<3/notes/

personal artificial intelligence

general intelligence is to do all of my chores for me my personal artificial intelligence will be to help scale me 

in a way this is really an opportunity to personalize your experience of life itself and take control over how you go through life. we think we have control today but i think that's largely an illusion. the information that we're fed what is prioritized et cetera is actually really in the hands of those running the platforms whether those are social platforms and increasingly the frontier ai platforms 

we are all incredibly susceptible to being subconsciously conditioned in ways that we may not even realize that has been shown time and time again that these platforms are capable of doing that to us.

so when i think of personal ai i also think about freedom gaining greater ownership and control over my own life experience. 

And I think that's a very powerful framing that is important to talk about. I'm not sure where. Also, the way I'm organizing this note is this is not just what's gonna be in the essay, because I think the essay is going to be the core foundation of this exploration, the North Star, the mission statement, what have you. But I'm also going to express a lot of the asides that I have, tangents that may end up being sort of smaller x-posts that quote the essay and things like that. So, basically, we will sculpt and figure out what goes where. This is mostly just a dump of everything that I have on this idea at the moment.

from a presentation perspective i think this is going to be one of the most fundamental essays that i write i think what i cover and how i cover it needs to be in a form or in a way where i think it'll be sort of timeless at least for a few years.

the flow that i see right now is first of all starting off with what is personal ai compared to general ai and the difference in objectives. there's a very clear difference. general ai is going to do all of my chores all the things that i don't want to do all the things that don't have space for my uniqueness. and personal ai is going to help me scale myself.

and then we get into what it even means to what is myself? so there's a theory of self that i need here. i view myself as basically a set of functions that take in inputs do some stuff and generate outputs. the functions are: observe understand value react and act. the core philosophy here basically stems from what the buddha said about the nature of the mind when we talk about sannyā viññāṇa things like that.

once we have a sense of what i am then it actually becomes pretty clear what it means to scale that what it means to scale that is to approach my actual function and to apply it in many more places across much longer durations than i as an individual am capable of

and then we basically get into specific use cases i think when i think about the use cases for personal ai for me it really comes down to three categories: scouting filtering and manifesting.

scouting is to go out into the world into the internet and try to find people ideas and expressions that i would vibe with that i would find incredibly valuable that may inspire me to do something or make me feel good whatever it is 

filtering is guarding my attention making sure that my attention is only exposed to the things that i want my attention to be exposed to. so this will basically act as a lens through which i consume information whether it's from social platforms my inbox or whatever all of those sources of information will now be abstracted away and i will consume things through my personal ai that understands this will increasingly get better at understanding what i want to give my attention to.

and manifesting is helping me translate feelings into ideas and ideas into reality.

and those are actually two different things. right now ai is actually reasonably good at translating an idea that i already have into reality but it still takes a very long time for me to go from a feeling to an idea. i feel like my personal ai can get really good at helping me out at that level as well but i think we're a bit far away from actually a bit of a ways away from actually implementing that 

and even for manifesting ideas into reality it's still taking a long time. if i want to ship a website or a video or whatever there's still a big gap between the way i want things done and the way general ai natively does it. understanding my taste and how i want an idea to be implemented will be part of my personal ai's training about me 

and then over time we can go toward the feeling layer as well which is basically "i have a feeling." it might even be indescribable and i might just prompt it with a few sort of descriptions of what i'm feeling. maybe it can even have access to my feelings in a more direct way. idk. it can now start suggesting things to explore directions to explore inspire me with works of art or products or whatever it is and basically help me pull that feeling into an idea (or into many ideas) and into reality.

and i also want to be clear that these functions aren't as distinct as i'm making them sound the kind of information i want the kind of things i want to discover what i want to give my attention to and what and how i want to create and express myself are all deeply interconnected 

in technical terms they are not separate models but projections of the same holistic model that operates through these distinct functions 

and i think that sort of helps us design the architecture for pi which is it'll manifest and appear in all of the interfaces that i use and things like that. it'll perform different functions that will capture data et cetera for those specific functions. it might even train function-specific models but it will also have a more holistic model of me generally that is used across everything every function and everything that i use it for 

so if we think about the architecture for this i basically view it as a ring around me. maybe not even a ring more like the super saiyan aura around me. are you looking for a haircut? there's me. there's the world the external world. there's my internal world. and there's a sort of semi-permeable layer that is my personal artificial intelligence that acts as the filter the scout and the tool through which i express myself to the world. this layer increasingly gets better molded to me to the shape of me 

a key aspect of this architecture is my personal data. i need to capture all expressions useful expressions of my taste judgment intuition et cetera. all expressions of me my uniqueness. i think that's the much more general framing: all meaningful expressions of my uniqueness.

and that's what my project ghost is about. ghost is essentially a repository/database of expressions of my uniqueness that i intend will get richer and more comprehensive as i build out the infrastructure tooling i need to capture more expressions of me and my uniqueness. you can check out the starting point right now in the repo.

and when i say "expressions of me" it's also expressions that are captured in motion. every chat that i have with my personal ai on how to manifest an idea tweak something or edit something are all expressions of me of my uniqueness of my taste which all have to be captured. i've gotten things off the ground in a very rudimentary way but this is one of my initial priorities: to essentially start capturing good data of me so that we can then start playing with ways to model that and use that for specific functions specific use cases

and the use cases don't necessarily have priorities. i want to do all of them. i think getting the mvp off the ground doesn't necessarily require fine-tuned models or any of that shit. we can get it off the ground pretty soon. i think it's important to find a structure for them where we can clearly observe in a very scientific way where it's at what i actually want the gap between those two and how well we're iterating to close that gap.

implementation notes for ourselves i imagine a dashboard for each function where we can see all of this information and we can use it to improve how it's doing things 

one of the key insights we have to be mindful of is that expressions of me are found in motion not necessarily through instructions. instructions may be a subtle or smaller part of it but the vast majority of information is capturing expressions of my instinct intuition which honestly cannot be represented in instructions. even if i try to represent it as an instruction it actually will fail to capture the essence because i myself am guessing at what the instruction is or the principle that the intuition is based off of. there are basically two very different representations or modes of operating: it's reasoning versus feeling. i strongly believe that most expressions of uniqueness or the most valuable expressions of uniqueness exist in the realm of feeling of intuition not in the realm of reasoning. reasoning will all be eaten sooner or later by general ai.

just some product notes. i think the browser app that we have is going to essentially be the primary core product. basically it'll be pi. browsing the internet is one aspect but whenever i open up a new tab it'll be an inbox of all the things that pi thinks i would want to see want to consume. this includes filtering things that have reached me as well as things that it's found while scouting. the way i interact with that app that page there will all be information that it uses to determine whether a suggestion was actually useful or not et cetera

coming back to the core objective of my personal artificial intelligence it is essentially to scale myself. we need an objective function that we're optimizing for and i think scaling myself is what we're going to do what it's going to try and do. but what is the objective function that we're aiming for here? that naturally will have to do with the personal stock. i think we can have a temporary one. it's good to have a temporary one i don't want to wait until i've fleshed out my thinking around the personal stock for us to have something that we're optimizing for. in the short term it has to be about relationships seeking truth and the friendships that i make on the journey to seek truth. we're optimizing for depth and quality of relationships but because we can't really measure that i think number of conversations we have is a good starting point. not all of them will count toward the objective. i don't want to judge them because i think if i start judging which ones i liked which ones i didn't that's putting negativity out into the world and i don't want to do that. that's like me saying "oh i didn't vibe with this person at all." but if it's presented well i think there might be a way to do it where it's not like i hate the person (because obviously i don't). it's just we didn't vibe and that relationship wasn't deep. maybe there's a way to figure that out. for now the rough objective that we can have is number of connections. it's a bad objective function i admit. i want to be very clear about admitting that but it is what we're going to use to get things off the ground.

because eventually the objective function is going to be the number of bidirectional ownerships that i have and the degree of the stake in each other's success. that's it. that'll be the ultimate. that'll be a much better objective function. but we're not yet at the point in the story where we can introduce that we'll start small and we'll evolve it as the story evolves.

okay i just want to put some thought into the product. i like evolving the browser as the core product here. we'll rename the browser to pi and pi essentially will be the interface between me and everything. i'll have an inbox that filters out unnecessary stuff and shows me what i need to see. pi will be able to notify me across my devices in case something comes up that i really need to check out. for example someone who would be great to connect with reaches out i want to be notified no matter what the time is so that i can respond to them and act on that. because once again that's the most important metric and so it'll be surfaced. it'll have the highest priority if it happens when it happens. a new tab on pi will basically show me the things that it thinks i would like whether it's from filtering what has already come to me or whether it's from scouting and finding things for me 

and that is a very clear starting point. it's a great starting point. we'll have another page that's basically a dashboard of how things are going how well pi is doing. that dashboard will have sub-menus for each function that it's performing.

so how many of the things it scouted have actually led to relationships? each thing will have a funnel.

if i found something do i engage with it? like if i found a post that i would vibe with:

  • did i like the post?

  • did i bookmark it?

  • did i go to the author of that post?

  • did i follow them?

  • after following them did i reach out to them?

all of these things reflect deeper success or deeper positive feedback for the thing that i found. we would want some notion of a funnel not too rigid. because maybe for example some folks don't have an x profile and so i can't follow them but i still reach out to them somehow and that's still very valuable.

it's important not to be stupidly rigid about these things or analytical about it but to have a rough sense of a flow or a funnel for each. then we keep tracking that. we keep tracking that. if it's proposing things or letting things through that i'm ignoring and not acting on then that's a sign that it's not doing a good enough job and that we can improve 

and we can also see all the data that it's actually using. once again all this data is going to be reflected in ghost in the repo itself so it's all there. this pi interface is just going to surface that data from the ghost repo and present it in a beautiful way that shows me what happened what information it's using to make its next decisions or that are influencing its next scout or its next filtering. we should even be able to see the things that it filtered out that may contain things that i actually want. and we should be able to see how it ranked them. there are so many things here to tweak. i think it's very helpful to approach this in a very scientific way especially these functions 

so the mvp for pyre is pretty clear to me now. and i don't think i will i mean eventually it can evolve to have all sorts of things in it but for now this is all i need. for everything else we keep it simple. we're using obsidian for all my notes and we're just working with chatgpt or claude in its interface having it save chats to ghost whenever i'm done with the chat. so that workflow can stay the same. there's no need to one of my biggest weaknesses is trying to do too much at once. i think it's really important to just keep the scope for pyre for this initial app really tight. i like what we have. we have an inbox we have a dashboard and we have a browser where i can go to the sites that i want to and consume or interact directly with them. and that's all we have. we don't need anything more than this at the moment 

and the inbox will reflect both scout and filtering both those functions. honestly there doesn't need to be a distinction between them because it's all the same thing from my perspective. from my perspective it's just what is worth giving my attention to and that's it. obviously that can be things that have already come to my inbox people have reached out to me. that could also be things that scout found that are on the internet. i mean both. in the early days it's probably going to be the latter: things that have been found for me rather than people reaching out to me because i'm a nobody right now

and for scout we probably want a separate browser instance for it that it can use to scour things. or maybe fuck it computer use and just do it on chrome is totally fine. don't overcomplicate things. just do things naturally in chrome and have it scroll stuff scour the web do things intuitively. not an issue. not an issue.

and that's sort of it. we have the mvp and i do want to make pi feel beautiful. i think beauty is a core principle. fuck i need some sort of design doc here that just captures the core principles i have. whatever we do we've got to make it beautiful. that's so important. beauty draws me into interacting more with it and that's a huge benefit. that's why i use an iphone rather than other garbage phones out there. it's why i use an incredible fountain pen instead of garbage pens. beauty in and of itself is incredibly valuable. and so whatever tools i create whatever i put out into the world i want it to be beautiful.