, all the biology that's within that organization plus the AI that. But the AI will automatically like. I don't think it fully addresses like the concern of will it hallucinate or not. But it's something and it's quicker.

Yeah, I don't answer trees for the specified. Oh, I guess another question I have to make a decision is if we go that path and next meeting they're like I want to do this user provided embeddings. I want the. You Know, more control. I want this to be in a.

More, you know, weight everybody's input and then come up with like the next best question to ask and like determine what's next.

Typical machine learning chatbot workflow. Then how would that. Would we have technical debt accumulation is I guess, is what I'm saying. Like, would it be a lot to change direction at that point? If we demo, we go with the broad, here's a prompt and then next meeting they're like, actually, I want to see this develop.

No, I think it's more maybe a question of first touch. Like, what do we bring to the table? Action that should be taken. So I think kind of like in the political structure in that sense.

Great. Okay, I have another MVP column. This is clearly mvp. Let me just put that. And MVP to me is like, Greg has used it and he has 100 private beta users.

And then I think about like, well, okay, what data do we put in? You mentioned biomarkers, right? I think they might be looking for us to say like, this is the best way to build it. And they may not know to like. I think we're really like taking a self model of, of every human and breaking it down to like all the homeostatic functions that we need. Like we talked about hearts and lungs and blood and muscles and you know, on it.

Like that's like pretty. That's my working mental model of mvp.

Yeah. So this will be an example app. Ask for additional things if we, we provide like a prompt technique that doesn't really work that well.

Like I'm thinking Brian being a philosophically minded dude, like he sees himself as a philosopher. When I said dialectic to him, he, his, he, he, he like understood.

And I don't know if it's just like it'll be a Python implementation.

He talked about the don't die dinners as a series of dialectics. So I think at some point he would probably like want to see this. And I also think that he would have a mind for, at sort of a higher level.

And then below that I think, you know, you have all kinds of cellular interactions. And so, yeah, like people could download it and use it, or they could just like use it as an example.

And build their own optimizing dialectics. So one question I have is, and.

You can imagine that like each individual person could be represented by say like 100,000 self agents that are all communicating data about their homeostatic needs to survive. You know, and so you, you want, you know, if you have, if you have cancer cells in your body. They, they are part of this represent thing, you know. Yeah, but I think it should be like baked enough, you know that you.

Could do it and yeah, yeah, I think this is a must have and performance benefit. Could we go the fast route for now and just like seed it based on our extracted belief systems from his content? Like give the this LLM that perspective and just feed that and just say hey, is V1 this is what we did. What do you think? Because it kind of fits into the model of okay, Brian's going to be making new content all the time.

That system that is saying like, you know, hey, I'm a threat or these cancer cells are a threat to me or something like that.

Right.

And so then you, you're going to bubble all those things up to a set where like what the AI is going to be looking at is kind of like these inter subjective representations.

Okay. So I've been Reach is like how much reach do we get if we implement this AKA how many users would use it? If we implement this way is less bottlenecked, this way is more bottlenecked. This is my gut feel at a high level. Implement it for our use case that that's how I'm thinking about reach. So of our hypothesis of all the.

Agent beans within its representation network and remember if you have 100,000 per one person, there's 50,000.

And I think it brings us a little bit closer to just evolving towards the overall vision. I'm kind of a fan of us riffing through this of just ICP market. Do we believe with what confidence this would reach that take perspective self of Brian, generate these. That's the demo. And then if we can have time get it working in dialectic designer. If it will take four hours, everybody versus nobody and nobody or very like one user niche. I'm just putting 1%.

Yeah, I think it's interesting anybody building a coach, an AIA health coach or people, you have millions and millions like billions of representatives that you need to account for. And so then you're going to take you know like any, any AI coach.

And I think this would be high impact simulated realities.

And you're going to have to understand the environment in which each of these agent independent agent actors works in.

I guess it's four hours. You mentioned if it's not like actually hooked up to the data, it's just stubbed out. But yeah, that would be my order is try to get this done. Go with the perspective self Brian.

And if you make changes to the environment, what is going to be the outcome of that. So this AI is going to need.

This route dialectic designer next. Whatever you can get done of that and then hook it up after that if you can.

Okay. Yeah it's really helpful. That makes sense.

Pretty confident about it. And how much more effort to take this through the acceptance criteria.

Let's make it a 3. Run sophisticated simulations across this population set and then it will understand like so like a little bit more than half a sprint.

Okay.

If I get. Because I would. So I would just like I would focus on like populating do we need to take immediate action such as like cleaning the air that these people live in or do we need to take longer term actions such as researching interventions that, that are pharmaceutical. Um and I think Brian's like the, the self model for Brian with his belief system based on documents and then use that belief system as con Lucky it'll be less but you know.

Yeah.

Okay. So here's one of the key questions on my mind that the common thread with all of them is sort of metabolic like if you can, if you can understand that holistics to ask the next best question. So what this trainer is going to be doing is going to be saying what's most empowering to Brian Johnson wants to learn about this user and it's going to be simulating that interaction and it's going to be promoting the next best question that Brian Johnson would ask in a given conversation.

I think that will get their light bulbs turned on for sure. And then from there you know we can go into the more we can.

Keep healthy metabolism amongst this group of like 5 billion, you know, representative or.

Technical specifics and provide them our thinking.

On the decision in my, my agent system. And then I can weight which metabolism sort of like is most important for survivability of the entire entirety. Then I think you've got like a, an epistemic institution like what we just.

Talked about and then ask them like here's the rest of the stuff we got to build for you. How would you.

That's working on top of a representation system that is holistically sort of like a consciously agent actor.

Are you good with this for mvp? And then we improve this part over time to give you a bit more.

Control for the, the AI coach developer from the, from the output of that training. Is it. You know, like for instance if you're building a knit coach one way you could have just constrained questions like multiple choice questions for instance. That's the most constrained thing is here's.

A provided Embeddings or user specified embeddings. As we work on these other items in the roadmap, I'm thinking the next meeting we show this show progress and then do another facility question.

Here's some multiple choice answers and then unconstrained is like here's some questions brainstorm.

Of where I really like this, this product exercise where I give them like a hundred bucks and it's like you have constrained capital. Pick these features and then that just.

Gets it to us works towards group survivability.

Okay. So just to piece together what I'm understanding to solve that problem we would need to allow the epistemic institution to be able to ask the right questions.

And get that will lightly like fine tune an LLM and then it can just ask whatever question as it wants, you know, based on those inputs. So like I guess my. Yeah, is it, is it like really.

Good loop where we're just like we know exactly what you want next. So that was my plan for this next meeting with them. The right answer is 100% of the time really fast in real time. So that means that there's a requirement.

Of is it most useful to have like here's like a Monte Carlo question.

Answer tree allowing them to run simple to highly complicated simulations on target. We show a bit of value. We show them what's on the backlog based on the first roadmap conversation we.

Had them that is heavily constrained or is it more like here's just like a fine tuning technique.

They'll have that context in their information architecture and mental model. We give them some play money kind of like we did last time when it's just the canva slides. Get them working in there on a shared artifact and actually just have them rank it for us. Cool.

Yeah. Okay, so for and for the demo in the meeting that you can use to then test out like a very broadly answering LLM agent.

And then part of this just for me to envision the user experience. So Greg sees this training hard population and have sufficient data for that to.

Have pretty would like. So I'll, I'll like run the trainer with Brian in the loop and like.

Show pretty much 100% efficacy of representing that target population in the real world. Did I get that right? Like that's the, that's the overall thing that's like required. Like that's the foundation that's required.

Right.

I mean he makes a copy of.

It on local and the question answer output in a way that like can all comprehend and then it would Be. I don't know if you want to. This would be a good use of time. But I think like if we acted like the user and then we asked Brian a question and saw response come back and like just did that together. Yeah, I think that would be like best. The best demo.

Okay. So yeah, I think now there's this in front of him. He could either just give the learning objective.

It gets complex in the details because just like our individual sensitivity. Yeah, yeah.

And then it is this broad LLM agent model where it makes. This is good direction. We'll obviously keep posted over the weekend. Asynchronous sense of that. And then it creates the experiment, the dialectics, the questions, answers honestly. In the next Wednesday we'll prep for demo. And I think this is a good direction. I mean this is like. I was just thinking how do we break this? Like. Like let's make it more complicated. So let's say like. So we only simulates it and then provides. Here's a bunch of things we tested. Cool. Greg, take a look. The first real like super backlog prioritization we've had. Right. It's like, it's fun to be in this cycle. Like. Like we slacked about. Right. I really want. Let me know what you want to do versus the other one would be. Greg goes in and he's already thinking, well there's certain who are looking at metabolic health. So now let's say. Because the bigger question is existential risk. So expanding scope a little bit instead of metabolic health, maybe it's on a micro scale smaller. I'm gonna spend a lot of my free cycles in the developer experience as if I were Greg. As if I were gonna ship apps on top. Because questions we wanna ask or a certain type of way we wanna ask things and I wanna input that. Then I wanna run the experiment. Is that. Are those the two paths in front of us?

I think it's like more like then.

I think that will help inform the greatest priorities. Right. Like, or. Or I will. I will be able to speak to scale. Like a bigger scale would be like how do we incorporate maybe the fact that nations are still waging war. That experience a lot better. So now that we actually have like okay, cool, we have stock partnership. Great. I see this.

What is the actual AI coaching experience going to be like? You know, is it just something that.

You could is on the path? Is it prioritizing your use cases? Then I could just funnel all my time to this and I'm working on.

Streamlining all the content you can ask Brian anything. In the more constrained example, I'm thinking like he's going to, he's going to identify that a user needs to answer.

And some of these populations live in a war zone. So that's, that's one use case I just had in my brain stuff. So it takes me less time. I think that the podcast editing is the thing that's killing me the most right now.

So once a certain set of questions and like they, they have to provide answers for all those before they can get to the next set of questions or something.

I streamline that. Then I'll just free up more cycles for the next big important things. I did want to share with you. I had a great conversation earlier with.

Something like that and then I think there's a lot of precedent like the product you're talking about.

Rasa last time a further off use case is like shit, what if there's a meteor coming? Like we don't. There's something unknown coming. Right. So that, that begets the requirement of oh, okay, we gotta have settings.

Yeah. And, and I used one called Landbot in the past and so on Landbot.

Like I would just Juan Carlos, I think from the sidebar community he is CPO at some mental together like multiple.

Choice questions like one after another. And you know, like that was fine.

To collect health company but on the side he's like bootstrapping an app. He was really interested in answers up the wazoo in the universe like somehow but. But micro interaction to break the information from the user.

So you know, just what's most useful right now.

Mental models. He's very obsessed with it. And so, so obviously we nerded out about that a lot and he was like wait, like what's your app? I'm curious. And so that was just.

Is one thing I'm trying to, I'm trying to hit like how can we make this for people that really know how to build an AI coach, You know, how can we make it useful? Maybe provide a set of evaluations across a question.

Something cool to share. I had another good conversation. I saw you had a couple good conversations too. What were they? The model and then let's evolve it. This is a fun thought experiment is being around people can also affect like very toxic people can also affect like your well being which affects your stress, which affects your emotional well being and stuff. And that could lead to some level of risk too. So maybe start there as a use case. Then we evolve into the maybe you're.

In war zone tree.

I think at the heart of this are two questions. From my perspective. It was about from your standup yesterday or I think it was yesterday.

Did I tell you Melissa works at True Diagnostics? Uh, so she's really like, she can be like a really key informant and contact.

Um.

Let'S see. I heard back from this one is.

Do people building AI agents or chatbots today with the existing models or workflows of. Here's a very constrained experience. The meteor use case. How. How might we take that into account?

I think it starts to look a lot, a little like the names of it.

You have to map together the multiple choice questions and stuff like that. Do they like that experience? Like did you like that experience woman.

Courtney that we went on the don't die hike with and she had the psychologist and the mental models for matching. It was like better than a form. So we, we had like one experience. It was a type form and then one experience making. She finally got back to me on those. Yeah. Yeah. And I don't remember what other conversations I mentioned in Stand up. Let me check it out.

It was just that the Victor and. Yeah.

So that nation state. Right. Or like, you know, because you need to.

Yeah.

Well if we're needing to maintain the metabolic health of all these 5 billion agents, then like let's make sure that we have nutrient supply chain.

That was yesterday. We didn't. I don't, I don't think we debriefed about that any. Anything there.

It was like that was a land bot and you know, people just like some people prefer one over the other. And so we just had boats that are, are like well stocked not just for today and tomorrow but for like weeks or months in the future. And that's available.

Okay. So. So then the other question is Greg's immediate experience would inform that Victor has.

This habit of like putting stuff on my calendar without telling me like. And that's how I woke up. I woke up after the question like.

How from a development experience do I like this to implement it to be useful.

Meeting started and I'm like, oh hey dude, you guys still there? So we only have 15 minutes. But he's been, he's been like building Melissa up to me and building me up to Melissa. And so we just met and one.

And then there's the bigger question that I think you're asking, which is which path technically one takeaway.

I think that Alyssa mentioned that stood with me was like. She's like, I'm so impressed that you're working with Brian. He's you know, like one interesting heuristic is like how long could a country survive today if it's like shipping channels.

Gets us the best value for the end user. Keeping in mind Brian would, would kind.

Of got such a high bar for who he works with. And so it really made me think.

That like the like his thoughts on this would matter a lot too. Like one use case I'm using to stretch this is what if somebody said hey Brian, I want to commit suicide, what do I do? Like what does the agent say? What do we allow them to say?

You know, we've done all the right things to get in position. But like I think we should like just over deliver, you know, in the sense of like expect that we'll like expect that we're like sort of applying for a very important job or something like that with the quality of work we deliver.

Yeah, yeah, that's how. And that's a question like I think the Don't Die team would, would need to help us understand, which I think.

And I think in like Taiwan you might have like 30 to 60 days worth of food or something like that. Um, so this is something that like you know, an epistemic institution that's acting for, for health purposes would want to like extend out pretty far. But on the other side of that I think you have, you know, you have.

Actuation of, of knowledge would drive a little bit of the philosophy here in my mind.

This new paradigm of, of CHAT GPT.

I'm reading it and that's also why I'm like so much trying to.

It's just like it like Brian's probably looking for something like that, but it's just like him and then it comes with all the guardrails of ChatGPT also and like you know, that kind of thing help them.

Yeah, because just my whole thinking is like just keep delivering them value without asking for anything in return. And at some point it'll be so much value that whatever we ask for it'll be like oh yeah, come on the pod. Give us a shout out, help us get more contributors, maybe fund this slash.

Work and physical labor that like you need your population, the aggregate metabolic sort of energy to be applied to so to like, you know, act on the actuation of the system. And so you're going to want to be.

Yeah, so then I guess because that's where my head's going is you want to introduce to Andrew Huberman once your AI is up and you're happy with it, like we should be treating every sprint set to Be a free form experience if we index on. I want to gather as much belief data from you as possible.

Taking measures to protect and make sure that people are in psychological conditions and you know, like working conditions.

Right. And so like I was think I was kind of imagining the chat GPT experience but with like the belief mirror.

That they're, they're, they're safe and they're able to perform tasks you know, according to the, the benefit of all.

They're sort of integrated and then you know, obviously health focus and it's Brian. So then I guess that begs the question how does cycle as like we're fighting for life. Right. That's how I think about it. So yeah, let's keep keeping each other accountable to like toeing the line of perfectionism with these deliveries. Yeah. And, and honestly it literally, I think it comes down to just me and you talking more about the focused use cases and then doing thought experiments of being in the user shoes, you know.

Assuming that they're actually responsible for keeping, you know, like we're taking internal resources.

Does ChatGPT do it its rules? Is it like a pretty basic rules engine where it's just data labeling tied to a rules engine that they maintain centrally?

Well, I think in our case we're, we're kind of just to like evolve health needs. Right. So I would start there like, like from the bottom up like at the GPT layer we're like, you know. Yeah. Like we'd fine tune it by creating a.

And then me with my relationship with Greg, I could just actually take these thoughts, go to him, get kind of.

Like metabolic supply chains and from the top down like real knowledge work and physical labor and GPT like as a set of questions that here's what AI.

Brian Coach inside feedback loop. You know what's been good is like I asked Greg like so are users going to own their own data in the future with health? And he just literally told me, yeah. And I'm like that helps us so much to understand where you guys see value and where you're going. I was just like bro, I literally just asked this and boom.

Productivity gains economically and that starts to look a lot like a nation state would ask please use these questions first when trying to understand the user and then extrapolate to ask the next question. Best question if it's not available.

Okay, so let's take that suicide use case. Let's say Brian's team is like we want to get roadmap revealed. It's so, it's so a very typical kind of response for that use case. And the reason I this is top.

Of mind is at least as I picture it.

So then, so then to solve the war problem it'd be model a bunch of nation states. That was actually there was a community member that posted that they're thinking about suicide and then somebody else pinged me. I was like yeah.

And it's funny to see all the people who meet this app experience and didn't don't die and kind of be on the inside of it because there was a guy who was like are you guys going to start like advertising soon? I just can't imagine this. All this valuable information is going to be free. And Greg responded like almost right away saying like, well we're going to have.

A reflect our reality.

Yeah. Because I think that's what happens is that you have. You have a bunch of these actors.

Hey, what should we do about this? And I was like oh shit, I should screenshot this. Send it to Greg. And now he's.

And there's like limited resources for metabolic health of their organizations. Right.

But that's literally I think about like how do we handle this use case? I suggested good stopgap is have something that detects that and then life right now. Yeah. I mean there's ego and pride and narcissism thrown in the mix. But like yes, those are just helpful tools.

Paid version with an AI coach and. And comma and other, you know, like other stuff he mentioned.

Yeah, yeah. An auto post suicide hotline. Resources nationwide resources like for that problem. That's funny. And then now they want me to like PM don't die. Yeah, I gotta wrap up that thread with Greg. But I think you know how. I don't know. I basically kind of demonstrated yo I could definitely know my shit with product in all of our interaction Persona to meet them where they're at. But also it would be good to vet with you know, an independent organization that is focused on this problem. We're sort of brainstorming here. But that basically tells me okay, if we're living at the GPT layer for that use case which stretches us a lot. Like this is a very extreme niche.

Use case but you know, like to solve the underlying resource dependency or constraint.

I think I demonstrated that. And then on our last call with them when we debriefed, Dean said this best. He was like we clearly have thought about this way more than they have. That was so fucking clear. I find it useful as a way to kind of min max or stress test our. Our thinking to drive direction of this is I would imagine that Brian would.

And that that's why I think like, you know, having abundant energy is one of the things that unlocks so now.

And so I think they might see that and they're like, wait, this guy's know. These guys know where the value is. Like let's kind of bring that in. So. But back. Back to what you said. Yeah, I think Brian does have a really high bar and yeah, it's, it's a testament that we made it.

But you have to optimize these epistemic institutions to like be militaristic, you know, or at least like I think boundary systems is something that all homeostatic life has and you know, probably want some.

Level of control over that. And so his developer would probably want some way to easily put in, hey, if it's topics of this type, we have a canned response. And they want to know that that gets to 100% so. So backing up from there, let's assume that is the case. We got to keep earning it. Yeah, I think that's a good, that's a good philosophy. Got to not. Don't let the partnership die.

There you go.

Yeah, there you go.

If you don't have to allocate all your energy towards like gaining new ground, taking new territory for resources for your.

Survival, then that seems reasonable to me based on the interaction I had with Greg about that particular use case. Okay, let me wrap the pod. Unless there's anything else you want to talk about.

Well, the just Dean's update and put it more into the epistemic knowledge work of your system and do more research to like, it's like, that's really exciting, you know that he's asked Toad Yellow and so this him in the wild.

Not them asking Coach, but them just posing it to the community. Then does any direction we're going here with user constrained versus broad ELM agent, Does any direction constrain that looking at.

His own personal health data is going to be a big deal? Like, you know, he just, he's just so adept. He'll be able to figure out like what we can pull in from like say an Apple health protocol or something like that and reduce other existential threats.

If we don't gotta worry about the lower level like requirement.

No, I'm thinking about, you know, how the CCP did this with the deep Seq Maslow's Archimedes.

Yeah. Then like it opens up the mind more.

That's gonna be really cool, dude.

I'm pretty stoked about that and Dean seems stoked too, which gets me service. Yeah.

Like they have, they have a parallel LLM like monitoring all the results. Yeah.

And he's taking care of his health, which I think will free up more capacity for him. Honestly, like when your battery's low, you just got low capacity at the end of the day. So once he takes care of his batteries, I think he's going to have a lot more energy. Yeah, I was excited to see that and see that he's just working on his health. Back to a parallel thing we were saying. It just allows more people to be asking the right questions and trying to solve.

You ask about Tiananmen Square and the secondary agent comes in is like so you could have a secondary die agent.

Those things collectively and the accumulation of that knowledge work over time. I think in your words.

And if like, if, if like something is higher on the die scale, it could interact, intervene.

And then he said he's going to share a prototype with us on Sunday that might be also interesting to incorporate.

And be like hey, we're hearing a lot of DAI from you right now.

Into the demo. Depending on the fidelity of that.

So oh, are you available on Sunday?

Kind of like that direction. Honestly that as we noodle through it, that seems the best to me. He said, yeah, he said work towards sharing a prototype someday. On my findings for the causal debugging system for his health.

Oh, okay. Asynchronously, because I thought you were, you were unavailable to meet will help the.

System actuate in the quote, next best way or right way. Okay, now what about the meteor thing? Is that in my mind I'm like all right, we really can't with that. Like we ain't at a gut level because that, that, that allows for control, but it's also flexible.

Yeah.

You know, so but and then from a user experience perspective, the developer wouldn't have to. Probably need somebody to go figure out how do we even sense things that are. We could sense things that are, I think hundreds of like on Sunday I'll probably just text you guys. But yeah, asynchronously, if, if you guys could.

Cool. Yeah, that's all good.

Maintain all these like rules in like a typical rules engine. It would just be an LLM with kind of a. Cool. Thanks a lot everybody for joining us. While we ripped about years away. But then some of the problems are like the image processing, image quality and like speed, you know, evolution of technology, sense making intelligence, fine tune prompt set and some content data.

Right.

Like and honestly models of self agents in the world. Everything how we're cows now its mirrors all get image. So then maybe this epistemic institution has to evolve its scope and be like we're going after that.

I don't know how additive we can be to that. Unless it's really like. Unless we can articulate it as a philosophical divide, you know where like the philosophy described is one of don't die.

And then like now to even get the right data to make sense of it.

Yeah.

What do you think?

Do you think that we can actually like auto identify things that are anti the philosophy, you know, on a scale.

That'S foundational requirement by the way, did.

You see that headline that the risk of an asteroid hitting Earth and this.

Was the way down and literally going through our backlog refinements together and talking through a use case for the Don't Die team and Brian AI. So thank you. Thanks a lot for joining us. As we build in public. This has been the ABCs for building the Future podcast and we hope you had a great time. Click subscribe.

But I'm just more thinking about like the utility that we can provide particular asteroid they increased the risk from like 0.2% to like 1.6% yesterday.

If you want to hear more from us and go to our website if you want to join the community, have a wonderful day.

I, I'd rather think about these edge cases later. Cause I know we can just overlay a layer, a safety layer on top of whatever we do or something like that. Yeah, this is my news headline.

Wait, what? Wait, wait. That.

Yeah, guardrails.

So I think actually what's going to help us the most is if we just get this down into a slack message and I just shoot off to Greg and then he just tells us because there's a more than 1 in 100 chance that we're just going to get hit by a meteor.

Oh, it's like 180 miles wide or something like that.

Because he's been very open in dialogue with us. So I think he's really open and just literally just giving us the priority at this point in the relationship building. So how about we do that with this particular one. Give you direction I guess.

Can we propose like this is what we were thinking, you know, we'll be working on this and you know, let us know if you have any thoughts because I, I think it's something. Because of the state of the world today, it's like sort of obvious like okay, ChatGPT found product market fit now, now that is huge. Yeah.

What? Oh, that would be like northern, like a fifth of California. Like hang on, that's a lot.

It'd be an impact.

What's the square mileage of California? This is also funny. I just asked AI because in the past I would've had to go figure out the square mileage of California and then do some calculations, conversions. AI just did that for me in real time. We are showing Benjamin's model is correct. I am making sense.

VOD is copying it now. Deepseek is copying it. And that's the form of AI Coach that everybody wants. Like, that's like Greg shouldn't need to tell us that. We should know that. I guess. So then I think what we're describing is here's how we would.

Who are thinking now from the artificial intelligence. And I'm evolving my own intelligence to.

Evolve about enabling you to program a chat GPT like AI Brian Coach, let.

Us know if the language that I use in this podcast.

California has it sounds about right. And that's where I would describe like.

You know, a total of 163,000 ish square miles.

180 miles. I mean, here's one option where we can provide a prompt of the. The question exploration space.

That's, you know. Yeah.

I mean that.

Yeah. So then it's 180 by 180. Right.

So here's another option. We can provide the embeddings in evaluations for the question answer evaluation space.

When you say embeddings, explain. Could you explain that a little bit more? What does that actually look like?

Well, just let me just look up Tharu.

Yeah, That's a lot. 32,000. 32,000. Oh, that is about a fifth. Wow. Thank you math teachers from the past for giving and engineering teachers from the past giving me some level.

It's like, you know, if we're going to encode a question answer pair in like a vector database, then it's like specifically encoded with the embeddings of the LLM where it's going to be true trained on. And then you can sort of like. So you can have that question answer pair as an embedding.

Good estimation. I'm kind of proud of that. Sick. Wait, that's.

That's almost actually on with an evaluation score. And, and you can like, you know, use that as a dot.

One fifth. Exactly. Like almost. Cool.

Oh, so it's only a 130 priority queue for how the bot would ask like, you know, query and ask questions and you know, a lot of this I'm Just, I'm just asking for time to like, let's, let's think this through and I want, you know, bounce some ideas around because it just.

Yeah.

You know.

Yeah.

330Ft long, the asteroid.

Oh, okay. Okay. 180 miles, bro. Like that's. Wait, 130ft.

Yeah. And it's, it's 1.3% chance of risk and it's 2032 where it would have a potential impact.

Okay, that, that makes me feel a little bit better because 130ft, I could better imagine that. That's like it's a building. Yeah, that sounds like a building. I'm just thinking about my apartment, square feet. Okay, cool. I don't think we're going to die from that one, homies.

Like a giant.

We got this. We just.

Yeah, shipping.

Cool. Sorry to scare everybody on the pod. We're looking at this. Does this change the scope any meaningful way going either way, or is it just more of a. I just need to know which one would be the most valuable. Doesn't change scope.

It just like.

All right.

It'S like a flight or flight kind of trade off. Right. Like the specific output that I have, you know, like I'll tell you, my tendency right now is to go and learn the machine learning best practices for building question answer bots and then like follow those heuristics. And I think that's like where you know, from a sense making perspective, there's all these things that we need to be doing at a higher level, cognition wise, like finding food mates in our world. It's, it's knowledge work going to guide me towards like formal question answer evaluations and there's probably some things I need to learn that I don't know yet. So that's like the feeling that I have. And if, if you were to say like no, like let's just generate a prompt with like some, some questions and like see how it performs. And then all of a sudden there's an asteroid coming at you and like nothing else matters. Right. Or if it's a tiger or whatever. And so it, then it, it just shortcuts epistemic capability, you know. I don't think it fully addresses like the concern of will it hallucinate or not. But it's something and, and it's quicker.

Yeah.

And, and you have to have sort of like pre programmed routines in your survivability history, things that your ancestors have already solved to figure out this problem that you're, you're relying on. ChatGPT won't help you.

Yeah, I guess another question I have to make a decision is if we go that path and next meeting they're like, I, I want to do this user provided embeddings. I want the, you know, more control. I want this to be in a more typical machine learning chatbot workflow. Then how would that. Would we have technical debt accumulation is I guess is what I'm saying. Like, would it be a lot to change direction at that point if we.

Demo the tiger in front of your face? Yeah.

No, but maybe.

And that was like what you're talking about the back propagation of like, of evolution, right? Like if we get our AI plugged into the sensory loops that happen at like the reflex. Reflex level of our nerd, we go.

With the broad, here's a prompt. And then next meeting they're like, actually I want to see this develop.

No, I think it's more maybe a question of first touch. Like what do we bring to the table system? You know, say 50 milliseconds and below, you know, like yeah, that can absolutely be like I think they might be looking for us to say like this is the best way to build it and, and they may not know to like ask for additional things if we, we provide like a prompt technique that doesn't really work that well. Like an adaptation that could evolve. Years of evolution have taught us that. We've now figured out like how to back propagate even more.

Like I'm thinking Brian being a philosophically minded dude, like he sees himself as a philosopher. When I said dialectic to him, he is, he, he like understood and he talked about the don't die dinners as a series of dialectics. So I think at some point he.

Would knowledge into that as system.

Definitely. And as you say that, I mean there's also the other take that, you know, humans and AI evolve in this hybrid way together. Probably like want to see this? And I also think that he would have a mind for. So like one example, I always think about what's that boat where if we replace every part of the boat, is it still the boat? There's some boat. There's that philosophical problem. I forget optimizing dialectics. So one question I have is like, could we go the fast route for now and just like seed it based on our extracted belief systems from his content? Like give this LLM that perspective. Enter the show notes. And so one of the things as I'm listening to Lifeforce, the book by Tony Robbins featuring Peter Diamandis, is there's all these like therapies going on and one wonders when. If you replace every part of the human body with some synthetic part and just feed that and just say hey v1 this is what we did. What do you think? Because it kind of fits into the model of okay, Brian's going to be making new content all the time. This way is less bottlenecked. This way is more bottlenecked. This is my gut feel at a high level. Or you're something new right. Like you're not Homo sapien anymore. Like I'd imagine especially if and I think it brings us a little bit closer to just evolving towards the overall. You tie it into literally like this has instantaneous reflex level feedback loops to your brain to do things with you sense making actuating all vision. I'm kind of a fan of us riffing through this. Then you've become something different. Yeah. Well I'm sure you've thought about this. So that take perspective self of Brian generate these. That's the demo and then if we can have time get it working in dialectic designer. If it will take four hours. Well I guess it's four hours. You mentioned if it's not like actually hooked up to the data it's just stubbed out. So do you have any evolved thoughts on. On that or additional thoughts on that?

It reminds me of my. The. The consciousness complexity index you know like where it. Yeah like if you. If you hold agency for what it is that that makes you conscious then and you don't like read.

But yeah that would be my order is try to get this done. Go with the perspective self Brian.

Reduce in complexity from that and in fact you increase in complexity from it.

Then that seems this route Dialectic Designer next whatever you can get done of that and then hook it up after that if you can.

Okay. Yeah it's really helpful that makes sense because I would so I would just like I would focus on like populating largely value directed. If you have you know sort of like an evolutionary lens on. On life and yeah you could replace yourself piece by piece. But what. What can't be replaced is Brian's like the. The self model for Brian with his belief system the ongoing conscious loop of consciousness since making an agency that you have between yourself and your environment and that's where you know like you could easily just kill somebody based on documents and then use that belief system as context to ask the next best question. So that what this trainer is going to be doing is going to be saying Brian Johnson wants to learn about this user and it's going to be simulating that and create something different in the process of replacing them piece by piece. Or you could have a measure and I think that measure of the consciousness complexity index where you really have like a mutual information score of all the interoceptive and extraoceptive processes that are going on. What the environment is, how we control it, what the interaction. And it's going to be promoting the next best question that Brian Johnson would ask in a given conversation, I think about the complexity of the environment, how many objects we can manage at a time, what's in our interception, how complex.

Internally that will get their light bulbs turned on for sure. And then from there we can go into the more technical specifics and provide them our thinking on the decision, like what we just talked about and then ask them like here's the rest of the stuff we got to build for you. How would you. Are you good with this for mvp? And then we improve this part over time to give you a bit more.

Control for the can we respond to our environment? Like if you can maintain that, then you can evolve it.

It's like the complexity, that factor is essentially like a weighing factor to balance right in this provided embeddings or user specified embeddings. As we work on these other items in the roadmap, like kind of complicated web of systems. For the audience that hasn't heard about Jonathan's cci, could you talk about that? And you touched on a little bit about interoceptive extra. I'm thinking the next meeting we show this show progress and then do another facilitated brainstorm of where I really like this product exercise where I give them like a hundred bucks and it's like you have constrained capital. Pick these features and then that just gets it to us sense making feedback loops. But can you just talk about a little bit of the origin of why you came up with it and then what problem does it solve to you?

It's interesting because I was mentioning, I.

Was watching the really good loop where we're just like, we know exactly what you want next. So that was my plan for this next meeting with them.

This Bratton presentation. He was starting to get on topics of evolution and I'd been working on this idea about assembly theory of consciousness. And then he introduced assembly theory.

We show a bit of value, we show them what's on the backlog. Based on the first roadmap conversation we had them, they'll have that context in their information architecture and mental model. We give them some play money kind of like we did last time. And it's just the K of the slides get them working in their. On our shared artifact and actually just have them rank it for us. Cool.

Yeah. Okay, so for. And for the demo in the meeting into this presentation which so it's just like, like you know, I was there, I was like this is. This is awesome. Would like. So I'll like run the trainer with Brian in the loop and like show the question answer output in a way that like comprehend. And then it would be. And the. Okay, so the assembly theory of consciousness, it was a problem solved. Like imagine you have data from a distant solar system through a telescope and you have like, you know, pixels with. I don't know if you want to this be good use of time but I think like if we acted like the user and then we asked Brian a question and saw response colors on them and stuff like this and you're trying to understand like what molecules might be on a planet in this. In this comeback and like just did that together. Yeah, I think that would be like best. The best demo.

Okay so yeah, I think this is good direction. We'll obviously keep posted over the weekend asynchronously and then next Wednesday we'll prep for demo and I think this is a good direction. I mean this is like distant solar system.

And so there was a guy who's. That was his job was like figure.

Out questions like that the first real like super backlog prioritization we've had. Right. It's like it's.

It's fun for NASA. And he like yeah realizes there's this inherent relationship with chemical particles that like the more complex they are, the more interactions they have gone through. And so he likened that to the.

Process to be in this cycle. Like like we slacked about. Right. I really want of evolution like to.

Say like that each individual interaction was sort of like an evolutionary step or turn in its.

I want to spend a lot of my free cycles in the developer experience as if I were Greg. As if I were going to ship apps on top. Because then I think that will help inform the greatest priorities. Right. Like, or. Or I will. I will be able to speak to that experience a lot better. So now that we actually have like okay, cool. We have SOC partnership. Great. I see this is on a path. This is prioritizing our use cases. Then I could just funnel all my time to this and I'm working on streamlining all the content recycle and that.

The more complex the molecule, the more ancient it must be, the more complex it must be. The more evolutionary cycles it must have gone through. And so it he. He came up with a formal like.

He calls it the stuff so it takes me less time. I think the. The podcast editing is the thing that's killing me the most right now.

So once I assembly theory and you know it's. It's about how particles are assembled and a chemical from a chemical, you know.

Streamline that then I'll just free up more cycles for the next kind of big important things.

Scientist and but it applies to then how biology happens. Right.

Like I. I did want to share with you. I had a great conversation earlier with Juan Carlos I think from the sidebar community. He is CPO at some mental because.

It'S just really the same kind of mechanisms and and this is something that I like. You know I just. I just love to study evil health.

Company but on the side he's like bootstrapping an app. He was really interested in evolution because.

In the sense that it's our history and I really like kind of you know spent a lot mental models.

He's very obsessed with it and so so obviously we nerded out about that a lot and he was like wait, like what's your app? I'm curious. And so that was just a lot.

Of time reading historical narratives of how these especially breakthroughs evolutions have happened. And I saw the same thing. It's just like there's always like it's.

Not something cool to share. I had another good conversation. I saw you had a couple good conversations too. What were those about from your stand up yesterday or I think it was yesterday.

Did I tell you Melissa works that evolution goes linearly through these more complex adaptations. It's just that like you do true diagnostics. Uh so she's really like. She can be like a really key informant and contact.

Um.

Let'S see. I heard back from this tend to get additive adaptations that like find fitness like with woman Courtney that we went on the don't die hike with and she had the. The psychologist and the mental models for match more complexity and sometimes you get more complexity that doesn't find fitness but it gets. But the more complexity gets passed on gets integrated and so on. So anyway I like that theory and I just thought like I was. I was trying to talk about Michael Schmicking. He finally got back to me on those. Yeah. Yeah. And I don't remember what other conversations I mentioned in Stan Levin's work and like how to explain because Levin's work has to do with like cellular biology and how Cells line up. Let me check it out.

It was just that the Victor and. Yeah, so that was yesterday. We didn't. I don't. I don't think we debriefed about that any. Anything there.

Um, it was like, have ion channels that form bioelectric fields that can like build our tissue systems and our embryonic process. And Victor has this habit of like putting stuff on my calendar without telling me. Like. And that's how I woke up. I woke up after the. I was. I was in a forum where somebody was talking about. Well, his theories really don't approach how stem cell drift happens. As in like the stem cells in our body have actual epistle, epigenetic and genetic drift over time, which is. Meeting started and I'm like, oh, hey dude, you guys still there? So we only got 15 minutes. But he's been. He's been like building Melissa up to me and building me up to Melissa. And so we just met and one, one takeaway. I think that Alyssa mentioned that stood with me was like, she's like, I'm so impressed that you're working with Brian. He's got such a high bar for who he works with. And so it really made me think that, like, you know, we've done all the right things to get in position, but like, I think we should like just over deliver sub intracellular. And I thought like, I think Levin's philosophy actually talks about this. His science doesn't as much, but his philosophy does because it's like about how things assemble. And so I thought, how can I articulate this in a way that like, you know, we can see this. And I kind of, you know, went on ChatGPT and created this theory about the assembly, you know, in the sense of like, expect that we'll like, expect that we're likely theories of consciousness. And it was fun because I just like, you wouldn't ever ask a human being to do this. But I was like, I need a theory sort of applying for a very important job or something like that. With the quality of work we deliver.

Yeah, yeah, that's how I'm treating it. And that's also why I'm like so much trying to help them because just my whole thinking is like, just that.

Explains all of conscious life through history and how everything evolved and how everything's ancestors are defined.

Keep delivering them value without asking for anything in return. And at some point it'll be so much value that whatever.

And I just listed out these. Absolutely crazy. And I said, these are acceptance criteria.

Here you go. Intern. Figure it out. Intern.

And it says, okay, here's the assembly theory of consciousness. And it like basically said, I've solved all your acceptance criteria.

Ask for. It'll be like, oh, yeah, come on the pod. Give us a shout out, help us get more contributors, maybe fund this slash intro us to Andrew Huberman once your AI is up and you're happy with it. Like, we should be treating every sprint.

Here's a Python script that demonstrates like this how, how things evolve and it.

Made the cycle as like we're fighting for life.

Right.

That's how I think about it. So, yeah, well, let's keep keeping each other accountable to like toeing the line of perfectionism with these.

It noted like, this is not an empirical model, this is just a thought experiment. But like, for like, you know, 30 minutes of work on my half, like, that was pretty good. And out of it came this idea of deliveries.

Yeah. And honestly, literally, I think it comes down to just me and you talking more about the focused use cases and then doing thought experiments of being in the user shoes.

The consciousness complexity index, which is like, so assembly theory has a complexity index. And so I just applied that to like agency to say like, you know.

And then me with my relationship with Greg, I could just actually take these thoughts, go to him, get kind of inside feedback loop. You know, what's been good is like I asked Greg like, so are users going to own their own data in the future with health? And he just literally told me, yeah. And I'm like, that helps us.

As you know, I just have enough background in thinking about agency subjectivity in consciousness to have mathematical so much to.

Understand where you guys see value and where you're going.

Yeah.

I was like, bro, I literally just asked this and boot a model to.

Say like, I think this is calculatable.

Well, what's that math model look like? Well, like in, in assembly theory, I might. I'm not sure.

Oh, the assembly theory. Oh, there's.

Yeah.

It's basically just like, you know, like it's an information theoretic number. I forget exactly the math that's used.

But it just roadmap revealed. It's so, it's so funny.

Yeah. And it's funny to see all the people who meet this app experience that didn't don't die and kind of being on the inside of it. Because there was a guy who was like, are you guys going to start like advertising soon? I just can't imagine this, all this valuable information is going to be free. And Greg responded like, Almost right away saying like well we're going to have a pitch that like here's. Here's the complexity of the molecule of water. Here's the complexity of the molecule of the. Of the element of oxygen. Um, you know and like you can get a number one's higher than the other and paid version with an AI coach and comma. And other you know, like other stuff he mentioned.

Yeah yeah. So he's coming at it from. So I'm really intrigued because I have a chemical engineering background so I used to love thinking about chemistry. That's funny. And then now they want me to like PM don't die.

Yeah.

All the time. So it made me think about the analogy of I gotta wrap up that thread with Greg. But I think you know how. I don't know. I basically kind of demonstrated yo I could. I definitely know my shit with product in all of our interactions. I think I demonstrated that. And then on our last call with them when we debriefed Dean said this best. He was like we clear a tree, we cut it, we see the middle. We could read its history. Right. But. But this is like the low have thought about this way more than they have. That was so fucking clear. Bliss level, tiny tree you have inside you the molecule molecular level. Right. I mean there's an even lower level but like. And so I think they might see that and they're like wait, this guys know. These guys know where the value is. Like let's kind of bring at this level is. Is quite sufficient I think for this. This theory. So just to get my gut in. So. But back. Back to what you said. Yeah. I think Brian does have a thinking straight. Is that every element and molecular compound has a different number really high bar and yeah it's. It's a testament that we made it. But we got to keep earning it. Yeah, I think that's a good. That's a good philosophy. Got to not don't let. Or is it also the actual matter like not just the partnership dying.

There you go.

Yeah, there you go. Compound in terms of theory like on the periodic table or. Okay, let me wrap the pod. Unless there's anything else you want to talk about.

Well the just Dean's update in organic.

Chemistry problems and equations.

But it's like that's really exciting. You know that he's kind of asked code yellow and so this him actually.

Like let's say for example, I don't know like even this piece of paper.

Has like looking at his own personal health data is going to be A big deal. Like, you know, he just. He's just so adept. He'll be able to figure out, like, what we can pull in from like, say an Apple health protocol or something like that.

And some carbon. Right. I've written on it. And now that carbon's changed in some way, it's been interacted. There's been.

That's gonna be really cool, dude.

I'm pretty stoked about that. And Dean seems stoked too, which gets me stoked. And he's taking care of his health, which I think will free up more capacity for him. Honestly, like, when your battery's low, you just got low capacity at the end of the day. So once he takes care of his batteries, I think he's gonna have a lot more energy. Yeah, I was excited to see that and see that he's just working on his health. Some abrasion. Right. Its molecular structure has changed a little bit. So does this theory account for that change?

Assembly theory? Yeah. Yeah, I think. I think so. I don't know exactly how it sort of like categorized the ink on the piece of paper, if that would be its own, sort of like metamolecule or whatever, but.

And then he said he's going to share a prototype with us on Sunday. That might be also interesting to incorporate the idea. But it's something in that direction, though, of like matter into the demo, depending on the fidelity of that. So.

Oh, are you available on Sunday?

He said, yeah, he said work towards sharing a prototype someday on my findings for the causal debugging system for his health.

Oh, okay. Asynchronously, because I thought you were. You were unavailable to meet.

Is changer aged or something? And at a molecular level, we can actually attribute a level of change. And that's kind of assembly theory or.

The complexity of that. And even an example you gave, like, where are they on Sunday?

I'll probably just text you guys. But yeah, asynchronously, if you guys could.

Cool. Yeah, that's all good. Where'd that paper come from? You know, it like it grew from a tree, you know, and how did it.

Cool. Thanks a lot, everybody for joining us. While we ripped about, you know, evolution of technology, sense making intelligence models of self agents in the world, everything, how we're house now, it's mirrors all the way down and literally going through our backlog refinement together and talking through intrigue.

It there and like it's got its own history. But then. So every single sort of step where the molecule gets more complex is just part of the assembly process. And then the last use case for.

The Don't Die team and Brian AI. So thanks a lot for joining us.

As like putting ink onto the piece of paper, it's just like that's the end of the process for right now.

We build in public. This has been the ABCs for building the feature podcast, and we hope you had a great time. Click subscribe if you want to hear more from us and go to our website if you want to join the community. Have a wonderful day.

Right.

Oh, well, is it. Because now it's in my mind what I've written down made sense of at a. At an even lower level, maybe not. Right. Because. Because like, there was something that happened.