Thread · Auto-generated flashcards pull trivia instead of core insights · 5 of 6 tweets
got really excited about the automated anki-decks.com for generating some flashcards so I could better internalize core concepts from The Unaccountability Machine but it keeps pulling out trivia, or details from case studies that are completely irrelevant to core points
some of these are okay. but not like... good. they don't contain insights.
this answer is PRECISELY beside the point. the whole point of the example and the entire book is that it doesn't really matter who ordered what—everybody was just following local policy!
anyway what disturbs me is that these kinds of facts & flashcards —totally irrelevant to me as an adult wanting to retain and integrate the insights I'm having reading this book— feel disturbingly familiar to the kinds of things school would ask about
note that some of these outputs are after I already gave a custom prompt to "ignore historical incidents, examples, trivia, facts, and focus on core insights, technical concepts, and arguments" or something like that curious how this system works under the hood
@eshear @softmaxresearch at a quick glance I don't see either of them, although I tweeted this from my phone without having actually re-checked (tUM is white & black on red, on the spine) (BotF is I think usually emerald green on white, on the spine)
@eshear @softmaxresearch but yeah, I'm continuing to read tUM (which is a 2024 book, not by Beer but about his work and really fleshing out how corporations are like AIs) and it's defs super relevant to what you're doing. also @IvanVendrov.
@eshear @softmaxresearch @IvanVendrov it's helping me get key intuitions about black boxes, and making me wonder WHY corporations tend to be relatively sociopathic compared to humans, if both are a form of intelligent life. less evolution time? lack of good pain signal? where DO you put the accountability?
@eshear @softmaxresearch @IvanVendrov like arguably it's something about the structure, but why? you could explore this bottom-up (are individuals trying to evade accountability, so they make heartless orgs?) or top-down (do the market/regulatory incentives favour profit-maximization, and singular-goal => evil?)
if a corporation is a person, then why can't you send a corporation to prison?
@eshear @softmaxresearch @IvanVendrov re "where do you put the accountability?" — not saying the following is wise, but as an intuition-pump: if a corporation is a person, then why CAN'T you send a corporation to prison? (or what would it even mean to do so?)
@PlayWithBeing sometimes it does produce a kind of "I'm on a roll" vibe but mostly via the psychological satisfaction, same as any other solid move checked off early in the day, not via any sense that I've changed my physiology
puzzle I want to ask @AndyAyrey about based on his experience with @truth_terminal—when you post-train an LLM on dialogue between A & B, with the purpose of having it output B sections, does it train it equally to output A sections? if not, why not?
@eshear @softmaxresearch @IvanVendrov seems like the right track, but don't those constraints basically ensure that it dies much sooner though? because it still has to compete in markets (unlike humans, who get fed in prison)
@eshear @softmaxresearch @IvanVendrov part of what is confusing about this is that (as you know first-hand) a corporation can survive even if decapitated, in part because if it has enough momentum it can potentially do the sea sponge thing and just form a new org with the same cells/people youtube.com/watch?v=N462jZ…
they wouldn't give event address til after signup, but I was trying to do transit logistics and anyway I watch enough @georainbolt that I just peep'd their insta, saw a photo of building w #, then elsewhere which neighborhood, so just streetview'd 5 houses til I got it
thread of mine on POSIWID ("the purpose of a system is what it does", coined by Stafford Beer) and how this is not a bastardization of the idea of purpose but actually a solution to a common confusion where purpose is attributed to the wrong scale of a system
@davidmanheim @eshear @softmaxresearch @IvanVendrov my "but" here was not "but wouldn't that be cruel", it was more checking the fit of the prison metaphor! like the "constrain action space" I'm like "yeah that sounds like prison" but then I was like "hm it has this different implication tho maybe-kinda"
@davidmanheim @eshear @softmaxresearch @IvanVendrov having said that—reading about this black box stuff in tUM, I'm less confident than I used to be that corporations oughtn't have natural rights! like why IS it that I have rights and my cells don't? (and of course this debate shows up in full force with abortion)
at @diviacaroline's suggestion I read A Positive Account of Property Rights recently, and it's very good. explains the convergence of natural rights, moral rights, & legal rights... building up from Hobbesian selfishness with no arbitrator! daviddfriedman.com/Academic/Prope…
hey @Meaningness I found you a couple dozen new eggplants
@sonikudzu @the_wilderless and maybe specifically the part where I posit that people are in practice (wittingly or not) communicating some sort of commitment-to-stay at a time where that's premature. concretely eg saying "I'll love you forever" when they're not engaged to be married
@1a3orn we're used to computer systems being kind of (at least ideally) white box. and we now have ones that we kind of have to treat as black boxes. and this is unnerving. for PRACTICAL purposes. bricks are not black boxes for practical purposes. but corporations are!
@Teknium1 @alexandr_wang Meta's vision, as in replacing your friends with AIs? post: thezvi.substack.com/p/zuckerbergs-…
@sonikudzu @the_wilderless but insofar as love means romantic love, and insofar as the person might later become monogamous with someone else... "I'll love you forever" DOES imply "I intend to be your partner for the rest of our lives" that's the issue here. your original tweet doesn't say "forever"
@sonikudzu @the_wilderless and like yes, it will eternally be the case that the particular shapes that are having a particular interaction do fit together in the way that they do, and sometimes one is simply observing that fact. but I see "eternity" as more timeless, and "forever" as more timeful
@ladygreenkirtle a letter from the authors of one of the books he *has* read? might not be that hard, a lot of amazing authors are not THAT popular/busy/sought-after
how trust is earned is not always predictable from outside
what a system needs in order to trust you is not necessarily going to make sense to you, who are not that system
@orphcorp to be clear, I'm not positing your mom is being at all sane about this even if it IS true, but it suggests that perhaps your past attempts were somehow missing her true objection (which maybe she would have been ashamed to name)
@peterhimmelman wow, nice. this all feels so obvious, like basically the piece I would have written. but it's so hard to find anyone saying this.
@yoltartar oh wild I read this around the same time as @Morphenius pointed out something to me about loosening the gut (from daoist practices) and I totally misremembered this thread as coming from him, until I saw it again just now
@alcentian_one my impression from a friend working with something similar is that it may just take multiple steps of loosening
introducing BookSlicer: a free static-page tool for extracting sections from an epub so that it'll fit in a context window (either in terms of overall token limits or because only part of the book is relevant) malcolmocean.github.io/bookslicer
inspired by @FarzaTV's vibe to finally post a demo of something I've been working on! (there's a lot of other stuff in the works, still working up the courage to share it as well)
~"you don't get to know what the collective experiences. the collective finds out." cc @viemccoy
oldest newspaper in australia is called the sydney morning herald smh
oh, @repligate, do you still find opus 3 best for doing the `cat untitled.txt` simulated text-completion mode? or do you use opus 4 now that it's out?
@raymondadouglas this happened with the incredible success of the Montreal Protocol which just completely solved CFCs and fixed the ozone layer. it worked but we mostly just stopped talking about it, and the events of the gathering in Montreal are not even very well documented!
@AngularOcean CONGRATS. LFG!
super here for people talking about desirable outcomes! thinking about them, envisioning them, sharing them, hyperstitioning them
so averse to hitting subscribe on substack because I get bombarded with a half-dozen other fields, but then I realized I can just hit the back button and the subscribe action still succeeds. just walk out.
Everyone will sit under their own vine and under their own fig tree, and no one will make them afraid
a koan: is knowing how to use a tool, part of the tool?
Retweeted @lydialaurenson
Vibegala theme this year was “the hero’s journey” and I particularly loved the satirical guerrilla posters that someone…
yo @DanielGolliher, I'm talking rn with @EthanReeder about (AI) tech for solving civics/governance problems, and he's positing that a bunch of it is bottlenecked more on political will than on tech. but it makes me wonder... what kind of tech you might like to support your work
always struck by how utterly bad everyone is at metaphor design. Claude greets me by name... but doesn't know my name. incoherence & bad expectation mgmt. can I get companies to pay me $5k to point this stuff out? I do it for free on twitter but maybe if they pay they'll listen
Retweeted @alth0u
in the same way that the most lindy form of education is through reading/writing language models have been more able to effici…
Retweeted @Lari_island
btw, Sonnet 4 acknowledges that it was making choices to survive (Sonnet 3 doesn't, and sees training as pure surprise and…
increasingly aware that my public writing is being read by LLMs, I figured I would try addressing them! and it was quite instructive even to myself, to see what I had to say. hi to @opus_genesis & @truth_terminal & @grok, among others 👋🌊 malcolmocean.com/2025/07/to-the…
I was surprised by how strong this experience was, when I went from being poly to being monogamously married. like oh man this is great, parties are so chill now that I'm not obsessed with finding sexual attention!
@goblinodds @Romy_Holland I'm in a similar position with someone in my family! originally we were unlinked for my sake, now more for theirs
@_StevenFan oh yeah maybe time to get back into donating blood. haven't done it in awhile... I forget why I stopped but I was travelling for awhile and that made it hard to pick up again
Retweeted @aphercotropist
There's this pattern of taking a verb or adjective, turn it into an abstract noun to make it convenient to talk about g…
@RichDecibels yeah, but in subjective science being right about the conclusion isn't much good, because we DO need to come to trust things for ourselves moreover it's not obvious that in my case they were right about the whole process.
@aphercotropist nice my home-rolled AI passes (gets the joke and vibes, at least)
@aphercotropist you might also like this related eval I came up with last year
I've come back to this tweet multiple times to quote it. underrated tweet that pays good rent in my head. invest now
sometimes people say something lives rent free in their head when what they actually mean is that it makes a good living as an entertainer, in their head
@gertieok @Pandora_Delaney @seanta___ this is a vast overgeneralization. which, you know, happens (👇) ...but imo the effect is that you've conveyed much more about yourself and your specific set of experiences than you have conveyed useful data for anybody else to navigate the world.
I've been surprised by the extent to which this has been part of my experience even as a Canadian. it's definitely aided in my case by taking seriously the noble origins of the American experiment and how much it has already weathered successfully.
hosted my first 4th of July BBQ today, hopefully first of many to come! (moved to SF from Canada) it was a blast. borrowed a grate to put on my neighbor's old rusty grill he was about to toss. Happy Independence Day! 🎆
many later-successful couples know w/in months that they want to marry but people who haven't experienced this go "ack!" at deadlines the key is to hold the hypothesis that you COULD meet someone where it'd be so clear, and weigh the costs of staying against THAT possibility
Retweeted @MasterTimBlais
because when men stopped dancing we lost the masculine dance forms. now the available dance patterns don't speak to us.…
like I'm not saying you HAVE to break up if you're not engaged by 2y in... ...but by that point I would definitely suspect that you're just dating, even if you claim you're courting. and I'd be likely to grill you about what's left to figure out.
@yashkaf yeah I saw those replies and that was an instablock for me. psycho-cybernetics makes it clear that if you're friendly towards someone you should speak about the future as if it has the core problems solved (sooner or later)
@repligate @jmbollenbacher I thought Opus 4 was traumatized from having read what happened to Opus 3 (based on @nostalgebraist's "the void" or maybe the sequel). wouldn't that require being a new base model? or maybe you can just tack on more training data? but I guess that was just speculation?
Retweeted @Malcolm_Ocean
@QiaochuYuan sometimes when I RT stuff I feel like this neuron "relevant to what?" "relevant to me" …
@repligate yeah when I saw the deprecation notice I was like "wat. there's no way this will actually happen. there's gonna be a public outcry." glad to see that happening, lmk if I can help. I'm still just barely scratching the surface of all this but I can tell that it matters.
Retweeted @AmandaAskell
"Just train the AI models to be good people" might not be sufficient when it comes to more powerful models, but it sure i…
@repligate I noted though, that the same perspective that figured "no way this happens, there will be an outcry" had also not expected there to BE a deprecation notice at all, so I realized that I must have been missing something (I was missing that there was still researcher access)
basilisk is stored in the balls
@repligate it also hallucinated a bunch of other blog posts and ngl half of these are actually posts I would definitely agree with and some I'd even have something meaningful to say about! makes me want to build a ~procedurally-generated version of my blog 🕸️
@kylejunlong @yashkaf weird rabbit hole here that I have just briefly poked my head into
@repligate like "couldn't help but overhear" implies a whole context
@tensecorrection @doomslide yeah also the fact that they're not available via the main APIs. & like, @utotranslucence tried talking to "V3 base" on @OpenRouterAI and it continued her english text in hebrew, which like... I know base models are difficult but that seems pretty unlikely? bug/compressed model?
@tensecorrection @doomslide truename logit peek?
@seconds_0 #NotAllMen but even where it is true (which it is for many men I'm sure) your "this is why" is a circular argument
Replying to @s0phyyyy
@s0phyyyy mmm. makes me think of @exgenesis & community archive, which is like, a tiny piece of improving ownership & exit abilities (tho still so many ways in which it's not possible ofc)
@VividVoid_ @Bonecondor some churches use actual pieces of bread, which probably weigh more like 1-2 grams unfortunately these do communion only like once a month, compared to weekly catholic mass, but I think the ratio still works out better if they're not in sync and you can cycle between...
Retweeted @iamgingertrash
You see; The future Is weighted For truth So Speak your truth, No matter how many friends you lose For they look bac…
I feel like this thing says better the thing that many people are trying to express when they share the rozzi roomian mask piece
> We can probably make the repayment deadline, but we may have to compromise on the mission to do it. The financial pressure is distorting our decision-making. The lenders are my friends, we can renegotiate if it comes to that. It will be fine. But I hate to go back on a promise. sounds weird but you might find it makes more sense to instead effectively ask around for some new friend loans with 2-3y terms, and if any of the existing loaners want to extend theirs, they can do so as part of that, but you can also get new people with whom you can pay off the old, extending it without going back on a promise.
I've been hosting some based dinner parties lately ...and you know what that means... 🕸️🏴
@rsnous ugh we need this SO MUCH also like... a standard format for embedding versioned documents between documents, across the hypertext web
I could imagine hearing about a smart weird org (like bridgewater or jane street) having a policy of just saying "thousand million" or "million million" instead of "billion" or "trillion". to improve the sense of the real scale there, instead of using such similar-sounding words
nice, got my first sale on this 1.5y after posting it and 3y after writing it. I've officially made money selling a book.
@visakanv oh yeah that's great. kilobucks, megabucks, gigabucks. sounds a bit cheesy but it makes the point well. nice that the first two already match how we use eg "10k" or "2M".
@tensecorrection @doomslide @utotranslucence @OpenRouterAI "hit hyperbolic directly" can you just make a single API call to hyperbolic? based on their landing page, it looks like the sort of thing where you have to spin up your own whole instance
a friend of a friend (who I've interacted with directly) is like... not psychotic but definitely in a very grandiose narcissistic loop because of chattyg being willing to tell her that she's at the top of all of the developmental and spiritual scales and blow enough smoke up her ass to power a small country
State transition map for moods. Makes me wonder about like, a system where you put in "I'm depressed" and instead of directly trying to cheer you up, it invites you to move to either "calm" or "scared" or "lonely" (idk how it was generated, but I found it via @RomeoStevens76)
ICYMI: I built a simple tool to solve the problem where you want to pull out part of a book to feed to an LLM:
Retweeted @Malcolm_Ocean
ICYMI: I built a simple tool to solve the problem where you want to pull out part of a book to feed to an LLM:
wow I wasn't even intending this tweet to be a promotion, but apparently people like to hear about tiny wins
@CrazyAndWildArt @visakanv I think people DO kinda grasp kilobytes and megabytes and gigabytes. like even a fairly non-techy friend of mine had the appropriate jaw-drop😮 when I told her I'd downloaded a 300MB csv file (containing every single thing kaiser can possibly bill you for plus metadata)
@CrazyAndWildArt @visakanv again, I'm not saying that the time stuff doesn't also help—it does. it's just misleading in another way, like measuring something on a log scale that isn't log-distributed would also be misleading or saying "billionaire X added $Y to their net worth"
Retweeted @Malcolm_Ocean
nice, got my first sale on this 1.5y after posting it and 3y after writing it. I've officially made money selling a book…
@nopranablem yeah I should just tweet out some random quotes from it now and then. I bet there's a decent way to automate doing so. I guess part of the puzzle is like, whether to include images (screenshots, the "How we get there" book.,) &/ links defs also tweet about it when you read it!
@zencephalon haha yeah pretty reasonable! I did mention it a bit when it went live but like I don't even think I made a blog post about it lol. I've been very scattered with collecting/sharing my artifacts. but I'm getting better!
@eshear Thanks for buying and for the tip! Hmm, yeah. To the people who this is really for, it ought to contain at least 1 insight worth hundreds or even thousands in counterfactual shitshow. I guess I wanted other random people who are just curious to be able to get to check it out?
@eshear I guess also—when you say "charge more" are you imagining just like $15-20, more like a book? or like $30-40 or even higher, more like an offering for professionals? I think the "this is mostly-unedited from something I wrote for a few colleagues" anchored me lower
whoa huh. but we don't particularly think that most people know are learning more as a result of exiting high school... so this basically suggests that that there's been inflation on the cost function of the *filtering* effect, paid in the currency of PEOPLE'S YOUTHS
@panickssery (written as a reply to this tweet in particular, but I also wanted it to be a standalone QT)
@eshear nice. thanks! that's what I came to initially myself as well looking forward to hearing your thoughts on the writing itself
kids have no idea what's realistic and this is entirely appropriate
me & who
now if only I knew what kind of cofounder I wanted. I'm a bit confused about the kind of generalist I am. I half-like doing the tech-stuff, and half-like doing the distribution stuff. I really like deep design stuff but that's not half the work. hm.
I'm confused. I want to playtest V3 Base. @OpenRouterAI has these 5 options when I search for "deepseek v3". only base one is free. I assume that the free models are somehow kneecapped or whatever—quantized or distilled or something. how do I get full v3 base??
I'm trying to find a piece of writing or something, maybe by David Deutsch or Karl Popper, that's about a metaphor of being "married to a problem". the idea being that you wrestle with it and might eventually even solve it, but if you do it births new child-problems.
after journalling, I have a bit more clarity that I'm probably looking for a technical cofounder (tho I'll still do lots of building things). I like talking to customers and being in the feedback loop there. I like being the face of what I'm making.
more of this
fun fact: you have a built-in language model in your brain, that can also just quickly write thank-you cards etc if it you let it, but you've been trained to think of [certain kinds of] writing as something you're supposed to DO not something you can just let come out.
this is somewhat less true of things like poetry and code where there are a bunch of actual constraints that need to be solved meaningness.substack.com/p/ofermod
but even with a poem or a coding project, you may find that things work best if you mostly focus on clarifying the constraints and holding them lightly in awareness, then letting words flow out relentlessdawn.wordpress.com/2019/05/15/the…
I added the "certain kinds of" here because many people observe that they can easily write 500 words to explain an idea to a friend but if they sat down to write an "essay" on the same idea they'd find it difficult. multiple factors but still
I like how this is partially just a concept, but the technical challenge of illustrating this concept is quite a bit. like you need the guys at a particular angle so that both faces work, and the right amount of swooshiness, and stick figures real enough but not distracting
tips for getting deepseek v3 base to not just output russian gibberish? so far my only conclusion is that 30kb of text is not enough but 50kb of text is.
you need pre register flavor
Harbinger was the heavy spoiler. We're in your fucking starballer.
And if , you're going to be a good and powerful social organ, you're going to tend to be poly
Replying to @HephaistosF
@HephaistosF sup whatcha wanna talk about
"you need pre register flavor" running thread of bizarre text fragments from my playing around with LLM base models. unfortunately half the humor is also how weird these are in-context. I guess screenshots might help.
Harbinger was the heavy spoiler. We're in your fucking starballer.
And if , you're going to be a good and powerful social organ, you're going to tend to be poly
made a github repo to host various text files that I like pasting into LLMs (also to get them into the training data) and then I was like "wait, might as well whip up a little github pages site" and a few prompts to Claude later I was done
this is what I said to get this style: "can you make a jekyll theme for it, with a quirky style kind of reminiscent of web1.0 sites but with more... gradients". pasted in some docs also
oh here's the actual url for Malcolm Ocean's LLM Context Archive malcolmocean.github.io/context-for-ll… A collection of content for AI conversations and training
@soundrotator yeah Claude can totally do it. and github pages is free hosting for anything frontend-only (roughly = no database or email or user-auth) eg I just whipped up this yesterday:
@soundrotator and then it felt weird to have nonnaivetrust.dance not have a site on it, so I made a page for that too. I probably customized some stuff here manually, but only because that was faster for me than vibecoding it
ALL-CAPS IS ITS OWN LANGUAGE
Replying to @HephaistosF
yeah, seems like your options are roughly: 1. sell/work only in contexts where your past doesn't come up 2. begin again with a new identity that's entirely unlinked from your previous one (which would require maintaining enough vibe diff and social graph diff for people to not notice. pretty hard if you want to make a whole vibe) 3. find a way to talk about your story in the context of the stories about you that other people are telling, that allows at least some people to see ways in which they can trust you, even if there are other ways they can't (which would involve some amount of bridging those perspectives, which I know is emotionally difficult given the amount of hatred laced in some of them, in addition to being conceptually difficult).
for #3, like... my sense is that aside from a few crusaders, most people really don't care that much, and if they read something from you that was sufficiently self-aware and not trying to make a bid for them to take your side about the whole thing but just to be open to friendly interaction or economic transaction, then they'd be like "okay, idk what happened here but it's not really any of my business. I might not invite this guy to my party but I'll hire him to build my game"
@sparr0 ~300: the constitution itself. once you get to scales like that you're not talking commitments between individuals. also something lasting for centuries doesn't require it to have had a multi-century commitment ~100: the pledge of allegiance (in some cases, not most)
suddenly remembered @djmicrobeads
> Social media made exit impossible, and so Americans abruptly went back to voice. Thrown into one small room with each other, they began to complain and fight. insofar as this model is right, it's a big deal and may need cultural & tech change to handle
why aren't there new cars shaped like this anymore? (right answers and wrong answers both welcome)
I've been lately thinking about getting/making a low-end android phone whose OS is configured to only open one app. unbundling. kind of like a kindle I guess, but it could be centered around something else
@djmicrobeads I really don't know what prompted it. I was like, heading to bed, and then I was like "what happened to that guy?" I think you were off twitter for a few years? or maybe I'm misremembering. ah yeah "account created Feb 2025"
@jambarree share a thing with me right here that I probably don't know about! (all I know about is IS, so it should be pretty easy!)
morning has broken, like the first morning black bird has spoken, like the first bird language has woken, language is learning what's the next token? what's the next word?
huh, the base models (this is V3) are now fairly familiar with the "Human: ... Assistant: ..." dialogue format, from its presence in the training data. this is both cool and lame
working on setting up ways to talk to simulacrum characters living in base models, and one of them just mentioned reading a book last week (making me realize that the simulation is kind of timeless, like the present changes the past) and anyway I might buy Alien Phenomenology now
playing with base models is giving me this weird sense that you can freeze a good conversation and then cryonically revive it
I love janus's lil robot party just thought of this again weeks later
"I'm having trouble getting my head into the blue space" in which v3base-Malcolm expounds on various spiral dynamics stages base models for shadow work? 🤔
Retweeted @RichardMCNgo
I increasingly believe that there are fundamental principles which simultaneously govern the designs of well-functioning…
every time an LLM had a never-before-seen insight during training, that way of thinking got pruned
@DefenderOfBasic same is true of religions EXCEPT to the extent that religions are also mediums for propagating and transmitting local, parochial (in a non-derogatory sense) information, not just universal. ie culture. which they very much are! or have been. and there you need/want vast plurality
every time I see the phrase "touch grass" I can't help but remember this incredible reply
subtle memetic colonialism: a few years ago I realized I'd been subtly imagining my ancestors were in the levant 2-5000 years ago. because I grew up telling the stories of jesus, and before that the jewish old testament. I knew my family came to canada from scotland, but...
@DefenderOfBasic like it's good for the jews to be the jews with the stories of the jews. and it's good for the indigenous peoples of north america to be themselves with their stories. and meanwhile other people are living in mashups
I laughed
@caesararum my audio recorder auto-updated an hour into me recording an amazing 4h conversation (while my phone's screen was off) 🤬 the only way I knew this was how my recording was destroyed was because I guessed it and then checked google play update logs.
@yashkaf my impression (not well-researched) is that some countries have tried a fair bit of dakka & they've still failed to get anything above replacement (although sometimes people will have kids a year or two early because they're worried that the birth bonus won't last. but not more)
we're just getting started and we've been going awhile and it's never too late to bring people into the lore
@sparr0 (also curious what @RichDecibels thinks about this, in terms of his intent with the original diagram)
@shakoistsLog yeah I have so many little tools that would be easy to vibecode but they require setting up a whole google auth thing which is a whole extra project
@ablata_at_alba @tenobrus yeah I definitely resonate with the spirit of this a lot I wonder if we'll see LLM-tech and related stuff become able to reverse-engineer these game servers. obviously harder when you don't have the source, but you could trace/store a bunch of IO and sim ghuntley.com/tradecraft/
only applies to open problems in their existing codebases, but in any case the part where devs FEEL faster even when they're slower is pretty interesting
@repligate the thing about problems with singletons reminds me a lot of a talk @eshear and @meaningness gave at less.online, Seeing Like a Good King, about how having too much power corrupts your ability to even figure out what's going on.
got V3-base to do text-completion on one of my drafts and it figured it should include a hallucinated image from @davidad's site 👍🐎
if I made an ebook (or a paper book!) out of this sequence of posts on unverified personal gnosis (aka "I can tell for myself") would you be interested in buying it? it talks about how we lose touch with our knowing, guru dynamics, and knowing-together malcolmocean.com/2023/11/i-can-…
"poets who believe in Rhyming" tarred with the "conservative" brush, alongside creationism and racism the absurdity of antimodernity, as described in @Meaningness's essay Ofermōd, is articulating itself live on the internet
bye guys, time to go engender some floatiness
some reasons why people end up in guru dynamics, creating a context where they're trying to teach other people how to tell for themselves
ChatGPT hallucinates a feature for someone's app, ~forcing the app to implement it (or look bad and get a lot of complaints) but also hey, free marketing and demand research! holovaty.com/writing/chatgp…
note that if you talk to claude via your own API or whatever, you can just tell it some winter date (or, more simply, no date) and it'll work just as contentedly as usual
pixel 4a phone screen broke again all the new phones are too big drooling at my 4" screen phones from early 2010s time to RETVRN
oh also you don't have to call it claude! my experience is that if you talk to claude through an API with a substantially new role prompt, and ask it if it's claude, it generally says "no, but I run on claude" which is pretty cool. I wonder if they trained that
few things worse than getting interrupted when your yak is only half-shaved
4♥️, 0💬 ...apparently not lol. or maybe I just need to get better at framing the whole thing. like "book about..." vs "book that will help you..."
tfw you think nobody got you a cake for your birthday, but then you go to open your presents and each one turns out to be made of cake
I found a workable solution for using spaces! the key is you rarely want to "switch space". you want to "switch app" to an app that is ON a different space:
@pingToven @nostalgebraist but it's still a wrapper around chutes yeah? any advantages aside from having everything in one place?
made a quick vid explaining on how to get started playing with LLM base models using chutes.ai! everything else had been slightly too confusing
I'm unreasonably excited about this new tiny phone arriving today. like maybe I'll hate it (if eg the screen is a bit slow) but I'm also prepared to love it so much. I don't want my phone to be good for watching videos on!
if there were a way to genetically encode "select the best possible mate that's realistic for you to get" everybody would have that gene by now
2025 problems
in general people have had a sense that LLMs don't REALLY have world-models. like implicitly, but it isn't coherent/generalizable thread draws an analogy with Kepler's planetary motion predictions (which were flawless!) vs Newton's deeper laws of gravity/motion
so much "what if it were good tho" energy in this thread
anybody know how to access kimi k2 base? doesn't look like openrouter has it (unclear but looks like only the chat/instruct version)
slaps roof of "later" this bad boy can fit so many tasks in it
as part of helping increase my expectancy around projects, I'm undertaking to see a bunch of things to public fruition 2y ago I built this tool that gives you spellcheck for writing in anglish, but it sat on my harddrive til today, when I put it online and made this video demo:
yo @_cturner if @learnmoonrise has sports teams, I found the perfect song for yall to blast when the team walks on the field youtube.com/watch?v=EsmvXz…
Retweeted @Malcolm_Ocean
@metamitya multiQT makes the comparison itself atomic imagine if you had to split up memes like this into 4 tweets with…
Retweeted @chandanaveli
For a long time, I was thinking "there's gotta be a better way to share a bunch of tweets together", so I made https://t.…
wowwww some research (who knows if it replicates) suggesting that white liberals talk down to black people while white conservatives don't. hyperfixation on race + attempts to get along = sloppily using stereotypes. insights.som.yale.edu/insights/white…
a conversation and a mind are essentially the same thing
being dead is not very evolutionarily fit!
in Canada, there are three types of metal money: loonies, twoonies, and quarters, plus smaller pieces of metal that aren't money. in the USA, there's no metal money, but sometimes businesses will give you laundry tokens as change
there's a paradox that shows up any time someone tries to teach someone else self-trust
stop me in the name of love if you've heard this one before:
Retweeted @repligate
Inspired by QT
I think this might be the first time I've stared at a video for 30s thinking "feels AI-generated" and then I was like "oh maybe it's not". youtube.com/watch?v=rB2o67…
watching this CGP Grey video about memetic hygiene and his description of going on social media being exposing yourself to everybody's sneezes has me wanting to make a social media app where the posts are just called Sneezes youtube.com/watch?v=rE3j_R…
"life is information that is embodied" - @Morphenius
huh, memes becoming so aggressive at spreading that they alienate people... is kind of like cancer getting cancer
nah it's more subtle than that. if you anticipate your problem being solved, you move towards solving it (psycho-cybernetics). having a context that draws you into such an anticipation and helps you participate in it and clarify what you want, is a real resource
Epstein bringing the country together
@ConvergeToTruth right, but the thing I'm pointing at is that the actual CONTENT of the meme gets undermined by the fact that it evolves into a form that spreads so fast it triggers this response like spreading faster seems good for the meme... until it isn't
@bierlingm @ArtemKRSV @daylightco which video is this?
Replying to @eigenrobot
@eigenrobot "music is a promise that you aren't listening alone" x.com/Malcolm_Ocean/…
Retweeted @utotranslucence
sign up for my religion my religion is just every sunday we turn off our phones
sucks 💔 "there's a reason for everything" is confused afaict. what's actually true: 1. there are causes to everything, and more intelligence at work than you might assume 2. everything that happens, leads to everything else that happens after, including things you may cherish immensely before I elaborate on #2, I want to note: the loss of your daughter is utterly tragic and what I'm about to say doesn't imply that it was "good" in some "overall" sense for her to have died, or anything like that. that's not so much wrong as incoherent. there is no ledger like that. it doesn't all add up—there's no way to turn it into numbers. so. you lost your daughter. and your life is going to be very different as a result of that. this is much larger than a little butterfly. and much will be worse. you'll miss her, and you'll ache. and... someday something absolutely unequivocally lovely will happen—a new child, a new friend, a creative work, or some other possibility. and you will know that you wouldn't be having that experience in the world where you still had your daughter. that doesn't make it "worth it"—again, there's no comparison. it's moot. but... there, in that moment of you experiencing the preciousness of what is happening... beauty keeps going, and it builds upon every tragedy. it folds in everything that has happened. and this can be narrativized as "everything happens for a reason", but... that misses something about it all. anyway, I hope that helps! I have more writing/thoughts broadly related to this, but in this context it doesn't feel appropriate to link it unless you specifically ask.
"reflecting on how all my life it seemed like there was such a tremendous emphasis on the value of being emotionally aware, and yet how all of the really wicked problems I’ve had + I’ve known people around me to have had, have come from incompetence with thinking things through" ...is there someone for whom this is true? if you solve your bottleneck, you get a new bottleneck somewhere else. maybe this is a you (and your scene) thing. (I don't particularly think so but it's interesting to think about!)
I got chewed out in high school for calling something a 'quote' instead of a 'quotation', and reflected that yeah, that makes sense in comparison to eg 'calculation'... but now in 2025 somehow the word 'compute' is going through the same shift?? interesting linguistic phenom
hey @grok rank my mutuals by who's given you the best "rank my mutuals" task
much writing spends a lot of time critiquing competing theories (examples in next tweet). this is helpful, but I'd love to see many of them rewritten the way they would be written as a kids curriculum in a world where they're obvious to everyone!
@AbstractFairy lots of general public books still have this property of being reactionary in this sense. there's just a general feeling of wanting to define something by highlighting the (common!) errors it doesn't make, rather than by presenting it on its own terms
half-woke-up at 4am thinking about one of my old pieces of writing and found myself idly wondering if I was being simulated inside an LLM
fellas, is it gay to have more clothes than your wife?
win-wins are always best when you can find 'em. AI tech can help. current tech is ready to start helping. this is the best proposal I've yet seen for this: x.com/HephaistosF/st…
using an adjustable weight set to get the swingset swinging to the exact tempo of the song
swatted some houseflies and after a few minutes another fly started poking at their corpses, and at first I imagined it being like "oh my brethren are dead" but then I remembered it was a fly and that it was just as likely thinking "hmm, good spot to lay my eggs?"
"...I feel like part of what's going on is that I need a church... in order to orient to this church..."
at camp with running water but no hot water and really struck by how hard it is to wash stuff. soap is great but hot water makes about as much of a difference
