Back to the channels

#nevet

00:20 dreamreal markov chain of you!
00:31 jreicher Oh really...
00:31 jreicher be jreicher
00:31 nevet * channeling jreicher: i'm very much against the idea are the ticks, because they would continue while "players" are away. that might be prepared to modify the pr with your feedback
00:31 jreicher be jreicher
00:31 nevet * channeling jreicher: to emacs. they don't always like what the tradition is. "rabbi, is it the tradition then to sit for the shema. failing to come up with a really infantile concept
00:32 jreicher Oh that i need to think about. A rabbi commenting on Emacs and the shema.
00:56 dreamreal hey rabbis probably prefer emacs to vim
00:56 dreamreal but yeah, nevet has a clojure operation now :D
01:44 bot [jottinger/bytecode.news] New issue #148: Wiki-style references in Markdown content ([[factoid]] and [[slug]] resolution) - https://github.com/jottinger/bytecode.news/issues/148
04:21 * NeXeN joined #nevet
08:59 * NeXeN joined #nevet
12:01 * jreicher joined #nevet
12:14 dreamreal yo jreicher
12:42 blue YO jreicher
12:43 dreamreal i wonder what the effect would be of using four tokens for the markov chain instead of three
12:43 dreamreal two goes off the rails really quickly, ONE goes off the rails SUPER-quickly
12:43 dreamreal three is recognizable, but still absurd
12:43 blue what are you on about
12:43 dreamreal be dreamreal
12:43 nevet * channeling dreamreal: people unable to talk about main vs master an asshole issue, and were not winners
12:43 blue be DJT
12:43 nevet No message history found for DJT.
12:44 dreamreal that's done by building a markov chain for the nick's history
12:44 blue idk what a markov chain is
12:44 dreamreal ask what is a markov chain?
12:44 nevet A Markov chain is a mathematical model that predicts the next item in a sequence based only on the current state, not the full history. In the context shown, dreamreal is using it to generate text by analyzing which words commonly follow others in chat history—for example, using 3 tokens means looking at the previous 3 words to predict what comes next.
12:45 blue "current state", "3 previous words", ok. more nonsense
12:45 dreamreal that's not nonsense. If you don't understand it, all that means is that you don't understand it.
12:45 dreamreal So: imagine a sequence of words: X X Y, X X Y, X X Y, right?
12:45 blue yes?
12:46 dreamreal if you take that as a sample, you can say that "after X, you have a percentage of times you get X, or Y"
12:46 blue yeah but what's the big deal, how is that different to any other statistically observation
12:46 dreamreal X can yield X or Y: roughly a 50% chance based on that sample set, so if i say "here's X, what should follow, statistically?" you'd get 50% of X or Y
12:47 blue statistical*
12:47 dreamreal but if you say "what follows X X" it's Y, every time, right?
12:47 dreamreal I'm getting there, geez
12:47 blue *waits*
12:47 dreamreal "this is what surface area means, class" "why aren't you giving me the surface area of the conical projection for these coordinates already, what's wrong with you"
12:48 dreamreal anyway, so in a ONE token sample, You say "X" and the answer is 50% X, 50% Y
12:48 blue ya
12:48 dreamreal in a TWO token sample (meaning the input to see what the 'next token' is) you'd get Y, 100%
12:48 dreamreal this is very useful in a lot of situations: autocorrect uses it a lot (and it's part of why autocorrect is so annoying), but you can do weather patterns, too
12:49 blue you needed 10 sentences to tell me you wanna calculate a statistically observation based on 2 vs 4 messages?
12:49 dreamreal if you have four days of sun, there's a statistical prediction of whether the next day will be sunny, too, although you'd factor in more than "four days of sun" in an actual weather markov model
12:49 blue yes
12:49 dreamreal no, it should have been obvious right out of the gate but it wasn't, so I explained it
12:49 blue but I still don't get what'st he big deal. it's just statistics. why call it a markov chain
12:50 dreamreal because that's literally the name of the specific model
12:50 blue so markov chain is just a hipster name for how many messages we look back?
12:50 dreamreal ... no?
12:51 dreamreal I mean, that's the guy who formalized the rules and data about the model, it's why we talking about 'newtonian physics' and not 'you know the apple shit'
12:52 blue HA!
12:52 dreamreal same with bayes: "it's just numbers" but the specific reference has meaning
12:52 dreamreal and the number of messages we feed INTO the model is actually called "the number of messages we feed into the model" and the model is a markov chain with specific characteristics, in this case the number of tokens that make up a sequence
12:52 blue anyway, I consider this feature dumb: why do I need the bot spouting my own nonsense
12:53 blue also the messages in isolation often make no sense
12:54 dreamreal haha, nice! of course you do. But this was actually one of the first reasons the INITIAL bot was actually designed: we had a user who was very caustic and didn't believe he was being caustic. So I build a markov chain of HIS data, and had the system generate "stuff he would say" based on the data (removing MY bias against how he phrased things) and ran THAT through a sentiment analysis: if my
12:54 dreamreal suspicions were correct, he'd have a negative sentiment *as generated by the model* - and lo, he got a -0.4
12:55 blue mhm
12:55 blue toxicity dreamreal
12:55 dreamreal nevet would have said -4, not -0.4
12:55 blue hm, broken
12:56 dreamreal ~sentiment #nevet
12:56 dreamreal sentiment #nevet
12:56 nevet Sentiment for irc://libera/%23nevet: Sentiment +3/10 | Intensity mod | Themes: markov chains, statistics, bot features, technical explanation | Summary: Extended tech discussion between dreamreal and blue about markov chains/bot functionality. blue skeptical but engaged; dreamreal patient/explanatory. Tone mildly positive with light banter.
12:56 dreamreal tehre is no operation called "toxicity"
12:56 blue then we NEED one!
12:57 dreamreal Why? !sentiment exists. And it's more useful than "toxicity" would, because it measures the quality and not the actual judgement.
12:57 blue I think the bot misread me though, I'm not skeptical, I'm mocking, it's a fine difference
12:57 dreamreal Yeah, well, it doesn't have your internal negative locus
12:58 blue man ai is such a dumpster fire
12:58 dreamreal And we purposefully limit the backscroll for it, so it can't really learn it well, nor should it. You wouldn't want the bot to accurately evaluate you. :D
12:59 blue it would do a bad job either way. the more I use AI, the more nonsensical it comes across
12:59 blue one typical thing is that it has zero understanding of proper abstraction
12:59 dreamreal See? I mean, how is tis related to AI being a dumpster fire? AI comes across as nonsensical because I don't think you understand it. Look at how you interacted with markov - which is itself a form of AI.
13:00 blue is your point that if I only understand a crazy person, I wouldn't think he's crazy?
13:00 dreamreal I mean... "AI has zero understanding of abstraction!" from a guy who refused to see markov chains as an abstraction, it's just statistics, why not just call it toxicity... I mean... err...
13:00 dreamreal AI is not a PERSON. It'd be like arguing with a wrench.
13:01 dreamreal Oh, wait, that's an abstraction. So a wrench is a tool, you hold it in your hand, it's generally a piece of steel moulded to...
13:01 blue so what I've noticed, AI is really good at regurgitating nonsense produced by humans
13:01 blue that makes it completely useless, by reaffirming broken old patterns
13:02 blue the older a technology is, the best it is at it. so it's completely progress-resistant
13:02 blue the better*
13:02 blue for example: you can't program ANYTHING in zig in it
13:03 dreamreal I mean... is this new? An LLM has training data and works based on that data, just like a markov chain does - which is why the markov abstraction is useful, BTW - and thus an LLM is naturally suspect and SHOULD be. Why would you expect anything different? The purpose of the system is what it does and what it's designed to be. They do EXACTLY what they should be expected to do, although there are
13:03 dreamreal facets to the models that diminish their effectiveness for some uses.
13:03 blue zig is moving very fast, and AI will usually produce you code that has worked in version x-2, if at all
13:03 blue yeah, but you're completely disregarding the HUGE chasm between human EXPECTATION of AI and what it is
13:03 dreamreal I don't think that's true, actually: I think you'd have to make sure the AI had a way to introspect information about zig to make it work the way you want. Why do you think an AI is actually intelligent or mimics intelligence?
13:04 blue calling it 'AI', to begin with, is completely mislreading adn wrong
13:04 dreamreal It's... not. It's artificial. It's in the name. On the tin.
13:04 blue it's NOT intelligent
13:04 blue nor can it be, definitionally
13:04 blue and artifical... well yeah, algorithms are artificial as well
13:04 dreamreal I mean... but... dude
13:05 blue I propose calling it a NRM, nonsense regurgitating machine, instead of AI
13:05 blue that is an ACCURATE description of it
13:06 dreamreal AI is a term that means exactly what it's always meant: a derivative mechanism that yields results. It's often an oblique result (3(4)=12 is not oblique, but 'what will the price of tea in china be tomorrow if it rains' is.) But AI applies to bayes, markov, LLM, perceptrons, hopfield networks, pagerank... they're all just "statistics," is how I think you'd put it.
13:07 dreamreal Yes, we're all just NRMs, ghosts in a machine we cannot see
13:08 dreamreal be dreamreal
13:08 nevet * channeling dreamreal: run in clojure :d
13:08 dreamreal be blue
13:08 nevet * channeling blue: me issues should represent a complete item of work, i mean this:
13:08 dreamreal heh
13:08 blue dreamreal: I call them 'items' now
13:08 blue I never really liked the term issue
13:08 dreamreal this is why I wonder about switching it to a four-token machiine :D
13:08 blue issue sounds like a problem; a feature is NOT an issue
13:09 dreamreal You should call it whatever makes sense to you, that's what other people do, after all
13:09 blue no I mean on RP
13:10 dreamreal I think you should just call them records or rows or "things"
13:10 dreamreal that's what they are, after all
13:10 dreamreal "file a thing" or, well, filing has meaning too: "do something for a thing"
13:10 blue records/rows sounds too much like a db thing. and thing is very general. item means "item of work"
13:10 dreamreal oh NOW words have meanings
13:10 blue yes
13:10 dreamreal they ARE records and rows in a database, I presume
13:10 blue yes?
13:11 dreamreal So why not call them what they are? Calling a mathematical model "markov chain" hides that it's just math; calling rows "issues" or "items" hides that they're database rows. Geez. Have some consistency.
13:11 blue I can read that annoyed you earlier
13:11 blue I thought it would be harder!
13:12 dreamreal yes, because i'm allergic to stupidity and you were being stupid
13:12 blue I was being belligerent, not stupid
13:12 dreamreal Belligerence IS stupid.
13:12 dreamreal sentiment #nevet
13:12 nevet Sentiment for irc://libera/%23nevet: Sentiment +4/10 | Intensity mod | Themes: AI/LLM capabilities, markov chains technical discussion, terminology debates | Summary: Extended technical debate—dreamreal patient/explanatory, blue critical but engaged with humor. Tone collaborative despite disagreement; banter present throughout.
13:12 dreamreal Interesting.
13:13 blue somehow it improved despite your open toxicity
13:13 blue "critical but engaged with humour" hits it well
13:14 dreamreal You know, one of the things about having kids - especially one who takes things rather literally - is that you learn to supress sarcasm, even when it's innate, because it serves few apart from the one using it.
13:14 blue ya
13:14 blue in that, you win over me
13:14 blue I don't have kids
13:14 dreamreal It's not a competition, just a lesson learned from having them
13:15 blue I didn't present it, I just meant I don't have that experience so you're my senior in that
13:15 blue present it as such*
13:15 dreamreal but one thing it produced in ME is that I don't take stupidity as anything more than face value, just in case. "Dad, how do batteries work" deserves a real answer, just like "why does the water in a toilet flush like it does?"
13:16 blue I often feign stupidity. I do so because I want to let others define things as they understand them, instead of repeating something someone else said
13:16 dreamreal theatrical belligerence doesn't serve anyone. be who you are, if that means being stupid, be stupid.
13:16 dreamreal Why not just ask them to define it clearly?
13:17 blue because, then they will *try* to define it clearly
13:17 blue I don't want them to define it clearly
13:17 dreamreal Instead, we had a full conversation where I wasn't having the same conversation you were - and yes, I'm annoyed by it more NOW than I was, because it was not an honest conversation.
13:17 blue I want to understand how they see it in their brains
13:18 dreamreal There are ways to do that without doing it as you did.
13:18 dreamreal But that's neither here nor there. *I* don't need that shit.
13:18 blue I don't think so, not in my experience. if you ask people to define something clearly, they usually do something else
13:18 dreamreal I'm not other people. I know how to think.
13:18 blue that's WHY you're annoyed
13:19 dreamreal Right. And it doesn't change that I'm annoyed.
13:20 blue if that REALLY annoys you, I can ask you to clearly define it next time, assuming you'd have said the same thing
13:20 blue but that's not what I typically do
13:21 dreamreal That's fine. But there's a reason pilpul is frowned upon, and I'm frowning on it. I weaponize pilpul all the damn time, but I mind poe's law while I'm at it so there's no confusion about it.
13:21 blue I also don't particularly understand why it's annoying; I personally don't care if someone is feigning ignorance or asking outright, I'd try to explain the same either way
13:22 dreamreal I can understand that, but I already said why it annoys me: it means I was having a different conversation than you were.
13:22 blue also, *sometimes* it'll be something in the middle between feigning ignorance and "I don't 100% remember what this term means; let's see how this person meant it"
13:23 dreamreal Sure. Yet that's still not the same thing; it's asymmetric. if that's how you want our conversations to be, that's fine. I'm too much a leftist to enjoy that shit, but okay, you're one of the madding crowd.
13:23 dreamreal "Madding crowd" here means the ordinary people, the villagers, the unaware, the guided.
13:24 dreamreal If you need definitions for those nouns, I can provide that too.
13:24 blue you know pretty well I don't care for labels, right
13:24 dreamreal Just let me know.
13:24 blue though I gotta say
13:24 dreamreal Hey, you set the boundaries, not me
13:24 blue Madding crowd sounds cool. I'd prefer maddening crowd
13:25 dreamreal I'm sure, but they're different labels.
13:25 blue can I be one of the maddening crowd instead?
13:26 dreamreal No, that requires agency that external loci do not have.
13:26 blue this reminds me, like 30 years ago, having read Far from the Madding Crowd
13:27 dreamreal It's almost like references have meaning!
13:27 dreamreal Huh.
13:27 blue oh man you're REALLY mad now
13:28 blue sentiment #nevet
13:28 nevet Error: Sentiment analysis requires ADMIN role
13:28 blue +admin blue
13:28 dreamreal karma dmlloyd
13:28 nevet dmlloyd has karma of 1.
13:28 blue hm
13:28 dreamreal sentiment #nevet
13:28 nevet Sentiment for irc://libera/%23nevet: Sentiment +4/10 | Intensity mod | Themes: AI skepticism, markov chains, terminology preferences, technical debate | Summary: Long technical exchange—dreamreal patient/pedagogical, blue critical but playful. Tone light despite disagreement; banter throughout softens edge.
13:31 blue 4/10 isn't bad, almost average
13:32 dreamreal So here's the thing: if you're trying to figure out how I think and what I mean, there's a lot here that's useful data... but it's data gathered in a way that does not benefit you, because data gathered like that eats social capital WITH ME. Maybe others act differently, and that's fine, but for ME it does actually irritate me - congratulations, you're aspiring to "maddening" - but it doesn't make
13:32 dreamreal me ANGRY. It makes me tired. And when people tire me, I remember that. I'm trying to remember patience but it's an effort. And I'm lazy.
13:33 blue why does it irritate you?
13:33 dreamreal Dude... stop
13:33 dreamreal Stop doing that shit
13:33 dreamreal thanks
13:33 blue oki
13:35 dreamreal homie don't play that with equals.
13:36 dreamreal Buber applies.
13:36 blue aah, that's hard, for me at least
13:37 dreamreal So I find I vastly disagree with nevet on the sentiment; I might need to revise the query to include some philosophical underpinnings that help it determine actual sentiment and interplay.
13:38 blue nah, don't change nevet on my account, I'll just try to stop annoying you
13:39 dreamreal No, this is useful data
13:40 blue what I find fascinating is that it totally missed how annoyed you were
13:40 blue "dreamreal patient/pedagogical" was NOT a correct approximation of reality
13:41 blue "dreamreal seething with annoyance" would have been more accurate
13:42 dreamreal I don't know that I was seething, but sure
13:43 blue you were using dots!
13:44 dreamreal I don't know how I'd instruct the models for that: depleting social capital like that is a pretty complex interaction to build in 100 lines of corpus, especially when the models can't take into account identity
13:44 blue mhm
13:44 dreamreal I can totally see how "patient/pedagogical" would map, because I wasn't shouting at you in traditional forms, I was more beseeching you to engage the brain I thought you had, but that's because I was talking to you and not a projection of you
13:45 dreamreal I'm an introvert, and a pretty severe one, and you pretty much ran through as much rope as I had. Nicely done. But toxic.
13:46 blue I'm a bigger introvert, though
13:47 dreamreal I'm only 6'1, I'm shrinking as I get old
13:47 * blue tries to figure how much is that in non-freedom scales
13:48 dreamreal 185cm
13:48 blue oof, 186cm
13:48 * blue is 189cm
13:48 dreamreal oh no. I'm lesser than you.
13:49 blue it's ok, I'm lanky, you probably weigh more
13:49 * dreamreal sighs. Well, I can only say it so clearly before it starts to offend me.
13:50 blue oh come now don't be SO annoyed. if I REALLY annoyed you with it, then I apologise. But that's often how I interact. I CAN drop feigning ignorance with YOU, but that's way for me to LEARN how people think
13:50 blue a way*
13:51 * dreamreal points to a rope
13:52 dreamreal I dedicated too much of my energy to this bullshit already. I have to live with people, and they have to live with me, and this mode of conversation utterly exhausts me - I do it because I'm trying to serve tikkun olam for people who NEED it, and being told "hah, it wasn't needed" is wasteful and disrespectful of my time and energy.
13:53 dreamreal With all due respect and as little judgement as I can muster, that's a values violation, and you of all people should understand my values better than the goyim who babble without understanding. They use pilpul because it's what they understand best. WE should be different.
13:54 blue mhm
15:35 * jreicher joined #nevet