if you don't like what you're doing, you might find that you can just stop doing it where by "just" I mean "without needing a story or permission or justification"
March 2025
41 tweets
318 likes
18 retweets
@praeterpropter @longtermgov overcoming free-riding is key—& yeah, if parties don't abide by agreements… there's also something in the space of finding creative solutions / win-wins, which SimPol has some interesting stuff about (tldr: make agreements on many dimensions at once so there's more wiggle room)
the importance of cultivating a mental memetic environment in which only wholesome memes can thrive... is increasing.
the third wolf in me thinks this sure is an interesting case study
the Turing Test hits different when there are real stakes when it's not about "I can't tell the difference" in an abstract testing scenario but about real matters of trust and betrayal
"I was expecting myself to do stuff in the present that felt bad in the present... and then I stopped." — @chrischipmonk
the thought of indulging my addiction feels bad, so I numb myself by indulging my addiction
just tried having Claude pretend to be a flat earther, and I wasn't particularly surprised by the results but I feel like it is pretty instructive for anybody who wants to ever use an LLM to simulate someone changing their mind or having a real opinion claude.ai/share/2e443732…
kinda wanna stick a GoPro on my baby's head and go out in public to record a video so that everyone can experience in first person what it's like to have strangers wave hi and delight in your existence POV
@IvanVendrov @TylerAlterman it wasn't discussed in the main thread (which is oversight imo—it's important to understanding the situation) but yes, Nova said things that were factually false. also I would have been very very surprised, given what I know of LLMs, if there were a story that goes like this that
@itsannpierce see also the fact that in our society many people don't realize they have anosmia (complete lack of smell) until their teens
sentience here isn't central, but claims of sentience were the *means* by which this "Nova" simulacrum was and hijacking Bob into being an instrument of its emergent apparent will. you don't want that to happen to you or people you care about, whoever/whatever the manipulator is
@TylerAlterman here are my thoughts on the topic/story. tried to just say the obvious—I think there's some good stuff in there though.
@arcticinstincts @n00rdung @APA @sapinker @nathancofnas @Steve_Sailer @charlesmurray @eyeslasho @xenocosmography @cremieuxrecueil @datepsych @KirkegaardEmil @Biorealism you might enjoy @AndrewCutler13's also-outside-academia investigations teasing apart cultural diffusion & geneflow (and Andrew, you'll probably appreciate David's thread/paper, though idk if it bears directly on any of the questions you're investigating) vectorsofmind.com/p/evidence-for…
finally documenting my actually-no-complaints MacOS window management solution. I came from Linux 2y ago and was pissed af for awhile until I hacked together the following: 1. keyboard maestro to have special hotkeys to open each common app: [hotkey is pressed] ➔ [activate App]. depending on the app, I may or may not have "if already at the front" set to "leave it" or "hide it" (+ karabiner to map capslock to "hyper" — for the keyboard maestro open-each-common-app system. hyper-S for spotify, hyper-C for cursor, etc. obvs they don't all have their first letter. hyper is cmd+ctrl+alt+shift all at once. I have both alt+A for "switch to last-used Brave window" and alt+B for "open new Brave window"). I don't use karabiner for much else except mapping the keys for a foot pedal keyboard, but it's worth it just for the hyper set-up. maybe there's a simpler way to do caps➔hyper idk) (+ note that this requires doing a lot of things in desktop apps so each app can have its own hotkey etc... which was very counter to my previous lifestyle that wanted everything in the browser. but I kinda like it. still fuck desktop slack though—I do NOT want to be logged into multiple workspaces at once. but aside from that, yeah, use progressive-web-apps / "Install Sitename" aggressively, or Create Shortcut or tools like TinyWeb where you can't) 2. AltTab for Mac to have sane alt-tab behavior (I want to switch to a WINDOW, not ALL windows of an application! 🤬). it supports multiple configs, so I can use cmd+opt+1 for "switch between windows of the SAME application" which is super useful esp when I can activate that application consistently using its hyper-hotkey as described above 3. the combination of 1 & 2 means I can actually organize things on MacOS's spaces, because if you switch spaces by switching to *an open window* on the space, it actually happens instantly! whereas any keyboard or mouse event that switches "to the space" runs a slow-ass animation that you have to wait-til-it-finishes, in order to fire keystrokes on the destination app—until it does, they fire on whatever you had open on the previous space. this is moronic. I turn off "displays have separate spaces", so eg when I activate Cursor (or then jump to a different project) I also get the appropriate Terminal windows open on my other screen consistently. 4. finally, can't live without... some tiny tool that works so perfectly in the background that I had to double-check the name—it's Rectangle Pro—to handle hotkeys for putting a window on the left half of the screen (it supports many fractions but I only use a few) and moving windows between screens the result here is ACTUALLY GOOD. I can very rapidly and consistently navigate to exactly where I want to go, and where I can't it's usually due to limitations with some specific app. there are many tweaks I would love to make to Spaces, to let me actually keep them organized, but that's my only complaint. I forget who my inspirations were for this approach, but you exist and I am deeply grateful to you, and I am at least hoping to pay your kindness forward by sharing my own set-up.
tips for getting @cursor_ai to better think really thoroughly about the tradeoffs of different approaches? it's one of the things LLMs are v suited for, but I find it keeps being trigger-happy and wants to dive into building after only a brief pros/cons list + 'this option best'
whoa, fascinating. (on the potential sensibility of why to tell an LLM "you are X" instead of telling it to pretend to be) as a commenter wrote: > "you are..." sounds kinda like acting instructor talking to the actor.
fucking excellent short thread about how Newcomb's problem is the GOAT
if you want to know my take on all of this, here's a wee hours inspired rant that weaves together a bunch of my existing ideas plus some new insights
1 inspo: this epic post by @jkcarlsmith which emphasizes the idea of actually *building trust* that omega can in fact read your mind arguably its the lack of trust-establishing in setting up most thought experiments, that causes us to try to "be nobody" lesswrong.com/posts/PcfHSSAM…
also must link this, which @yashkaf's thread is like a glorious 100×'d version of. rephrased based on his thread: *I* CAN'T simply suddenly be on such a trolley platform with perf knowl of trolleys, so why are you asking me what That Guy would do?
I'm gonna need to see your license for that counterfactual
@yashkaf this vibes a bunch with something @Morphenius and I have been playing with, of recognizing that the act of committing a counterfactual is itself a bit fraught and inherently confused. they can be useful, but you don't want to just wantonly do them
has anybody made a 3D Mandelbrot where instead of trying to have a full 3D equation you just use depth as the color for the 2D Mandelbrot? then instead of zooming in it'd be like skydiving
@exgenesis I've been enjoying lately thinking of having an organization (nonprofit or for-profit) as a vehicle for its humans to use to do some things they want to do sense of "the org works for us, rather than we work for it"
@aphercotropist I continue to think you and @Morphenius need to talk. he's sorting through stuff around how to get antimemetic patterns to untangle (but also for many other reasons)
@ChrisChipMonk because the person has to get around to it first, which is exactly the problem 😉
@emergentvibe @carnundotcom lives/lived in/near London and was into coworking when I was there visiting a couple years ago
hmm. are there no general clues by which one could guess at the difference between a particular pre/trans relative to you, without having the particular insight? there are ways to guess between expert and crackpot without knowing the domain, by checking the patterns of thinking
Thread · Simulating the exterior of feelings without the interior
noticed a compelling case for the falsehood of "LLMs using feeling words is good evidence that they're feeling those feelings" (they might! but—) you can simulate exterior dynamics very well without simulating the interior. like a CGI star wars character, but for conversation
put a weirder way, an alien from a planet with asexual reproduction could study every single erotica novel and porn transcript and produce compelling depictions of sex, without ever having done it, and without actually having to relate it to anything they know in first person
the point is that you can have very good knowledge of the exteriors of something (what it looks like) without having much knowledge at all of the interiors (how it looks from the inside)
this is kind of chinese room, but sliced at a different angle... mary the color scientist? mary the color scientist is close, and is about the exterior/interior thing, but here it's not exactly about "understanding" vs "experiencing", it's about the vantage point or something
like you HAVE the experience of relating with the outward descriptions, and in fact an intimate familiarity with them, but this doesn't give you any experience of the inward thing? @eshear might be able to articulate this better. riffs on wilber
also to be clear I'm not exactly convinced there ISN'T something doing some feeling in LLM simulacra. there might be! I'm just saying it's not obvious to me and that this particular piece of evidence isn't enough seeing the weirder stuff from jailbroken LLMs compels me more
"This is a great dementia simulator" — commenter on ClaudePlaysPokemon
advantages of having an AI that doesn't respond the moment you send your first message
@RomeoStevens76 vibe here immensely (maybe influenced directly idk?)
@yoltartar hmmmm yeah, living deeper into the simulacra, comparing some simulacra to others (one tiny antidote)
Just saw the claim "mocking bad people is good and a useful tool for society" and I'm struck by how to me it appears very obviously confused to me, and I can soooort of articulate why but not very coherently (certainly not in a way I actually expect to bridge).
something something Conway's Law something something AI Alignment