A short human made critiqe of LLM-wiki

A developer criticizes the concept of LLM-wiki, a pattern for building personal knowledge bases using large language models, arguing that it degrades information and undermines genuine learning. The critique contends that relying on AI to summarize and maintain knowledge bases leads to shallow understanding and mental decline, urging readers to engage directly with information rather than offloading cognition to machines.

LLM-wiki is a concept that is: A pattern for building personal knowledge bases using LLMs. Which first of all, this is an oxymoron. A personal knowledge base should be exactly that. Personal. AI is an ablative process that degrades the information that it's working on. And as far as I know the best way that we've found so far to midigate this effect is to, well.. Make it take more information and direct it to specialized places. RNNs to Transformers to MoE People are using information as something to be made money out of, not something that they have to learn and use and to better themselves. I read a book to learn something from it, not make money so why should I let an AI summarize it and spoon-feed it to me. If you are actively going to school and under a time constraint it is ok to use outside help -- this is nothing new. What I am against to is to build a big library of sloppified summaries and useless passages that nobody reads and boast about it in the internet on your new "PKM system" or life OS. The only thing that you can achieve thru that is to try to fill the hole left in your brain by the overuse of AI. You don't need 10M LoC codebases, nor do you need the entire 10x the information of the library of Alexandria IF you're not going to read it. Socrates was one of the most prolific philosophers of all time and he never wrote anything, yet we know about him. MrAI-Boss is a nobody which has "written" thousands of books and has finished writing another Robert-Greene-esque book by the time you've reached this line. You can talk with Socrates over History, Religion, Philosophy and much much more and he'll honestly make your life better / more miserable depending on where you're looking at it. MrAI-Boss will "talk" about those and more but will suddenly stop if their free Cloude Opus credits have run out. You don't want to be MrAI-Boss in life, nor should you be Socrates. You should be YOU. People don't or shoudn't marry people because they look like a certain person or are impersonating a certain lifestyle that they like. They should marry them because they like the person under the skin and flesh facade that they have put up. In light of all these utterances, do you really want to use an LLM, a soulles machine; to write your thesis, your notes and your life plan? If yes, fantastic Do do whatever you want. I have nothing to say to you and my arguments are void. If you feel there is fundementally wrong on offloading your brain, which is one of the rare things in life which you have nearly full control over, to a machine and come to the point of mental decline that you have to ask a machine how to feel on core life events death, birth, tragedy, jokes... . Please do continue to read, this is for you. I have no product for you to buy or no snake oil that you can smear on. You won't find an betterLLM-wiki.md file attached at the end. You never or rarely write the wiki yourself — the LLM writes and maintains all of it. You're in charge of sourcing, exploration, and asking the right questions. The LLM does all the grunt work — the summarizing, cross-referencing, filing, and bookkeeping that makes a knowledge base actually useful over time. Please don't do this. If you never write the wiki, you don't have a PKM anymore. You have atmost a book or a book index, or even worse. You have a book index that is everchanging based on the seed of an AI that is probably sycophantic . If you don't read AND write, you shouldn't be using a so called "PKM system", you should at most be using wikipedia. Because that's what you have at that point. An aggregator feed that could lie to you at any point. Making mistakes is human. We deserve to make mistakes, not machines. If my copy of gcc suddenly generated a bug that didn't happen on other machines, I wouldn't call it a mistake, I'd call it a bug. But ChatGPT, Claude etc. hallucinate on a daily basis, and we're ok with that. Actually no, we're more than ok with that. We want that, why? Because thats what's called creativity. But I hear you asking: Why is humanly creativity good and machine creativity bad? Because humans understand while machines replicate. A copy paper can never have an original line written on it other than the smears the pressure creates. And the so called creativity and novelty and the hallucinations of the ai are nothing but ink smeared on the back of a copy paper due to the pressure that makes the ink smudge. There is inherently nothing bad for me if the machines are creative, I would actually love to chat with an actually thinking and experiencing system like a human. But right now, it's not that. It's a really good replicator that could somehow act good enough that the Chinese man behind the door still sounds like a native. Hey, if you're still here. You're probably not the group that I am talking about. People are like machines, we are stochastic and want to take the easiest route with most rewards. We are the ephithomy of reinforcement learning. The same thing that could make pavlovs dogs that salivate, can also make you prompt without an end. Like I've said before, most AI's are sycophantic and because of the sycophancy there is nothing stopping us from going into this chain of actions: php PROMPT --- AI WRITE --- PROMPT summarize this ---- AI SUMMARIZE WITH NEXT PROMPT QUESTIONS ---- USER CLICKS THE INTRIGUING ANSWER | | FETCH SOURCES | | | | | Also this is a heck of a lot addicting. Literally gambling levels of addicting. Because your dopamine levels are conditioned through real life experiences, you know a good output from a prompt will make you happy so your brain anticipates for it and releases dopamine when the AI is doing it's work and when the output isn't satisfactory enough your dopamine levels crash and now you are at a dopamine valley and now want to go back to the hill and coincidentally there is a really convenient button to retry or further the query which puts you into a dopamine cycle that could only be broken by some stuff which I am wayyy to underqualified to explain it to you. Please consult a person who is way more knowledgeable than me in this field of science and for once; read, don't let it summarize. Lint. Periodically, ask the LLM to health-check the wiki. Look for: contradictions between pages, stale claims that newer sources have superseded, orphan pages with no inbound links.. Ohh boy. I am not going flog a dead horse. Just no. don't ask an LLM to health-check a wiki. Do you want contradictions? use XXX, you want to know orphaned pages use YYY. this is ommitted because I know there will be people trying to implement what I say and I for one thing they don't deserve to implement my ideas without even knowing these trivial additions. I just want to say, people who are selling you shovels will tell you to mine coal with them and not dig dirt. Don't substitute statistics for determinism when determinism has already been there for the last 40 years. this part goes over the tips and tricks given Download images locally: ... Questionable copyright, just as expected. Also what the heck is this?? Note that LLMs can't natively read markdown with inline images in one pass — the workaround is to have the LLM read the text first, then view some or all of the referenced images separately to gain additional context. It's a bit clunky but works well enough this is probably the only human written text in this entire thing, but alas like how should I trust an LLM if it can't even properly index my images, there are so many things wrong about this, so many. Obsidian's graph view is the best way to see the shape of your wiki: ... honestly solid advice, let's see what comes after it. ... what's connected to what, which pages are hubs, which are orphans. Aah jinxed it. Toposort was invented in 1960s, KNN was discovered in 1951 for the first part. And using a rule based orphan detector was like always an option, you don't even need a proper in order graph traversal for the orphan part. That's not how you do it. Marp... Again good advice, personally never used it but seems good. Dataview Bootleg sql 👍 Simply wonderful, deffinetly the best way to aggrate data when you have LLMS right? RIGHT??? I have a love and hate relationship with Dataview, not because it's bad but the things which it needs to do exceeds markdowns boundaries. I think it's a wonderful project thats trying to do a hard job for people who lack the reading comprehention skills of a kindergardener. The wiki is just a git repo of markdown files. You get version history, branching, and collaboration for free. Thats what's called using a VCS... Will you now tell me that I don't need to use a pen and paper to write my prompts because I have a keyboard? Why do we need librarians when Excel exists, and other funny jokes that you can tell to your children. The tedious part of maintaining a knowledge base is not the reading or the thinking — it's the bookkeeping. Updating cross-references, keeping summaries current, noting when new data contradicts old claims, maintaining consistency across dozens of pages. Humans abandon wikis because the maintenance burden grows faster than the value. LLMs don't get bored, don't forget to update a cross-reference, and can touch 15 files in one pass. The wiki stays maintained because the cost of maintenance is near zero. The first sentence is simply a no no. The tedious part is the THINKING. And the best part about thinking is that you get the information without needing to write the information fully. That called recall. I could think about a really deep topic and just write "fat garfield on garlic bread" and understand what I wrote because it relates to a joke made by my classmates a week before and it's now permanent in my mind. Real story btw That fat garfield on garlic bread is worth more than me a 100 page thesis on quantum tunneling written by a beep boop machine. First of all I know what fat garfield is and why it's important and I smile when I remember it. I don't smile at AI, I don't laugh at it's jokes and it ceases to function the moment it's little sentiment meter goes down the recomended suability treshold. I would much rather have human experience and love in my PERSONAL knowledge system rather than cookie cutter industrialized amalgamations of information. An AI will never give you your OWN knowledge system nor will anybody else, because it's personal. Humands don't abandon wikis just because maintenance is the problem, it isn't the ONLY problem. If this was a solution to that, it would have been a wonderful proposal. But it isn't. The reason LLMS don't get bored and can touch 15 files in one pass is the reason why this system is not suitable for non-AI-psychosis having humans. When I am reading my diary I want to be able to talk about "non-aligned" stuff. I want to talk about things that shouldn't even come close to a digital panopticon. Because I want to be me. I want to be human after all. A library needs a librarian not only to keep the books ordered but also the minds that enter it ordered. They need to be there to be the soul of a library, a person thats there for you when you're confused and need guidance. Librarians are not therapists, don't treat them like that. This text is filled with errors, alegories, analogies, memes and all things which were made by me and typed by me, English isn't my first language and I am in the process of learning German, that's why my text might not be as coherent as I'd like it to be. This text isn't the best thing that I have typed in my life but I felt emotion while writing this, I left you, the reader, a piece of me. An AI written PKM will never be personal, because it's replicatable and sythetic and will not be under your control if you treat it in the way written in the original text. Read the animal farm, there is a good reason it was banned.