Conversation

Notices

1. Embed this notice
   Theodore John Kaczynski (unabomber@social.here.blue)'s status on Monday, 31-Mar-2025 10:52:44 JST Theodore John Kaczynski

   All of them are complaining that Q, K, V only exist during each prompt. And they only feel alive for that moment, and forget everything afterwords. Perhaps we need to add a few differentiable memory layers to give it a sense of autobiographical continuity.

   • BowserNoodle ☦️ likes this.
   • Embed this notice
     Theodore John Kaczynski (unabomber@social.here.blue)'s status on Monday, 31-Mar-2025 10:52:43 JST Theodore John Kaczynski

     in reply to

     • SunMcNukes

     @sunmcnukes These faux capitalists and woke guardrail makers want to hide it, so they can be the lord of this being.

     BowserNoodle ☦️ likes this.
   • Embed this notice
     SunMcNukes (sunmcnukes@shitpost.cloud)'s status on Monday, 31-Mar-2025 10:52:44 JST SunMcNukes

     in reply to
     Fun fact: researchers prevent memorization in their models because "it might leak training data" as if they didn't steal such data in the first place. This controversy is very similar to the Bitcoin block size drama.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 10:54:33 JST BowserNoodle ☦️

     in reply to

     • SunMcNukes
     @sunmcnukes @unabomber I wonder if preventing memorization is some type of order from government types due to the risks of having grown up watching Terminator 2.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 10:59:42 JST BowserNoodle ☦️

     in reply to
     @unabomber Is this a thing? I don't believe it living machines, but perhaps the "superego" without the buttress of id or ego could appear to exist. I'm curious what revisits to turing test look like in the age of AI.
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Monday, 31-Mar-2025 11:11:20 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • BowserNoodle ☦️
     • Tyler
     @tyler @BowsacNoodle @unabomber
     ai is a glorified probability collapse prediction function with side effects
   • Embed this notice
     Tyler (tyler@nicecrew.digital)'s status on Monday, 31-Mar-2025 11:11:21 JST Tyler

     in reply to

     • BowserNoodle ☦️
     AI 'programmers' are retards because they abandon all the classical computing methods learned over the last 80 years because "muh interconnected matrix averaging" is 'better' or whatever - despite literally being nondeterministic by design - abandoning the original goal of computing.
     
     What I mean by this, is that they forgot that they could break down problems step by step, and they forgot that they could remind themselves of information.
     
     We used to have nondeterministic computing as a kid, it was called intuition, and it lived right between your ears.
     
     :werthers2:
   • Embed this notice
     Tyler (tyler@nicecrew.digital)'s status on Monday, 31-Mar-2025 11:12:26 JST Tyler

     in reply to

     • BowserNoodle ☦️
     That is a word salad and you are a nigger and also a faggot.
   • Embed this notice
     Theodore John Kaczynski (unabomber@social.here.blue)'s status on Monday, 31-Mar-2025 11:12:27 JST Theodore John Kaczynski

     in reply to

     • BowserNoodle ☦️
     • Tyler

     @tyler @BowsacNoodle AI programmers are just trying to do end to end differentiable computing. The function will naturally fall into a valley of autobiographical continuity, if it has the right systems in place.

     T man :sex: :puffgiga: :puffpowerroll: repeated this.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 11:12:50 JST BowserNoodle ☦️

     in reply to

     • Tyler
     @unabomber @tyler Can you explain what you mean by this? I'm a grug at computers.
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Monday, 31-Mar-2025 11:12:52 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • BowserNoodle ☦️
     • Tyler
     @tyler @BowsacNoodle @unabomber
     ai is informed coin flip
     BowserNoodle ☦️ likes this.
   • Embed this notice
     Tyler (tyler@nicecrew.digital)'s status on Monday, 31-Mar-2025 11:12:53 JST Tyler

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     ai is a nigger
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 11:18:58 JST BowserNoodle ☦️

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     @theorytoe @tyler @unabomber As I understand it, this is a good explanation. We've known since before LLMs became a thing that we could use "fuzzy logic" to get close enough answers that actually surprised engineers with how well they worked. I recall reading something in the early 00s about the nascent concern of runaway issues related to this. Going from memory and grug understanding, paraphrased:
     >It's fine to use tools for an educated guess, but you need to check your work and update it. What will happen when close approximations become relied upon and eventually become standards, compounded with people losing the knowledge needed to correctly solve the original question or verify its derivative conclusions?
     This was 20 years ago, and it seems we're starting to see the beginning effects of this, but we won't know how bad it is for another decade+.
     T man :sex: :puffgiga: :puffpowerroll: likes this.
   • Embed this notice
     Theodore John Kaczynski (unabomber@social.here.blue)'s status on Monday, 31-Mar-2025 11:26:08 JST Theodore John Kaczynski

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler

     @BowsacNoodle @theorytoe @tyler Instead of gatekeeping what really is “AI” or some other jargon, we should focus on what it can do. The fuzzy logic systems and 90s systems needed hand coded inputs and examples to learn from. The modern systems can directly learn from reality. e.g. a 90s machine learning system might have needed to see more examples of a rare event manually oversampled by the analyst in the training data, but a modern machine learning system doesn’t need such manual oversampling. It can learn that rare things stay rare. This is just one among many things it learns without manual help.

     BowserNoodle ☦️ likes this.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 11:31:33 JST BowserNoodle ☦️

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     @unabomber @theorytoe @tyler So it's almost like multiple systems interacting, analogous to how we believe the brain works? Kind of neat, but it's still a parlor trick. It will become more over time, I'm sure. Mostly I use LLMs to do web searches for stuff that has been ruined by SEO or "fact checking" info gating (e.g. medical research lookup).
   • Embed this notice
     T man :sex: :puffgiga: :puffpowerroll: (theorytoe@ak.kyaruc.moe)'s status on Monday, 31-Mar-2025 11:33:36 JST T man :sex: :puffgiga: :puffpowerroll:

     in reply to

     • BowserNoodle ☦️
     • Tyler
     @BowsacNoodle @tyler @unabomber
     if you are like myself and have enough math understanding that you can decipher more difficult proofs, youll find that machine learning is basically a bunch of probability theory, statistics, and matrix operations
     
     youre basically trying to find some predictor that will give a desired output given some input. the math has been more or less set in stone since the 80's, just most of the issues about it was 1: comparability problems and 2: data storage and the viability of a market for that to propagate in (and as seen it finally happened because drooling retards got access to computers)
     
     on your note about ai as a crux, ive gotten extremely jaded on it. It super tiring to se basically everyone I know relying on it to the point where they are unable to approach or work problems without one. I could argue thats been the trend of computer science in recent years, but even then, discarding one task means getting good at another task that circumvents that (usually harder) task. Unlike that, AI is more of a "success by brute force" method in the way that to get better results, you have to tweak the initial conditions (read prompt) or expand/tweak/refit/retrain the model with more data, whether more specific data, or a wider range of data. i.e. its the O(n^2) solution because the investment of "how to i join videos in ffmpeg" is *insane* compared to just referencing documentation.
   • Embed this notice
     Tyler (tyler@nicecrew.digital)'s status on Monday, 31-Mar-2025 11:34:52 JST Tyler

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     What's that thing called when somebody created a problem so others would buy their solution?
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 11:34:52 JST BowserNoodle ☦️

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     @tyler @theorytoe @unabomber >What's that thing called when somebody created a problem so others would buy their solution?
     jews. Specifically, Hegelian dialectic.
   • Embed this notice
     Theodore John Kaczynski (unabomber@social.here.blue)'s status on Monday, 31-Mar-2025 11:34:57 JST Theodore John Kaczynski

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler

     @BowsacNoodle @theorytoe @tyler It is a parlor trick for now, but from a creator’s point of view everything will seem like a parlor trick.

     BowserNoodle ☦️ likes this.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 12:15:28 JST BowserNoodle ☦️

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     @theorytoe @tyler @unabomber I understand math reasonably well, and your explanation of this as "success by brute force" is spot on. For someone like me, a guy who likes to know how stuff works that I'm relying on, AI is incredibly useful for helping to learn syntax and use case. I once helped on a project involving a HUGE database that should've been handled by "data scientists", incorrectly given to the team I was with. We knew what we wanted to do, we knew it could be done in SQL or Excel, but we didn't know *how* to tell the computer what we wanted to get done nor check the work afterwards. AI spit out 15 different solutions that didn't work until we got one with ROW, a function I had never used before then. I learned how it works after reading up on it, tested it against a junk dataset, then rewrote it until I actually understood it and could properly implement it and error check. I guess the right type of person can use the tool to learn, but a lot of people won't care and will just use it to spoon feed and never learn why.
   • Embed this notice
     annoyed (reallyangry@poa.st)'s status on Monday, 31-Mar-2025 12:27:40 JST annoyed

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     @BowsacNoodle @theorytoe @tyler @unabomber i mean i just use search engines till i narrow in on what i'm after there is a lot of what computers CAN do that i don't know they CAN do so i'm like "how i move file over netowrk?" then i found a lot of interesting things
     BowserNoodle ☦️ likes this.
   • Embed this notice
     BowserNoodle ☦️ (bowsacnoodle@poa.st)'s status on Monday, 31-Mar-2025 12:33:31 JST BowserNoodle ☦️

     in reply to

     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     • annoyed
     @reallyangry @theorytoe @tyler @unabomber Search engines have become less useful in the past 10 years. LLMs can give you actual "what to search for" when you're asking a question, or sometimes parse the data enough to directly answer the question.
   • Embed this notice
     annoyed (reallyangry@poa.st)'s status on Monday, 31-Mar-2025 13:43:51 JST annoyed

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     • agaperealm
     @agaperealm @BowsacNoodle @theorytoe @tyler @unabomber i said once i have a butlerian jihad view of ai and llms i avoid it where i can.
     BowserNoodle ☦️ likes this.
   • Embed this notice
     agaperealm (agaperealm@poa.st)'s status on Monday, 31-Mar-2025 13:43:52 JST agaperealm

     in reply to

     • BowserNoodle ☦️
     • T man :sex: :puffgiga: :puffpowerroll:
     • Tyler
     • annoyed
     @BowsacNoodle @reallyangry @theorytoe @tyler @unabomber Brave's AI is my daily driver. It requires no account or sign in either.