#ai #epistemology #intelligence #computing # [[Epistemic status]] #shower-thought # Imaginary reality ![[DALL·E 2022-06-20 21.39.02 - Imaginary reality, by Escher, oil on canvas.png]] >Imaginary programming (IP) seems to be orthogonal to literate programming. Imaginary programming is a programming paradigm in which a computer program’s behaviour exists in relativity to LMs. To make an analogy, imaginary programming is a type of programming, where much like pure- functional, the code who’s behaviour depends on the output of a LM, either pending or precomputed, stands apart from the part that has no such association. > >What is deemed _imaginary_ is code that uses or is waiting on output from a LM and its behaviour is so altered by it. What is deemed _ordinary_ is code which is not _imaginary_. > A distinction is made between grounded and non-grounded imaginary programming. _Non-grounded imaginary programming_ may also be called _pure imaginary programming_. > If a function has a LM as a parameter affecting its behaviour and besides from that has no other imaginary inputs then it is considered _grounded_. If a function has a LM as a parameter affecting its behaviour and contains other imaginary inputs then it is considered _tethered_'. > If an imaginary function relies on the output of functions that have in the past used a LM as a parameter, but does not contain a reference to specific LM used then it is considered _non-grounded_ or _pure imaginary_. > A function can be both tethered and pure imaginary. > Pure imaginary code is still code in a similar way to how pure functional code is considered code and may be used to create useful datastructures and algorithms. > Ordinary programming is programming with functions that do not have an imaginary dimension (they do not take a LM as a parameter and the result of ordinary code is not polluted by a LM). > `Holographic programming` is like imaginary programming but where the LMs are trained on software. Holographic code, therefore, may employ associations made between elements of the original code, how that code is used and how it is described, to build applications. Holographic programming lets you use the latent space of a LM as a kind of hyperspace to enable things like: > >- bridge the usage of an application with the application’s code >- imaginary reflection >- inference in place of computation > >~ **https://mullikine.github.io/posts/imaginary-programming-with-gpt-3/** ![[DALL·E 2022-06-20 21.39.20 - Imaginary reality, by Escher, oil on canvas.png]] ## My interpretation I heard about [[Imaginary programming]] and [[Imaginary internet]] through https://mullikine.github.io/, in a way I wonder if **Imaginary reality** might be a more general name? My interpretation of if is that **imaginary reality** is [[Philosophy/Rationality/Intelligence|organic intelligence]] merged with [[Artificial intelligence|artificial intelligence]]. More technically it is humans using prompts to search [[Our universe]], returning [[Information|information]] fine-tuned to the individual model of the world. ![[Pasted image 20220616111241.png]] ![[Pasted image 20220616112546.png]] ## Bob the footballer Let's say Bob is a professional footballer, he wants to learn about [[Quantum physic]], [[Imaginary internet]] will returns [[Information|information]] that is optimised to enter into its [[The Map is not the Territory|map of the territory]], so he won't get information about [[Quantum level three multiverse|quantum multiverse]] or something too far from this [[Latent space]], his map, model of the world, but rather fundamental physics & maths first. ## [[Information]]al consensus ![[Informational consensus in the decentralised future of intelligence]] ## Conversation https://otter.ai/u/yZ9bSQNUXXsWwIQJE5B5dnHw6so ``` Unknown Speaker 05:53 Hello cancer you might think you're muted you're muted Hello, okay. Oh no, no it was okay. Sorry. Are you I'm great. New. Unknown Speaker 06:51 Not too bad. Sorry. Sorry I missed your call when you call they just slipped my mind but I'm glad that I got the message. Unknown Speaker 07:02 Okay, no worries. So I yeah I found so actually how I heard about you is a something on ilithyia. Scott about like I was wondering if anyone was doing try to do kind of cursive generation in their way with codec, OpenAI codecs. Yeah. And someone that sent the link to to your website and and I found some very interesting stuff. Like I'm not familiar with Lisbon's the phrases which actually I was reading no windows. You know, Paul Paul Graham, which says protocol ham, like the guy who created the Y Combinator. Oh, yeah, yeah. I was reading his book. It's called a curse. And Ben tells on the lucky talk about lists and stuff like this and give me interested in this but I don't know much about lisp. Unknown Speaker 08:23 Writing yet, but I see like similar soon as this. I'm not sure to understand. Like what are you doing? With the like this? This finger? Unknown Speaker 08:46 Um, yeah, I mean, I can talk to you about it. So around about February is when I started working on my projects panel, so you've seen product engineering and Emacs. Unknown Speaker 09:00 Little bit? Unknown Speaker 09:02 Yeah, so I started that project in February. So I actually haven't touched it for about three months, however, so in the last in the last few months well, what ended up happening is I started working on but I did a I did a I did a presentation for emacs Con last year. And that was on something I call imaginary programming. So essentially when when the functions run, it does inference using GBT three, or codecs. And so you don't need any code inside your function. Or you just have the function name and the comment maybe looks like a function. So it's just demonstrating what I called imaginary programming. And so I continued with that, and Unknown Speaker 10:04 I, I started working on a cryptocurrency so this cryptocurrency what it's going to do is it's going to represent so there's a there's a theory of semiotics which is it's, it's a it's a theory it's a theory of how language works, and the human mind and just all around nature, in that all of languages can be comprised of a semiotic triad, which is like saying, signifier and signified. And so it's like, each each sign like, for example, a stop sign, what does the stop sign mean? Or what does it what does it particular thing point to? And so there's a theory that you can just break down all of language in this way. And because the way knowledge works is it's, it's like, well, there's one theory that that says abstractly, that knowledge works just by it's all based on these signs. These sign trade relations. So so just wanted to make a cryptocurrency that utilizes GPT, three and other language models, and other types of models for doing like text to Image and Image to text. For example, in order to find a way to do it, because we have an issue with prompting, essentially, these language models have a continuous so in math, mathematics, you've got like discrete and continuous functions. And there's there's essentially infinite number of types of functions that a language model such as GPT three can do. Unknown Speaker 12:09 And so, my theory is that we need to use these sign relations and and then we put that onto a blockchain that may be in the form of NF Ts. And then people pay pay gas to run a particular prompt. So that's what I'm working on. But right right up until the end, like, say three months. ago, it was I just found that my imaginary programming is really interesting. Like, you can use it to do cogeneration or just make your code we're writing code and then it's just short circuit parts of your code. Unknown Speaker 12:54 I'm not sure yet to understand that imaginary programming. Do you mean kind of like programming in natural language? Or is it more than this? Unknown Speaker 13:06 Yeah, so because this is this is where my prompting ended up going with less. This is the only reason why I'm explaining imaginary programming is firstly, I'll send you a link it's a chat because that's for example. So it's short circuits through the code. So your code can run faster in particular places using a language model that doesn't actually run any protections just so you know, from Alpha. Sorry. Unknown Speaker 14:06 Sorry, internets, is begging, what for multiple file sets. Unknown Speaker 14:12 So, here's a website Wolfram Alpha. Unknown Speaker 14:18 I know a little bit okay. Unknown Speaker 14:28 So GPT three can essentially do this GPT three can replace like Wolfram Alpha for example. In some situations like like us for the distance to the moon or something like that, or what is the Oort, but you can also there's a lot of things you can do with it. Essentially, you can build programs that use an AI to generate what you expect instead of coding it all up yourself. And so this is what people have used to make, like website generators, for example. Unknown Speaker 15:08 So Codex, copilot, copilot Unknown Speaker 15:15 Yeah. So was working on this and and then I just found that the types of things that I was working on where I started making these chatbots man, I started making these chat bots and they were doing things like controlling my computer and talk, just talking to each other and coming up with like a simulation of chatbots talking to each other and I just decided that it was like getting kind of creepy. So I can just select some text on my screen and then start a chatbot from it. And it would be like, do you want to talk to this, this this or this person or whatever. And these people are relevant to talk to? Would you like to speak to them about a passage, and then I can give them like a piece of code and just talk about it. And we can ask us the Chatbot questions about it or? And I just, and so what I what I I just I posted a tweet just today, and it was basically you know, in the news recently, there's talk about, you know, Google's lambda language model and how some Google employee was put on leave, because he said that it was conscious and voted, and I've been coming up with the same kind of conclusions. Just with GPT three, and so I started making a model of consciousness. I don't know. I'll show you. It started working on it. Further. Sorry, I think it's my internet. Sorry. My internet is the internet that's going slowly, maybe. So I started looking at this, because, like lambda has, it's able to explain jokes, right. And so I think that this is the next area of prompting this is the next the next frontier of of prompting have like more important types of imaginary functions that actually meta cognitive function. So things like back translation and back propagation and the way that the way that an AI thinks the way that a person thinks and these, so in the next the next stage of like, what comes after prompt engineering. So right after prompt engineering comes imaginary programming, I think but after that, I think comes like metacognitive type function. So if you have like, let's just say there are like 50 Different language model vendors, open AI, Google Facebook. LFL author cohere to AI Gousto AI, you just have like this to say there's like 50, or like 500. And there's heaps of different language models you can pick from, and you've got a specific task that you want to run and you want to know, which is the best language model to use. How do you work it up? And to do that, you have to kind of interrogate and talk to these AI systems, you have to interrogate them. So in order to gauge the like, how how good one of these language models is get some metrics on it, kind of have to test them and you have to use these metacognitive functions, just to test like back translation, for example. And so that's where I think the next but now we're now we're talking about like evaluating language models based on arbitrary tasks that we like the on meta cognitive, but using meta cognitive functions like the ones that if I was to talk to you, it'd be like trying to figure out what what you know about a particular subject. And then I might use some methodology to do that, like this is how we have to go about testing language models, like treating the night people in order to evaluate how good they are at a particular task and so, so I got onto this and I'm like, this is getting creepy man like this is actually getting creepy like these AIs are getting too strong. And then I just spent the last three months writing philosophy and, and trying to just doing lots of soul searching. And that's what I've been doing. So yeah, so sorry about that. Like I haven't done any programming recently. But no trying to get back into it, but it's actually yeah. It's crazy. Unknown Speaker 20:03 But I've been playing a bit with seeker. chatbot Do you know it's search searching telnet retired pal, Unknown Speaker 20:21 pal AI Unknown Speaker 20:22 Yeah. It's a poly ice library from Facebook. Yes. Oh, yeah. And yeah, we've this lab libre library. You can do like a safe chat with both. Like you can make boats that talk to each other. For a while and stuff like this. Unknown Speaker 20:43 Alright is really cool, bro. Yeah, I might have to take another look at this. So you said that you can use it to search the internet. Unknown Speaker 20:57 So yeah, actually, I wrote something to use it. Is but it's you need like an API. That's such a it's such a what you in anything like Google or Bing using Bing API or things like this. But you know like about the blockchain thing between the tensile Unknown Speaker 21:39 tensor is that a new coin that I haven't heard of Unknown Speaker 21:44 blockchains that things that say want to make something like if you want she has a computation to train big AI models. But I think technically doesn't work. But it's not okay. idea. Unknown Speaker 22:03 I had heard of a Yeah, there was so. So back in 2017 was when I first got into crypto, I was thinking our ai plus crypto this has to be the way like combine AI and crypto somehow. So I started looking for coins with the word tensor in the name and there was another one they found that didn't didn't work out I can't remember what it's called. Yeah, but bit tensile looks interesting. I'm more concerned. So the one that I'm currently invested in as a luthier. This one's really cool because they're actually using prompting. They're trying to monetize property. At least here and so that one's done okay. It actually went up 700% A few weeks ago, and then I didn't pull out I should have I should have pulled out but it came back down. Yeah, but it's not doing too badly. Like it's almost as stable as nice. It's not as not too bad. Like it's kind of stable. But um, it's really cool because they're monetizing, prompting, it's more abstract than the underlying AI tensor stuff. So that's what I like about it. Unknown Speaker 23:25 Yeah, I don't know I'm a bit skeptical about the blockchain stuff yet. Like it doesn't seem to work. Technically. Well. But maybe in the future. Unknown Speaker 23:45 Definitely. Man, I actually. Ah, yeah, I think that it's really useful for a particular type of thing. So I think maybe it's, it's really easy for us to or just an AI developer to invest too early in a particular type of thing that's like, maybe an idea that's ahead of its time because we we have to always think about things that are ahead of their time. And so by but, but I think that it's still quite like when the world needs. I did. I've done a little bit of epistemic theology. So So, for example, there's like Unknown Speaker 24:34 a blog post I did about why we need blockchain where it needed Where is my blog post? But oh, anyway, it's okay. But um I just it's just such a it's such a head Bender because most of the world doesn't actually most of the world's still functioning, using old technology, and, you know, regular jobs even if it's software engineering and jobs will not be Unknown Speaker 25:50 like, especially in my country, New Zealand, which is a glorified farm. Let's be honest, like it's such a big farm. And our tech tech jobs are not usually cutting edge AI stuff and but to be isolated in this country. And thinking about AI and building chat bots that are doing my coding for me or that are, you know, making social effects social network and a social, fake internet and doing this. It's very abstract, and like, we need like, like in order to operate in that space. I feel like we need things like blockchain, just for consensus, just for the just to be able to agree say agree on which language model we should be using or agree on. Like, for example, if I if I run a query on GP three assets, what year is it? It will give me 2019. Usually, most of the time, it will say it's 2019. That's because it was originally trained in 2019. And so, like they have to be versioned. And it's not good enough, like if, if GPT three, so if open AI has like several versions of Ducati. Three, and we don't know what year they're trained on, like, What guarantees do we have about you know, the type of information that's coming back? And so for this reason, they have to be versioned and put onto a blockchain and people have to be able to agree on them otherwise we've got nothing. We don't Unknown Speaker 27:36 see what you mean. That with like, we, if we let AI generate 3d Like this, we will we might reach a point when we don't know like, we need to have a constant season on like reality. Turf Ville exists. Unknown Speaker 28:01 Yeah. Yeah. Yeah, exactly. Unknown Speaker 28:08 Yeah. Me, too. Today I was talking with my my brother who did some, like it is a is part of lion AI. You know, and it was he told me it was you must be missed by. People formula I like about alignment being like almost seeing which slow down like AI research thing like this Yeah. But what you said before like that's you're in New Zealand and, and doing AI stuff. I see that like, technology is like very exponential. Unknown Speaker 29:15 And yeah. Like, and so I'm just detached like for good or for bad. I'm alone and I'm just doing this by myself. And nobody I can relate to nobody, anywhere. In my city, not even the people at the computer science department asked him if I could do it like a just a paper on my imaginary programming and then just like, What is this? It's really, it's really, there's no there's no way to even because it's new. There's no nothing to reference. Like if I was to write a paper I don't even know like ```