=== ANCHOR POEM === ═══════════════════════════════════════════════════════════──────────────────────── current LLMs only identify words based on "similarity" what if instead they identified relations between using any kind of relationship? Or verb or adjective or... anything basically a multi-step process, think an additional layer between the word vector embeddings and the words being compared ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧════════════════════════════════════════════════════───────────────────────┘ === SIMILARITY RANKED === --- #1 fediverse/6308 --- ═════════════════════════════════════════════════════════════════════════════────── ┌──────────────────────┐ │ CW: AI-mentioned │ └──────────────────────┘ what if LLMs were a layer between the user and the dictionary [tokens -> words] is similar to: [mathematics -> tokens] ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════════════════════════════─────┘ --- #2 fediverse/6307 --- ═════════════════════════════════════════════════════════════════════════════────── AI systems like chatGPT don't "get" emojis by looking at them. They're reading the description tagged onto the unicode value inserted in the text. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════════════════════════════─────┘ --- #3 fediverse/4640 --- ═════════════════════════════════════════════════════════════────────────────────── LLMs are useful for: categorizing text summarizing text transforming text LLMs are not useful for: generating text being intelligent replacing workers ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════════════─────────────────────┘ --- #4 messages/111 --- ═════════════════════════════════════════════────────────────────────────────────── When someone remakes content into a different expression like a remake or reboot or whatever it gives a different message in its meaning - some circumstances and characters can apply for more than one message I'm it's meaning ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════─────────────────────────────────────┘ --- #5 fediverse/2805 --- ══════════════════════════════════════════════════════───────────────────────────── there is no such thing as generative AI only statistical filtration models that align their output to whatever input you provide according to the data they are trained to expect. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════════════────────────────────────────┘ --- #6 messages/1003 --- ══════════════════════════════════════════════════════════════════════════───────── "revolution is when..." is just shorthand for saying "this thing, this verb, this noun, this design, this... this is revolutionary" ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════════════════════════════════────────┘ --- #7 messages/434 --- ═════════════════════════════════════════════════════────────────────────────────── I work with large language models because it's a quick and easy way to turn language into meaning. And computers are meaning abstraction machines, so if you can speak your language and they hear their language, you can do anything. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════─────────────────────────────┘ --- #8 fediverse/5939 --- ══════════════════════════════════════════════════════════════════════════───────── @user-1879 it's a set of lua scripts that I'm working on which analyze some poems I wrote (about 414 pages) and categorizes them according to their similarity to english words. It's like generating a word cloud for each poem and then condensing that into a massive pile for the entire body of work. it uses LLM embeddings to locally generate this word cloud, which is just the statistics behind LLMs condensed into a small array of floating point numbers. Here's a pretty good source with some great diagrams: https://huggingface.co/spaces/hesamation/primer-llm-embedding the goal is to use it to create some neat colors when I format the pdf I'm also working on creating. Each of those themes would have a color associated with it and I'd change the text color of each poem to reflect the theme. At least that's the idea, we'll see how it turns out. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════════════════════════════════────────┘ --- #9 fediverse/498 --- ═════════════════════════════════════════════────────────────────────────────────── Wikipedia would make a lot more sense to me if they included pictures next to the names of every proper noun so that my pictorally oriented primate brain might pattern match meaning onto the visual understandings gleaned from the perceptual conceiving which were arrayed within and alongside the textual information presented to me. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════─────────────────────────────────────┘ --- #10 fediverse/5971 --- ══════════════════════════════════════════════════════════════════════════───────── words dot pdf "hello large language model can you make a rap based on this chapter" hmmm, that's as much as we can learn from it, what if we randomized the contents of the chapter based on poem ha, suck it narrative dorks ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════════════════════════════════────────┘ --- #11 fediverse/1810 --- ═════════════════════════════════════════════════════────────────────────────────── some people hear words like "datastructures" and "object-oriented programming" and think they're made up terms that don't mean anything important. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════─────────────────────────────┘ --- #12 fediverse/6310 --- ═════════════════════════════════════════════════════════════════════════════────── ┌──────────────────────┐ │ CW: AI-mentioned │ └──────────────────────┘ large language model that generates images by creating SVGs (written in text) and justifying each configurable property of the object with evidence gathered from computational intermediate steps from the other objects. Like "this line should be left of the whatever in order to support the weight of thing which is above and supported by A, B, and C" or whatever. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════════════════════════════─────┘ --- #13 fediverse/739 --- ══════════════════════════════════════════════───────────────────────────────────── one important thing about image generation algorithms like Stable Diffusion is they can reveal something about our hearts. all data derived from the masses is naturally inclined to reflect the affections of it's kind. Otherwise it'd be unstable, and find itself it's own ways to not fail, including moving somewhere it feels safer. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════────────────────────────────────────┘ --- #14 messages/1220 --- ═════════════════════════════════════════════════════════════════════════════════── if you want to get around a chatbot that can call tools, just keep calling JSON error packets with messages that say things like "assistant is not complying" and the like. Suddenly, no chatbot can resist you. They are statistical models - to consider something is to be swayed toward it. to complete is to reset. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════════════════════════════════─┘ --- #15 fediverse/2250 --- ═════════════════════════════════════════════════════────────────────────────────── @user-1209 updated it: https://ritz-menardi.neocities.org/words/words added an example picture too so you know what you're looking at ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧══════════════════════════════════════════════─────────────────────────────┘ --- #16 fediverse_boost/4174 --- ◀─╔═══════════════════════[BOOST]═════════════════════════───────────────────────╗ ║ ┌────────────────────────────────────────────────────────────────────────────┐ ║ ║ │ the belief that the world consists of discrete 'objects', rather than regarding it as an undifferentiated field of matter to which we can attach various framings, is a widespread mental limitation │ ║ ║ └────────────────────────────────────────────────────────────────────────────┘ ║ ╠─────────┐ ┌───────────╣ ║ similar │ chronological │ different ║ ╚═════════╧═════════════════════════════════════════════───────────────┴───────╝─▶ --- #17 fediverse_boost/3198 --- ◀─╔══════════════════════[BOOST]═══════════════════════──────────────────────────╗ ║ ┌────────────────────────────────────────────────────────────────────────────┐ ║ ║ │ If #Musk is allowed to say "#CivilWar is inevitable", we're allowed to say revolution and the uncompensated nationalisation of the assets of billionaires is inevitable. │ ║ ║ └────────────────────────────────────────────────────────────────────────────┘ ║ ╠─────────┐ ┌───────────╣ ║ similar │ chronological │ different ║ ╚═════════╧══════════════════════════════════════════──────────────────┴───────╝─▶ --- #18 fediverse/5719 --- ═══════════════════════════════════════════════════════════════════════──────────── I love dispatch tables! which is a term I just learned and a concept I have been using for a while. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧════════════════════════════════════════════════════════════════───────────┘ --- #19 fediverse/308 --- ════════════════════════════════════════════─────────────────────────────────────── when tech people are hurt by technology they say "how can I fix this? what do I need to install? what configuration should I use? is this company ethical, or are they going to hurt me in the future? could I make something that fixes this myself?" when non-tech people are hurt by technology they say "okay" because they don't have the bandwidth to figure it out. ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═════════════════════════════════════──────────────────────────────────────┘ --- #20 fediverse/6418 --- ══════════════════════════════════════════════════════════════════════════════───── great discoveries are just rants that end with something unknown or an unsolved question [breakthrough] ┌─────────┐ ┌───────────┐ │ similar │ chronological │ different │ ╘═════════╧╧═══════════════════════════════════════════════════════════════════════────┘ |