Found a bunch of undelivereds 2/24 ---------- Forwarded message --------- From: Undescribed Horrific Abuse, One Victim & Survivor of Many <gmkarl@gmail.com> Date: Fri, Jan 12, 2024 at 5:20 PM Subject: Re: [spam][crazy][fiction][random] Non-Canon MCBoss Spinoffs To: mailbombbin <mailbombbin@gmail.com> idea for robo-cypherpunk-that-writes-code [whoops never mind? 1705 anyway basic draft concept: - pursue-goal-by-writing-code - maybe a small set of _high level ideas_ that it iterates through in a decision tree like early game-playing engines? leaves out: how to navigate the space of concepts [without either -- 1706 hrm :S sounds like [unresolved connect-part hrm :S 1706 ok um i guess the biggest issue is personal dissociation around an area of unfamiliarity. maybe with more familiarity with the area, more ideas would seem [graspable? findable? makeable? .] 1707 what kinds of high level ideas to write code? - access and summarize APIs ok hmm so one basic idea is [making use of existing interfaces, e.g. email similar to um descriping inside computer-code-system, e.g. decision tree, these existing interfaces. ] [so maybe idea of translating text from API specs and source code into something useful. this might be easier to think about if the "something useful" were more clear. _lots_ of dissociation around concept graphs: could we generalize? what we want is utility-information. how to make use of. so, each thing in an API spec or source code, provides information on how to use it for stuff. goals have parts, and the stuff needed to reach them has parts. what do i want, what do i have, what parts get from one hting to another thing. each part of an api spec provides something that gets to B given A. what does it need, what does it make? maybe? seems at least tiny bit helpful? ] specs are usually buggy/wrong so then we can move idea to source code, is a little bit bigger, maybe a separate area of translation or summarization maybe around -- what does it do. near, what is it for [how do i use it] [how was it made] [what were they thinking when they made it] [why is it made this way] so maybe, what does it do / how do i use it how do i use it, this can be similar to what-does-it-need,what-does-it-make [then you have, what is what it makes useful for] [and then you can index tasks by what something is useful for, and form steps to reach them !] 1710 inside code, we have, what does it do ; [ how to form parts of what does it do like needs/produces? maybe little steps? why-is-this-this-way, what-is-this. what-is-this is like part of what-does-it-do. is this a for loop? an algorithm? maybe it would be fun to explore! what would a language model think with what-is-this? 1711 1712 ok as soon as i poke at, i notice there are two parts: word-choice, and structure. traditionally when analysing code developers/cypherpunks/whatever think only of structure, as far as i understand (could be very wrong), but structure is what actually defines behavior. as opposed to variable names, function names, etc; these are very useful too and inform the production of an api spec, but don't actually define the underlying behavior. they are useful for accurate summaries tho, can be fast to use and then behavior used to check. the issue is, something like a language model is likely to use these words a lot, far more than logically appropriate. there are also other concerns when doing something like sending e-mail, system administration concerns 1714 1715 here's one of chatgpt's example codes they have it produce to show it off: window.onscroll = function() {makeSticky()}; var header = document.getElementsByClassName("header")[0]; var sticky = header.offsetTop; function makeSticky() { if (window.pageYOffset > sticky) { header.classList.add("sticky"); } else { header.classList.remove("sticky"); } } pasting made it all run into one line. maybe that's helpful, more obfuscated. noting that: - forms of code exist densely. for example, javascript has a syntax, and writing styles have a syntax. code is mostly boilerplate specific to the form it is presented in [; the comprehension of code could be maybe considered a job of processing and summarizing boilerplate away this boilerplate has a similarity in that it exists similarly in many places. variable assignments, function calls, etc. ...] so one task could be to kind of normalize a chunk of code into some representation, some normative boilerplate then one could ask: what does it need, what does it produce. maybe this is the answer to what its purpose is? to what it is for? i think it might be concievable that way. but there are a lot of things that can be needed and produced. the code maybe is too simple. but it is also obscure! it raises a concern. let's say we processed it successfully to something like "the functoin sets a class to be present if and only if a global variable "sticky" is less than window.pageYOffset" this explanation is incomplete. it depends on external information :D and then we [get a little to [informaiton propagation 1720
( >( :S :s :D :s :D :D
6/24 I'm using Inbox When Ready to protect my focus.