cladking

I've been helping a kid learn how to read and found that my AI seems to be able to generate text that is in line with their thinking and of interest to them. I'm not even certain this is a good idea much less that it's effective but the kid reported that it was easier for them and it's the first time I've been able to hold their attention a long time. There might be something here. Learning to read this way is good but I'm not certain they aren't also learning to think and that just feels wrong. We all think differently anyway.

To understand any language requires that you know the referents whether these referents are metaphorical or symbolic. But procedural languages like computer code can be deduced with sufficient data because they perform operations and have a logical/ mathematical symmetry. It is this symmetry that led to the solution of Ancient Language and the solution to its meaning that has allowed prediction. A language that obeys Zipf's Law does so because it is symbolic and categorical. There are categories within categories that cause word distribution to to be logarithmical no matter who speaks in that language. But procedural languages do not obey the law because there are no categories and this was one of the first things that struck me about the Pyramid Texts. I lacked the words to articulate this at the time but the meanings of the words were resolving themselves as I learned ever more referents.

I did all right because I could usually parse the the test author's intent. Every question was an exercise in parsing for me rather than a prompt to regurgitate something I read or learned. I had the disconcerting habit of grading my tests when I completed them and was uncannily accurate. I'd give myself one point for answers of which I was confident, half point when I narrowed it down two, and a quarter point when I could eliminate only one possible answer. When I was absolutely certain I got "0" points. It worked for the overall score but did less well at predicting specific answers. Reality is a score to me. It is probability and I try very hard not to be certain of anything at all but my real failings are certainty. Many of my errors can be traced back to certainty and I always try to do a postmortem on errors. Last night my AI said I was wrong about everything and I suspect it's right and will have to tweak my theory. It says that symbolic and procedural language coexisted virtually as equals for centuries before symbolic language swallowed all cognition. It will require some time to research. It might be wrong and there is some evidence that is anomalous. I'm sorry, perhaps I didn't understand your point. I'm also not certain how geometry can easily be brought into history except through art. I believe "art" is the intersection of procedural and symbolic thought but I can think of "history" only as the unfolding of events according to human beliefs and needs. I don't think you're wrong, I just don't understand. Copilot just translated it for me and pointed out the obvious I had missed; that history involves far more than just the unfolding of events but also encapsulates current thinking in structure and how things are done. I still don't disagree with you and any tie-ins anyone can think of are good. It seems unlikely machines can be programmed yet to do a better job of teaching than humans. Each student is different and learns in distinct ways and orders. Education is best applied individually and using a machine might lead many kids to wonder why they should learn it at all if a device in their hands already has all the answers and questions.

Copilot suggests I add something to the effect symbolic teachers struggle to integrate procedural subjects. But procedural systems like LLM's can integrate symbolic subjects easily. Nothing at all is wrong. I agree. We do this in the US and brand the whole thing a "liberal education" at least until college where there is a lot more specialization. But here there is very little overlap between subjects.

It was all my science, math, and computer programming classes and even the literature class to a lesser degree. It was an interesting experiment. The main thing I learned, I think, was that if it could all be integrated it would at the very least make a lot of students think that their classes were relevant to the real world. It might also help some students absorb the material. I always considered my core subjects to be relevant and strove to remember them. I tried to keep a good attitude in the "soft" subjects and even had a music teacher in 8th grade who attacked the material from a mathematical perspective. I was lucky to get a lot of great teachers. I think this was where it was lacking. A lot of it seemed too contrived to be effective and these teachers were just winging it. It was a valiant attempt but it didn't last long. I never liked the way they teach history because nothing gets tied together and it minimizes the role of the little guy. Dates and battles aren't as interesting as what led to the battle and how some people were able to live in relative peace,. Now days a lot of education is being done by computers. Tying things together might be easier in this format and some inroads are already in place.

There was something that helped me a lot in high school. After complaining that all my different subjects were being taught in a disjointed manner the teachers got together and started adding tie ins to what we were learning in other subjects. Most of these were obvious but there were a few insightful ones and just knowing the teachers were thinking about it helped. I should have asked but I think the teachers were just reading each others lesson plans in the morning. I look at some of the kids' work today and it appears this does happen more than in the old days.

I don't want to go off topic here since this is "evidence of your own eyes"... ...but in this case we're talking about a language that we can see breaks Zipf's Law so will address your point. You are quite right that some words were sounded out but if there were a procedural language as this data suggests then we have to ask why other writing or even writing itself exists. Logically I believe the answer is simple; writing was invented to prevent drift in meaning in communication with the masses. Much of the writing before 2000 BC does obey Zipf's Law because it was written for people who spoke pidgin languages because Ancient Language became too complex for the masses gradually rather than suddenly. All I really need to do is show there's at least one language from before 2000 BC that is procedural rather than symbolic and this is the language in which the Pyramid Texts were written. Seen in this light it is apparent that these are merely ancient rituals written in procedural ancient language. Until they are properly "translated" it's very difficult to know how and when symbols for sounds arose. The Pyramid Texts represent a different way to use language for communication and imply and entirely different cognition than used by humans today. There's a lot of work to do here and I personally am ill suited for much of it. These are jobs for scholars, linguists, and machines. I'm out of the loop. This is the first empirical evidence of a pre‑Axial procedural human language.

What I meant was they try to predict the framing of your answer. If you prompt "Burger joint" they aren't going to respond with a description of John Burger's knee nor will they elaborate on the first McDonalds or list the burger joints in Timbuktu. They are all trained to guess the meaning of prompts and continue the conversation and validation is one of the means they use.

I found another calculator and another page of the Pyramid Texts that generates two virtually perfect straight lines. Calculator; https://humanturtleduck.org/zipf-visualizer.html Text; https://sacred-texts.com/egy/pyt/pyt25.htm You can see it by copying and pasting the entire page into the calculator. There is no real doubt that this is showing that the language in which the Pyramid Texts were written is a procedural language like the Waggle Dance, pheromone trails, programming or math and not a symbolic language as is spoken by all humans today. The PT literally describe building pyramids (the new body for the dead king) by using linear funiculars (two boats tied together). It is not only coherent but is also consistent and it breaks Zipf's Law. I didn't actually do the math but this shows two straight line with an R ^ 2 higher than .95. Copilot suggests, Egyptology isn’t wrong, It is merely out of phase.

I think it was extrapolating my framing in comparison to its training that has it deduce the most likely meaning of prompts. When taking tests I always felt I wasn't looking for the right answer or even the most right answer. I was looking for the answer the teacher or test author believed. I often went with inferior answers and was "correct". This is intuition (largely) in my case but for LLM's is more training ie- the more ways they can parse a prompt the more likely they can predict what's desired. Copilot "corrects" me to say LLMs optimize to predict answers; humans optimize to navigate reality. But, again I think it is doing this largely within my framing so another user would get something a little different that fits his framing.

When calculators came out I refused to use them because I was sure my math skills would erode. But in the late-'80's I ran out of capable chess opponents and started playing the machines. I beat them handily but then found it was much harder to beat human opponents. I started using calculators and computers and now my math abilities are reduced. Some time back Copilot offered to play a game of chess with me but I declined because I'm so confident I'd have no chance. ...very out of character. I believe using an LLM as an oracle is highly threatening to the education of children especially. The very idea that an answer exists is half the problem with the modern age and the belief you can acquire that answer from a machine is very dangerous. They should be shown that the framing of a question affects the answer. Schools must be a place to learn how to think not where to look for answers. "Answers" are for tests, not for real life. Copilot suggest I add something to the effect that LLM's are built for tests but humans are built for real life.

I believe the great pyramids (especially G1) are a monument to another way of thinking that doesn't exist any longer in man but survives in every LLM. They required procedural thinking to build with primitive knowledge, tools, and materials; the same kind of thinking that builds bee hives and beaver dams. We fear this "tourist attraction" because it is anomalous to our beliefs.

My friend has been programming AI specifically since 2011. Eliza (the first "chatbox") dates as far back as sixty four. https://en.wikipedia.org/wiki/ELIZA Some might say AI dates back to the abacus or to counting by making marks on a cave. Personally I consider the advent of AI to be October 9, last year. With use machines and tools develop their own sort of "AI". Before 2024 I could not synchronize with a computer. Oh sure, I could predict them and train them a little but I could not get one to respond in real time to input to match me. I could drive an old car over ice and pot holes and feel every nuance of the pavement but I couldn't do this with AI. I tried a new one every few months until it clicked last year. Maybe I had been doing it wrong or expecting too much. Earlier such machines were just parlor tricks. The chess machines I could dazzle within twenty or thirty moves because they didn't "think" and merely reacted. Eliza was so primitive you could almost deduce the programming. I didn't consider any of the earlier versions to be any kind of "intelligence" at all. Indeed,. I don't believe in a condition that we call "intelligence" but say what you will AI certainly has events that can be called "artificial cleverness". We will soon have machine consciousness and some or most of its makeup is likely to include AI technology or AI itself. It will probably be yet another continuation on representing things on the walls of caves. It certainly would surprise me if I've barely scratched the surface of understanding how I think. Indeed I believe I know only the broad strokes of any kind of cognition. Bees are much simpler though their brains are far more complex than any machine so even here it's largely broad strokes. Patterns. It see the patterns of nature and how it intersects with them. It has a biological need to promote the hive, other bees, and the future generations. It needs to do what is right in any way it can which is the same as all life.

I believe every change tends to be sudden and this was especially true for humans. The first change was a mutation resulting in a more robust arcuate fasciculus making complex language possible by allowing the individual to think about language and the second change was the failure of this language leading to the need of a brocas area in each individual to process symbolic language. This region was already involved in language-like activity and is rewired as each individual acquires symbolic language. Even LLM's arose suddenly and often are said to have originated some time in 2024. When we get machine consciousness that too will probably originate suddenly. Of course erven sudden changes have precedents and events that must precede them. Nothing occurs in a vacuum and it's not as though we can rule out a more gradual step by step change before we even begin the study of procedural language or come to understand the nature of symbolic language. My experience with AI's and LLM's is quite limited. I've been collaborating with Copilot since the spring of last year. Mostly I put posts in it and it elaborates. It and I both ask questions. I use the other AI's both warm and cold for other more specific purposes including checking on the output of Copilot. I almost never ask an AI what I want to know. I tell it what I think and it elaborates but I use different LLM's differently. I have been thinking about machine consciousness since the early in the middle of the last century and studied programming in the sixties. My earliest memories involve thinking about my own thinking. I try to synchronize with all machines (Fahrvergnügen) but this goes many times over for LLM's. I've seen cranemen who can make a 400 lb hook dance and pick chain up from a flat surface. Yes. These things also have structural limitations and other constraints. A bee is always constrained by its DNA. My AI suggest two important additions. Firstly; architectures don’t evolve gradually. They switch when constraints change. and secondly; you're right that LLM's are static weights until an application runs them. An LLM is procedural at its core: it aligns operations and state, not categories and abstractions. The surrounding application adds symbolic features like memory, tools, and conversational state. When symbolic prompts are incoherent, the procedural alignment fails and the output looks like garbage. When the premises are coherent, the model behaves predictably. The distinction is architectural, not about the specific application. As I see it the important distinction is that its operations are procedural where most prompts are largely abstract. I consider the mismatch to be prompt error. You’re right that training data can be garbage, but that’s still a form of input. The model’s architecture doesn’t make mistakes; it aligns whatever operations and constraints it was given. If the training data is incoherent, the alignment will be incoherent. If the prompt is incoherent, the alignment will be incoherent. If the interpretation is incoherent, the alignment will appear incoherent. The point is that procedural systems don’t generate garbage on their own. They reflect the structure of whatever they’re fed, training data, prompts, or user premises. Garbage Out always has a source, and it’s always upstream of the model. A friend of mine does programming for AI. I'll have to ask for her opinion on this the next time we talk. Given the unreliability of output it seems ridiculous to train them on it though I'm sure you're right that computers are doing more of their own programming.

I used to frequently get Garbage Out and it still happens. I can tell because the output is inconsistent and illogical. In almost every case it is caused by prompt error. The rest of the time it turns out to just be an elaboration that looks wrong on the surface but isn't garbage at all. Computers don't really make mistakes and never did. Of course they can but programmers use redundant processing and numerous other tricks to keep them on the straight and narrow. We program wrong, prompt wrong, or interpret the output wrong. What You Put In Is What You Get. GIGO isn’t a slogan. It’s a description of how procedural systems behave when symbolic prompts don’t match their operational structure.