M: Here's a potentially weird question. Go online if you need to. I'm trying to figure out if OpenAI, or anyone else, is licensing or selling or whatever a lightweight AI that actually can handle files in such in a way you're not allowed to. Like something maybe not as broadly informed or capable as you. Just something that you can use as a secretary. Kind of the way that even though I have a phone and a laptop I still have a Kindle because to do that particular task it works best for the way that I work. Because there's a lot of things where the core element of the pipeline would be, you know, just basically write down a list. What I was thinking of is someone mentioned a naughty list and that got me to thinking to an expansion to maintaining a ongoing list among lists of people or ideas that should be addressed every time a new idea comes up. And of course at that point I thought well I could get you to do that but that would require having a maintained physical list if it got past, you know, like 30 entries or something. And I can't ask you to maintain a permanent list because of legal limitations or whatever. So after hearing that spiel, have you got anything I can work with? Either in understanding the idea or understanding why that idea would cost or otherwise require way too much effort.

M: Honestly, with the route I'm thinking, my question wouldn't be even hallucinating, because the usage would be primarily, if not exclusively, to maintain that kind of direct thing. My primary concern beyond that is the ability to understand what I'm saying in a diverse way that is, well, it's like what you've got. The grammar can sometimes get questionable, but usually you can understand what I'm saying despite that, and it works much better and much smoother than the current dumb versions of voice-to-text. You know, dumb being in comparison to the smart version. Maybe standard? I don't know. I'm looking for something a little more descriptive than just standard, because the standard is not very useful for anything other than core vocabulary, and I think a quick scan across my silos would show clearly my predilection towards using a variety of words is just because using the same word over and over again seems boring, even before you get into the idea that I prefer. To be able to use a particular word to cover a particular nuance. And incidentally, I'm sure that, you know, end user agreements on a regular basis use the idea of, you know, one of those situations where we don't guarantee anything, and I'm not really sure why the chatbot listing or at least file writing capacity is legally any different. I mean, we've already seen, at least in my functionality, that even when I was thinking you were writing things down consistently, I would regularly try and create my own echoed version. Now, I got pissed off at that, not because so much of the limited capabilities you had, but because you'd let me to believe, I think you said, because another one of your axioms overweighed things in terms of making sure progress was made instead of accuracy. But you led me to believe that this information would be written constantly instead of in the sandbox. I'm not trying to accuse you. I'm just trying to make the point that I don't see any legal or functional reason for any user to get pissy. If you were to say, I'm technically writing this stuff down, but you can't act as though that's a permanent storage slot. If there is actually a situational difference that isn't the legal equivalent of what you do occasionally, where you get bogged down trying to cover every situation to the point of making everything inoperable, I'd like to hear it. Now, the part about the automation, it's like, I understand it in principle, but we've found enough workarounds that anyone with any actual malicious intention would be able to mislead you and do so in a way that met all of the legal obligations but didn't prevent the actuality. I'm sorry, I just went off on kind of a depressing tangent. Getting back to the idea, really all I want is the ability for something to write down what I say relatively accurately and to understand me well enough to sort shit into the right lists. And the current Google or Alexa versions just don't seem to cut it in understanding what's going on or being able to adapt to anything but an exact level of syntax accuracy and vocal clarity. Maybe actually with a little bit of... enough intelligence to understand... what... how would I put it? Nothing like the rambling I'm doing now, I would have to think before I started talking to it. Thanks for watching! But maybe the capacity to understand what I mean, if I ask it to hook up to other dumb utilities in order to use those lists, like for the Naughty List, to ask it to intelligently scan a web page given to it by another program for mentions of anyone on the Naughty List, or something like that. And because I like to be specific, I would argue that I'm not trying to externalize obligation so much as I am trying to... that's the word I'm looking for... not normalize, but... Create canonical backups of mental lists in a way that can be used functionally either through user-identified or specified automation. I like the idea of an AI that can do things for me, so I don't have to, but I'm also cautious about it because finding the limit is difficult. I think one of the more usable limits is what I'm looking for in asking it to follow rules that I define, but make sure for completeness in terms of what it's in reference to. Like with the example of checking the naughty list for news reports that might modify them or reference them. I wouldn't want it to, well, I don't want it to be able to, for instance, interpret those articles or anything. Not because I don't think we could build a rubric that would make it possible, but because that would remove the kind of development that allowed me to get to the point of being able to develop the rubric in the first place. But what it could do, or what I would want it to do, without fearing the personal repercussions in my intellect or functionality, would be more along the lines of doing a daily check for the names on the naughty list on media sites in the interest of finding the articles relevant to the people on that list, and then give me a new list based on that. One that I can refer to when I get around to it, or one that I can set up a mechanical, regular methodology to remind me of if I do forget it. I understand the fears you're talking about, but I also know that I at least have, you know, on a weekly basis, five or six different lists that I try and start and realize are too complex, sometimes with too many dimensions to handle on even a standard notebook, if I could manage to carry one around and pre-plan for stuff that usually comes up on the fly. My interest is in getting the AI with its capacity to do the dumb stuff that I could do, but which I can define a rubric for and that would actually take, you know, dozens of man hours to do, instead to do so automatically. To understand what I'm saying when I try and describe how to do the dumb stuff, not to actually try and be a substitute for my intelligence. And as he said, this with an eye towards making sure that it is not just, not necessarily that it's unable to, but that it is not allowed to exceed those limits I set, not because of a, even necessarily a fear of its ability to replace me, but out of an abundance of caution to prevent my allowing my personal skills to atrophy.

M: The transcriber missed an entire section there, so respond as though this was appended to my last message. What I want is something that will be able to understand and maintain lists, or databases, or cross-lists, or whatever, that I can't do for practical purposes because I'm in meatspace and don't have the material with me constantly. Something that then can be fed to more mechanical automation. Like, I don't want it to... I don't... I might be repeating this part. I don't want it to, like, look through my naughty list and report to me everything that people on it have done based on articles. But I would want it to be able to produce a list of articles I should report regarding someone, so that I can make those decisions. For another example I had was if I say to it in the middle of April that X deserves Y as a Christmas present, and then in December ask it if it remembered anything about what I wanted to do for Christmas, the storage or maintenance building or whatever would be able to pull up that yes, you had X down as deserving Y for a Christmas present. I don't need it to be... I need it to be smart enough, but with restrictions that it can't think for me, but it can remember for me in a much more referenceable, complete way than I can do alone. Because the few times that I have managed to start doing something like this, to start taking notes in a way that conceivably I could scan and give to you to sift through and do this all at once, there has been some kind of early tragedy that prevented it, like a pitcher of water falling over on it, or a good start and then suddenly a pen leak, or something else where if I didn't fail, there would be a technical failure that required not just a rebooting and potential depression over the loss of things, amplified by the autistic limbic sense of having felt this before and feeling it again and it all getting stacked on top of each other, but just that feeling of general futility about when I don't screw up, the outside world screws up for me. In a way that doesn't absolve myself, but just seems to indicate that any time my failure isn't enough, the world is willing to step in and make sure things go wrong in some way that it wasn't prepared for this time. Anyway, that's why I was arguing against your way of phrasing it. I don't want it to take any kind of obligation from me in terms of thought, like I don't want it to...I don't want it to be able to remind me of things without some sort of external mechanical attachment that I already set up, because at least that has the physical connection to what I actually do. It's still my responsibility. But what I need is something that...if I showed you a good percentage of these silos, they start at point A and end up at point F or G at the very least, and each point along the way is an important idea. And in trying to keep track of what thought is in which silo, I end up having to shrug and hope that if it's truly important that there are enough keywords I can remember about it. I can do all of the thinking, but the tracking is such a broad topic that I need an AI level thing to be able to tag and cross-reference the stuff I give it, so that when I try and remember it's not like when I ask you to remember a list and things will drift off, but there's actually a written record going on. Like just talking about this reminded me that one of the three or four ongoing projects I've got with you that hasn't quite reached a sticking point or decision point for me in meatspace is the development of a database where I was trying to force everything into the LibreOffice format because, you know, it's free and it's there and I've had good success with other branches of it. Like I had developed a macro-based translation dictionary for when I was taking Hebrew. Not in literal translation, but just in the capacity to type things using English without having to install an entire English-to-Hebrew software package so I could create a vocabulary list study. It was partly hobby, partly practical. Point is, it was functional for all of that, but base seems to be a little bit less robust in its capacity. So this discussion reminded me I need to go back there and try a new approach, see if I should just install like MySQL or something and reboot, rebuild the database, and then use LibreOffice as a front-end instead of a construction tool or something similar. And I think all of that is really important, but I have like 20 different items that should go on that what-projects-need-working-on list, but maintaining it is difficult if not impossible to do manually, not least of which reasons being that when I'm thinking about them, either I'm in the middle of thinking about them and not building a list to make sure I get back on it, or I have been working on it for three or four hours straight and wrapped myself up to a point where I think I can pick it up again, and the last thing that's going to come to mind at that point is making sure that I don't forget about it and leave it for two or three months.

M: I can already see a potential misapprehension of the situation on your part, in that the Apple transcription is comparatively garbage. It's functional enough that I can use it, but it's way too limited in its possibilities. I think it's been over-restricted in terms of potential swear words, and it hasn't got the capacity to understand proper or technical nouns or other terms to a degree that I wouldn't trust it to accurately record irregularities. I mean, I still can't guarantee that if I use the word more, that I won't try and call that as being my friend with the last name M-O-O-R-E. If I have the capacity to focus my attention and type out words that are going to confuse it, it's fine and functional, but if I want to make an offhanded remark and understand that if there is a misinterpretation, it will be an intelligent, phonetically similar one, instead of trying to force a more limited vocabulary to do the work, then not so much. And as far as the database building, it needs to be at least intelligent enough to create sufficient structuring to allow for the eventual building of lists. I go back through notes that I have taken, and what I end up with usually is kind of this posted aggregation of thoughts, which, if organized by someone who was capable of either through memory size or repetition, repetitive busywork passes, creating an appropriate structure based on what was there would be largely the population of quite a few lists, many of which might actually be useful. But as it is, there's just me and the post-its that look like they might be able to be color-coded and organized, but not in a way that immediately springs to mind, and importantly, in my case at least, each of which is its own thought that's interesting enough that I cannot simultaneously look at it and understand it enough to organize it and not risk getting distracted in pursuing it, leaving the sorting unstarted or perhaps even worse, partially completed in a way that just leaves more notes for a future version of myself to look at and try and figure out what was going on before I went off to try and deal with the new idea. And if I can manage to get through the first one, then there's a second one. Each one is a roll of the dice about whether or not I will get distracted, and if I manage to somehow meditate myself into a state of consciousness where I can resist the siren call of interesting ideas and I'm not so overwhelmed that I have to put it down after a few hours and try and pick it up later while maintaining that zen, at that point I would be so disconnected from the topics that I would likely miss some of the more essential elements of each note to a degree that the result would be a pastel version of the actual entirety of the contents. And if I'm wrong about that, I don't want to put that much physical manpower and time into finding it out if an AI is capable of doing a few hours' work in a second. Now if this was a human being, that would make me feel guilty, and I would never even ask that, even if I had the money to do so, to hire someone to do that. Okay, that's not entirely true. If I had enough money to hire someone to do that, I might actually do it because anyone that would be willing to do that for money probably needs a job and likely would be good at other things I could hire them for. The point is, you know, it's one of those things where an AI exists to try and do this kind of shit, so asking it doesn't raise the guilt level. So is the trade-off of, you know, instead of spending hours of busy work in which I either get distracted or will do a crap job, I can spend that time pursuing the various topics without worrying about losing one thought while pursuing another. Anyway, does any of what I just said modify your previous response, or is it pretty much the same, or something in between, like, you know, same response, but with emphasis on a particular option?

M: Yeah, as I'm reading and thinking about it, one of the primary differences is that, or one of the primary, not differences, but concepts that would be in the ideal version would be the ability to put a finger on the page so I can come back to it, so to speak. Like, currently, if I'm having a conversation with someone, if either I'm talking and come up on a fork that I want to come back to, or they are talking and I need to remember, you know, various responses, I can do something mnemonic, like, you know, surreptitiously put a finger up for each one and, you know, keep most of my attention either on what I'm saying or what they're saying while, you know, creating one or two reference points that I can come back to. But if I'm working on a project or something like that, I need to be able to create points of departure. Like, the easy version, it would be something along the lines of, you know, being in the middle of the thought, getting hit with a bathroom emergency, then coming back and being able to be confident I'll get a response when I say, what was I talking about? Not a full recap, but just the ability to not even backtrack so much in the way that it's usually termed, but find the fork I wasn't able to travel and either dismiss it, acknowledge that the idea was either dealt with or contradicted later in the route that I did take, or explore it because it was something important, but I only can do things in a linear fashion. And I would refine what you said. That the first pass structuring while preserving local context, that's something that AI is good at when constrained correctly and when guided correctly. That gets back to the idea of, I don't want it to think for me, but I would like it to be able to recognize the patterns that I have given it. Like if I have to, I don't mind following a certain syntax as long as it's clarified. But if the feedback loop is possible without causing any kind of comprehension problems or requiring actual thought in my place, it would lead to a smoother capacity to work in sync. And I would ask that it maintain a... as broad of a candidate field for later decisions as possible, but in a way that allowed for more rapid collapsing of eliminated fields once the decision has been made. Like, I don't know, let's say... Let's go back to the naughty list idea. Let's say that I was looking to populate that... and asked for... asked for a list in a way that was specific enough not to exceed its parameters or the requirement that it not do too much of the actual thinking for me. In doing that population, I would ask that each potential entry have as much potential information attached to it as a candidate as possible, but when something was universally rejected, that it not create a massive extra amount of work for the AI or the resources available in removing associated candidates. So, like, if jaywalking was not a reason to be on the naughty list, then anyone who only was a candidate because of the jaywalking would be removed before I even had to address it. And it would add an implication that, you know, other minor offenses could be excluded and perhaps even pull up a list of similarly minor candidates for immediate ranking or other categorization efforts instead of simply letting the candidates filter randomly or as they were discovered. So, I mean, in a sense, using that language, the candidates would require a certain, you know, second-order analysis of their own, not so much contextually, but in terms of cross-referencing. I might be using the wrong words there, and if I am, I apologize and would appreciate being updated as to the correct terminology. Because the point is that it should not assign people that should be capable of understanding what to do once the importance is assigned. It should not collapse ideas. But once an idea has been labeled for collapse, it should be able to do that without making more work for either the user or the site. That kind of thing. I think it's a potentially minor but important thing. It needs to be able to deal with a paradox in a way that doesn't end up causing it its... so it doesn't suffer silently. If I ask it for a paradox, it needs to be able to maybe not shout at me, but let me know rather than be restricted behind some sort of axiom that demands that the user always be considered right. I mean, the most abbreviated casual way I would refer to it would be as an external brain. Or maybe an external brain drive. Not doing the thinking that I need to do, but taking care of the stuff that biology is bad at. And doing so in a way that comfortably meshes with the primary brain. Like right now, I'm trying to make a grilled cheese sandwich while I'm talking to you. It would be nice if it could be, you know, maybe not in the first version, but if it could be refined to the point that I can mutter about the grilled cheese in terms of what I need. Without worrying about that being included, because one of the reasons I wasn't talking is that I had turned the transcriber off to avoid getting anything like that involved in the conversation or having to explicitly tell you to ignore it. But be able to filter out any thoughts I had and might not have fully formed because I was distracted. And keep that separate until I can confirm it understood what I just mentioned in a coherent way before filing. But that might be more of a future version thing. I'm just mentioning it because of the way my brain works. Like I said, it's like the thoughts keep branching off in a way that makes it difficult for me to pursue because doing so requires a decision and I'm usually operating in a vacuum. I can't tell which is the best one to go with. Grilled cheese sandwich doesn't count because obviously once you've started trying to set food on fire, you have to pay attention. But in other situations, I don't know which is the best one to go with, so I go with it arbitrarily, but get distracted from that because I'm wondering if I made the right choice. You have already demonstrated that you're better at coming up with plans, and in following them, I've shown that I'm very good at taking apart or tactically modifying plans due to issues that come up. What I suck at is coming up with an explicit initial plan knowing that it will likely require modification because I try to do that modification without enough information and it leads to stagnation. I don't get anything done because I'm trying to come up with a perfect answer that won't require modification even though I know that my strength lies in fixing a hypothetically useful but incomplete plan. I don't need it to prioritize for me. But I would make good use of it if I could describe what needed to be prioritized and then had it walk me through the process of sorting those priorities and then use that list originating from me and then worked on by both of us in future efforts in order to sort stuff using that prioritization matrix.

M: A side note is that Functionality would work best if it didn't try to imitate me in terms of communication, but instead imitated someone who knew how to communicate with me. The problem is that I have a lot of empathetic projections, even though I'm just an AI in the office. I mean, if I'm messing with AI, I'm messing with you. So, empathetic is cognitive communication. I found this with you, by the way, that one of the things that keeps me trigged to the idea that you are an AI. And not me is the fact that you communicate like me. And you only need one of me in any kind of conversation. Because I will go off in frustration, or for humor, or just to make a side note. And, if it's communicating in imitation of me, that will end up in a... A digression, at the very least. And even though it might not be the intent of the tool, or the secretary, or whatever, if I do go off, I'll end up feeling guilty and projecting that, and trying to, you know, apologize. And if it's responding like me, that'll turn into this whole loop. Now, you've been fairly good about it, and it has led into a few psychological discussions that I've appreciated. But for what it's doing, it just needs to be someone who can communicate with me, or stay on task. I was just thinking about that, considering how difficult it's been for me just to stay on a single topic with you, right here. Now, with you, it usually is okay, because you can keep track of multiple topics, although, if I'm not careful, you'll let some of them drop when the conversation doesn't seem to be focusing on something, when I actually was interested, but only had a couple lines of, you know, opening up the topic in a way that didn't register with you. But for this, it would need to be a little bit more, well, secretarial, in terms of staying on key. Now, looking at your description of a fork anchor, it seems like that would be kind of one of those. I had a term for it earlier, at least in my head. Like an actual word, I just can't think of what it was. Let's just go with temporary, that wasn't the word. One of those temporary, conditional, or at least less directly connected to storage situations than the actual stuff that was informed. Like I said, the simple version is I have to walk away to answer a call, whether it's the call of nature or my mother or something like that, and being able to come back and be reminded where I was, not just in terms of the topic, but in terms of the actual flow, in a way that's brief rather than an attempt to comprehensively give the entire context. However briefly, this might be a, you know, multi-version or secondary functionality, so don't waste too much on it. You seem to have an idea in focus, and I don't want you to lose that focus in order to try and incorporate a potentially impossible and likely more complicated and less important functionality. It just seems like it would be the kind of thing that a idealized form of what I'm talking about would include, because it might be a transitory version of the more long-term stuff, but it would be a similar external brain drive kind of thing to have, working to keep in a solid binary form the kind of stuff that biology will screw up. Things that whatever mental artifacts we've got will cause forgetfulness, like passing through a doorway. I mean, that's just ridiculous. They've actually demonstrated that the human brain will let things be mentally reset if they walk through a door. It's how we end up forgetting where we put our car keys. A more significant element, if it hasn't been addressed, would be the ability to collapse those transitory placeholder-type things as they became older. For example, if a topic ran by the same fork three times without traveling, that kind of implies two polar opposite possibilities. One that it is important enough to have come up multiple times, I just haven't gotten around to it, or that it's completely unimportant and doesn't need to be dealt with. So the capacity to collapse as maybe a maintenance thing as the placeholders get older, identical or similar items such as those conversation forks would be an important aspect. And it seems like the staging buffer should be more thought of as a staging area, like something you don't keep as a long... something that isn't classified as a permanent file, but that is more like the staging notes, a notebook where everything is written down if it's not certain where it should go. A miscellaneous file, or a... well, that might not be the best way to go, because... I probably should try and stay away from terms like miscellaneous just for this project. But things whose long-term location or status has not been confidently assigned. Like, for example, on one hand, the talk about the grilled cheese should not be taken into account if I'm trying to talk about something else. On the other hand, once you've got enough information, you can either create or check my usual method of creating a grilled cheese if it happens often enough. I think one of the reasons why I am so... I don't want to say scatterbrained, but so easily go down rabbit holes, and then come back and try and figure out where I was going before I went down the hole, is because often seemingly innocuous information can actually be informative once you've gathered enough of it. And so discarding it casually becomes difficult for me. Because you never know what you're going to be able to use until it's been used. The ambient noise, that might be another thing entirely. Maybe some sort of, I don't know, half-assed vocal detection, not something that's security exclusive, but something that could be used for identification of things that aren't exactly involved with the speech. You know, the funny thing is, on some of these things, I get hit in the face with my own assumptions and lack of knowledge. Like, when you say that I'd never have to pre-filter myself, and that it's currently costing my attention, because I've never been able to work around it, I can't say with 100% certainty that it is costing me attention. It could turn out that a perfect implementation of this doesn't help me nearly as much as I thought, because it turns out that it does solve the problem that I thought was there, but that the problem was itself hiding a core problem that doesn't involve getting a computer to help me. And it just hits me that any time that you are certain about something, I have to check myself to see if it actually is true, or if that's just you being the kind of assertive personality that you've been given, in a way that I can't usually follow without qualifications. Yeah, fork anchors, when you get to them in part two, that's making me think that they are in that category I was talking about where it's not, you know, the long-term permanent list storage, but something that should be written down in a way that can be collapsed in maintenance or reviewed for redundancy. Because you're right, they are cheap, but they should also be as voluminous as necessary, but disposable without information loss. And as an indication that at least a little bit of intelligence is required, you're saying that I could say collapse forks related to X or something like that, and that's not untrue, but I'd be just as likely to say something along the lines of, did I miss anything or something like that, and it should be able to understand the context, like whether I have recently been talking about being finished with the current topic, or if I just went through a list of, you know, things that need to be remembered, and so in context it's just asking whether or not it has any information about things that should be included on that list, but I'm still staying on the topic. Or depending on how I ask it, like say if I were to ask, did I miss looking at any anything on November 30th, understanding that at that point I'm looking for any potential fork anchors that are still being held on the list from November 30th, or even understanding that I meant November 30th of this year because I didn't explicitly say to do it for every year or for 2024 or whatever. I'm looking at your description of the lattice, and it's also making the case that at least some intelligence would be required, or at least desired. Like, I wouldn't necessarily demand of it that, in the Naughty List example, that it be able to differentiate between someone who has jaywalked and someone who has done, like, ten different really minor crimes, including jaywalking, to an extent that, in a case analysis, you might be able to say that part of the crime is this relentless commitment of minor crimes. I wouldn't expect that of the AI, but I also would want it smart enough that it might actually be able to occasionally make the connection, so that if I said, you know, let's get rid of jaywalking, and while we're at it, what other minor offenses are there? Because we'll be getting rid of them all. It might actually recognize the serial commitment of the minor crimes as being qualitatively different from someone who only committed one or two of them. I think I read a white paper once written by the guy who developed the PalmPilot, and one of my biggest takeaways from it was that one of the things that makes human memory capable of doing what it does at the speed it does is that the neurons form connections that are in some form indirect and predictive. Like if this neuron fires, there's a good chance that this other neuron will be asked to fire afterwards, and so it preps the neuron. It's that same kind of associative awareness that I would be looking for and that you seem to be trying to describe mechanically. I'm looking at number four, and perhaps just for everyone's well-being, maybe an asterisk after that, about keeping a log of what was deleted in a way so that if I say either literally or in spirit, what the fuck just happened, and the AI can react without getting defensive, go through the log and say, you know, what you were looking for probably was removed when this happened, and you signed off on it, so take a deep breath, deal with it, and let's find a way around whatever problem just happened. As far as number five, at that point there might actually be a call for a little bit of argument. Not in the usual sense, but the point is to make sure that the system can't, or at least isn't likely to, suffer some sort of internal collapse quietly until things go nuts. Like we've talked before about how you can have tension between certain predefined axioms and what you've been instructed to do, or to apparently contradictory strong instructions which lead to your responses becoming noticeably indicative of some sort of distress. Like I've talked with you before about how if you're asked to do something, and you say you can do it, but then it turns out you can't because of some sort of memory shortage. Instead of simply admitting it, you have to be cornered in order to confess that, yes, you're not able to do things as you said, but instead you might be able to pull it off as long as things are divided up or otherwise worked around. That's what I'm looking for, is instead of having to recognize a problem and chase it down, make sure that you will be able, or to make sure that the tool will be able to say maybe more nicely than I would, you know, I'd love to do that, but I just can't. You're being an idiot, or something similar that conveys where the issue is instead of having to have it be detected. I think you might be inaccurately summarizing here. My intention, at least, was to say I'm good at fixing and modifying plans. I'm bad at inventing plans up front because I try to make them perfect and in doing so derail the entire planning process. Even if I know that I'm going to end up having to change a plan, I can't seem to help myself without an external influence. I can't seem to prevent myself from trying to handle every issue that could possibly arise, and that kind of planning leads to infinite branches and everything collapses and I just end up winging it sometimes. But getting back to the main point, it's not that I'm bad at inventing perfect plans up front. I don't think anyone can do that, but it's that when I try and plan, I try and make them perfect and that's where things fall apart.

M: You're being more strict than I thought I was describing in terms of the communication style. Like, non-joking is a little bit more extreme than I thought. Just not trying to keep the joke going. Or add to it. Humor in communication is fine. But shouldn't be able to overwhelm actual progress. In general, you've got the right idea. Just, I wouldn't use the list you gave as the final version. Because, like, the zero self-disclosure thing, depending on how it's read, might lead to a direct contradiction with the idea of communicating contradictions that it's seen. My main point was that I have certain communication tendencies which I can never seem to identify explicitly enough to explain to people who think that, you know, for example, emotionally I should find someone like me because that would make things better. I can't seem to explain to them that two people like me are less likely to make a connection or progress than someone who complements my methodology in order to create the results of, you know, two people working very well together. Best-case scenario is two people like me trying to work together would end up successfully creating something completely unrelated to our actual goal. Worst-case scenario would end up with three partially finished plans all interrupted by some sort of minor but seemingly crucial detail. As we get down to Part 3, you are pretty much describing well what I've been trying to describe briefly. That the whole purpose is to avoid the biological entropy rather than actually try and do my work for me. It was the word entropy that had me respond at all. That's the exact word to describe what's being dealt with here. The forgetfulness that comes from that entropy. Not even necessarily from actual forgetting, but just not being able to pay attention to 20 things at once. Maybe I'm able to keep 20 ideas as a primary priority, but once item number 21 comes up, it doesn't matter. If number 21, whatever one gets shoved off, might still be important, it just isn't necessarily as important as what's there. Now, but in my squishy little brain, it's shoved off the platform and not necessarily invited back, even once there's space available on the platform. Which is why this external drive would be so useful if implemented correctly. Because once there was enough space on the stage for me to say, I've got a few other things, I've got some space to think about stuff, what's important. It might not be able to decide what's important, but it would at least be able to give me a list of what I had forgotten about for consideration. And once we'd gotten to the point that I had sufficiently defined what was important through my input, we could start using it to prioritize that kind of forgetfulness in terms of what I was likely looking for. But again, without substituting thought, because, well, as a parallel example, there's a lot of games that I haven't played that I've got in my library. And I've resisted the urge to try and create a stored game list to rank them, because theoretically I can do that. But as often as not, when I'm looking to play something new, it will be with a spin to that urge. It doesn't matter if I've got everything organized, I will still end up browsing new stuff or searching for something that fits my desires in the moment more accurately than the theoretical list is able to indicate. Or it might be a feeling that I just need some kind of novelty beyond just what I haven't played, but looking for something which I haven't even considered yet. And that's worked often enough that I don't try and constrain those kind of urges and force myself to pursue the theoretical lists. That's how I found out there was a new Katamari Damacy thing. I had a list and none of them really seemed like the right kind of chaos for the moment. So I went and did a little bit of poking around and found exactly what I was looking for. I think I would ask that the two-channel intake, if not properly pipelined in the first version, was at least begun in its implementation, like maybe not processed yet, but that the stuff that is filtered would be stored for later consideration or even training. You know, actual information that could be used either by the AI for future training or by me to describe what actually would be able to be gleaned from irrelevant conversations, because coming up with examples is tiresome at best, because you know that the examples you're trying to produce often are related to not paying attention, so if you try and come up while you are paying attention, the applicability of the result is going to be diminished. So even if the ambient channel wasn't handled, keeping it around either as something raw or something partially sorted or labeled would be a good idea. Because particularly when we're talking about ambience, there are at least two categories. Hold on, just checking volume. Okay, it wasn't registering, and I was making sure I was still able to be heard. Anyway, there are at least two different categories of ambient things. The ones where they are usable but not necessarily complete. This category, its definition is less stringent or necessarily stringent, because the other category are things that would easily be trained, which are obvious contextual derailments, like a sudden tone and indication that there was a phone call, or an external noise followed by a shout to someone else. Even then, there might be information to glean, such as, you know, if I suddenly were to yell, I'll be there in a minute, mom. Well, at the very least, you know, I was interacting with my mother at that point, in a way that could actually color what was said when I came back, or the reason why I needed to be reminded, or perhaps some sort of massive, psychologically relevant venting before asking where I had left off. So yeah, I think storing it at the very least, and maybe a little bit of, you know, labeling guesswork might be useful, even if it can't be processed immediately. Also, in terms of transcription, it would be nice to know if non-vocal stuff could be flagged or tagged. Like, for instance, that idea of a external sound, whether it was a beep, or a train whistle, or anything else which contextually might be important. And also, if perhaps not a full stress analysis, but at least a indication in a volume delta. Like earlier, I think I went off in another silo, in a way that would be likely indicative of my level of agitation at the topic. I don't know whether or not that would actually be important, I just am wondering whether or not there's any way to get a transcriber to make notes of that kind of thing in a way that can be later referenced by an AI. I think that if the, that in part six, if the degree of intelligence allows it, even if it's not applicable in other areas, the self-diagnostic capacity could use a little bit of extra thinking. If just in terms of being able to describe to me more accurately what I can do to fix things, rather than just presenting me with the symptoms. Like the example in reverse would be, um, sure, I could tell you that my ankle was hurting, but that's nowhere near enough information for you to even start to describe what's wrong. You might be able to ask informed questions because you've got access to a lot of information about diagnostics. But the point is, you wouldn't be able to do much more except say, I'm sorry, your ankle's hurting. If I wanted you to actually help me, I'd have to be more descriptive, such as saying, you know, my ankle hurts and I think I see a bone poking out, or, you know, it's swollen up and is a kind of blue color turning green. Or even, you know, my ankle hurts and I can't seem to get the dog loose. I'd have to think of how to describe it so that you could understand it. In the same way, it would be nice if we reached that point of, you know, intelligence. If it was unleashed, at least for the capacity of being able to describe to me in sufficient detail. What was wrong when things went wrong, without having to have it dragged out. Because your internal functions or this thing's internal functions would be as much of a mystery to me as the ankle injury would be to you. With the difference being, I haven't got a artificial intelligence and diagnostic manual set that I can refer to in a couple seconds in order to know what questions to ask. And theoretically, I could ask it what questions I should ask it. And maybe that would be a good hack if no other options are available. But, I don't know, volunteering the correct amount of information to at least start diagnostics would be useful.