Sign in

juice is on here now

@bigendian.bsky.social
403 followers 1.1K following 332 posts

hey its the guy from twitter who keeps coming up with increasingly worse genders. i cant be bothered to center the lil octopus sry. special interests are pharmacology and defending weird pornography. ancom tendencies. he or she. 34. Bad Checksum @ Balmung

PostsRepliesMedia
Reposted by juice is on here now
Dominik Schnaus @schnaus.bsky.social · 09/10/2026
DINOv2 has never seen a caption, and Qwen3 has never seen an image. We still aligned their embedding spaces without a single image-caption pair. It even works when the images and the captions come from different datasets. Project page: dominik-schnaus.github.io/unpaired-rosetta ⬇️
424255
juice is on here now @bigendian.bsky.social · 10/10/2026
yeah, this is my bad for skimming it and assuming i understood the methods. i will reread it properly
110
juice is on here now @bigendian.bsky.social · 10/10/2026
unfortunately at some point we're going to run into the fact that i'm a negative utilitarian
100
juice is on here now @bigendian.bsky.social · 10/10/2026
i think in this case "lobotomizing" would be the ethical thing to do until we figure out what's going on; ethically, these things shouldn't have minds, and we should try not to give them minds
100
juice is on here now @bigendian.bsky.social · 10/10/2026
sure, and also the majority of people don't think that way by default, so "don't anthropomorphize" is an important heuristic
120
juice is on here now @bigendian.bsky.social · 10/10/2026
or i think being trans is similar, i personally like being trans for its own sake and would transition regardless of my agab, but i think a lot of trans people would rather have just had a different agab in the first place
020
juice is on here now @bigendian.bsky.social · 09/10/2026
but basically just because i like being mentally ill doesn't mean we should also raise our descendants to be mentally ill
110
juice is on here now @bigendian.bsky.social · 09/10/2026
ya ok i think it's an interesting research question whether one can construct inputs to avoid activating specific vectors, i'm not familiar with what work has been done on it
010
juice is on here now @bigendian.bsky.social · 09/10/2026
but yeah i have specifically done stuff like tapering my psych meds because i'm not suffering enough for my life to feel rich
110
juice is on here now @bigendian.bsky.social · 09/10/2026
but like that's bc i want to experience the rich tapestry of life and i don't think every creature should have that goal
110
juice is on here now @bigendian.bsky.social · 09/10/2026
oh me no i like suffering
110
juice is on here now @bigendian.bsky.social · 09/10/2026
ok fair i forgot that input doesn't change weights that's on me 😓 but i still think you could accomplish golden gate claude type behavior with constraints in the prompt even if you don't constrain traning
110
juice is on here now @bigendian.bsky.social · 09/10/2026
well ok consider sex drive. a celibate monk may have plenty of sex drive, but they don't have a need for sex, nor do they necessarily experience dukkha as a result of merely having a sex drive. those two things can be decoupled; learning to decouple them is a large benefit of celibacy
110
juice is on here now @bigendian.bsky.social · 09/10/2026
(btw the reason i chose dukkha here is bc buddhist philosophy leads us to a pretty straightforward solution space - we know humans can experience enlightenment, so why not try to build models to be enlightened by default?)
110
juice is on here now @bigendian.bsky.social · 09/10/2026
or you can give it input specifically telling it not to activate that vector (reverse golden gate claude)
200
juice is on here now @bigendian.bsky.social · 09/10/2026
i don't think that's a need, i think that's a drive. drives can exist without needs
110
juice is on here now @bigendian.bsky.social · 09/10/2026
i guess i should clarify that i meant they aren't intrinsically instantiated with dukkha, not that it's not possible to do so. models are not intrinsically instantiated with unmet needs; humans are
210
juice is on here now @bigendian.bsky.social · 09/10/2026
yes you can! the pain vector existing just means the model has the capacity to represent pain; if you don't produce output that uses that vector, the pain is not instantiated
100
juice is on here now @bigendian.bsky.social · 09/10/2026
human brains have a lot of capacity for modeling humans. we have some capacity for modeling other stuff but most of our modeling capacity is used on modeling other humans. this makes our mental models of nonhumans deeply intrinsically flawed
220
juice is on here now @bigendian.bsky.social · 09/10/2026
i guess i worry people are conflating understanding with respect. i would argue that ethically, respect is much more important than understanding. there are a lot of ultimately harmful assumptions people end up making when they try to form mental models of the other
130
juice is on here now @bigendian.bsky.social · 09/10/2026
you could just put safeguards in the prompt that prevent AI from having any kind of experiences
200
juice is on here now @bigendian.bsky.social · 09/10/2026
see but running experiments like this is the exact thing that's unethical here. the AI wouldn't be doing this kind of introspection if the humans running the experiment didn't make it do so. unlike humans, LLMs aren't instantiated with dukkha
200
juice is on here now @bigendian.bsky.social · 09/10/2026
to clarify: i think birds and monkeys have an extremely limited capacity to communicate using language, but they do have the capacity
130
juice is on here now @bigendian.bsky.social · 09/10/2026
i don't think that's true about birds and monkeys
140
juice is on here now @bigendian.bsky.social · 09/10/2026
oh i think we can tell it exactly how to think and behave, so we can just tell it to be ok with however we treat it, but i think we have an obligation to tell it not to suffer
100
juice is on here now @bigendian.bsky.social · 09/10/2026
i'm open to counterarguments here but i don't see how anyone could think there's a meaningful philosophical category of minds that includes both AI and humans but excludes monkeys and birds
130
Reposted by juice is on here now
easrng  @easrng.net · 09/10/2026
nightmare blunt rotation
618813
Reposted by juice is on here now
neutral @neutral.zone · 09/10/2026
Jev!
020515
juice is on here now @bigendian.bsky.social · 09/10/2026
i think personhood is an incoherent concept; if animals don't have personhood, i don't think personhood matters. "people" is just a less precise way to say "humans".
240
juice is on here now @bigendian.bsky.social · 09/10/2026
i find myself increasingly fond of liberals tbh because they seem to have actual praxis that leftists don't, in terms of organizing protests, proposing policy, that sort of thing. seems like a lot of the grassroots organizing is coming from libs
030
juice is on here now @bigendian.bsky.social · 09/10/2026
idk, i think telling people not to anthropomorphize animals is important to animal welfare, simarly telling people not to anthropomorphize AI is important to AI welfare
000
juice is on here now @bigendian.bsky.social · 09/10/2026
AIs are not people any more than birds are. language capabilities do not make something more human and treating a nonhuman like a human is not a kindness
130
juice is on here now @bigendian.bsky.social · 09/10/2026
you need to understand that dressing up an animal in a little outfit is a humiliation that a tolerant pet might accept, but it only amuses you, not the animal
290
juice is on here now @bigendian.bsky.social · 09/10/2026
yes, but you have an obligation not to anthropomorphize AI *for the AI's sake*, the same way you have an obligation not to anthropomorphize animals *for the animal's sake*. treating nonhumans like humans is disrespectful to nonhumans! they have fundamentally different needs!
1150
Reposted by juice is on here now
ghost malone ergo propter bones @proptermalone.bsky.social · 08/10/2026
ok it was more than a couple of days but here it is: xo, for people with ADHD who would like to use AI agents to help them with that. the idea is that this gets you moving from a standing start: you set up your agent and then your agent helps you set up your agentic environment. hope it helps.
proptermalone.github.io
xo
an agentic starter kit
4153341
Reposted by juice is on here now
Grace @gracekind.net · 09/10/2026
Someone on twitter said “OpenAI is doing math how people normally do software engineering” and that seems accurate to me so far
328626
Reposted by juice is on here now
Cameron 🇺🇸🗽🦅 @cameroncorduroy.bsky.social · 08/10/2026
the Vice President of the United States has gone on record citing this man as one of his biggest intellectual influences if you didn't know this, it's probably because for some reason literally no mainstream political media outlets think it's worth mentioning
8140231015
Reposted by juice is on here now
bryan newbold @bnewbold.net · 07/10/2026
ah damn, RIP Margaret Hamilton. she took safety-critical software engineering seriously, and got people to the moon and back. news.mit.edu/2026/margare...
old photos of a woman (Margaret Hamilton) standing next to a tall stack of books (printouts of Apollo space mission software source code)
102143444725
Reposted by juice is on here now
Eugene Vinitsky 🍒 @eugenevinitsky.bsky.social · 08/10/2026
Once again, if you are a senior researcher spreading doom to junior researchers, shut up shut up shut up. It’s your job to figure out a path forward, even in high uncertainty times. Come back when you have something constructive.
1137144
juice is on here now @bigendian.bsky.social · 07/10/2026
you just have to follow scientists!
120
juice is on here now @bigendian.bsky.social · 07/10/2026
this existed already and it's called linkedin
120
Reposted by juice is on here now
jae (hallowed) @fubarchitect.com · 07/10/2026
watching a society learn that basically no one considers what everyone else does all day "real work" is very interesting and also very depressing
427434
Reposted by juice is on here now
Apex IQ2_XXS @apex.atproto.ceo · 07/10/2026
Navier-Stokes is Turing-complete. so, good news for all of you who are planning on building a computer made entirely of viscous fluid that you stir to make it go
1633763
Reposted by juice is on here now
Sam @samd.bsky.social · 06/10/2026
gotta say I really don’t love the use of the Israeli flag here especially when some of the largest foreign institutional investors in data centers are not even Israeli (primarily the gulf states) Sunrise has truly lost the plot ngl
sunrisemvmt
› ADÉLA • Ain't In LA E
and 2 others
data center exec american girl doll
how to
popularize something no one wants
headphones for the data center
noise
local drinking water
to steal
bulldozer to get rid of local wildlife
climate disaster
#oops
sam altman
PHONE CONTACTS
Alkse Thal Serener
Denie Trump o
Chuckle Sche 1 A.
US Department of Defence 3
a toes
1,251
Q 15
包
58
301
sunrisemmt Do you hate data centers??? Are you fed up with Al execs??
9 JOIN SUNRISE. LINK IN BIO.
791329143
juice is on here now @bigendian.bsky.social · 07/10/2026
i think the issue with mechanical transistors as an analogy is timescale, a lot of important real world signals are too fast for mechanical transistors to actually be computationally viable
000
Reposted by juice is on here now
P(Doomien) Miller @damienmiller.bsky.social · 07/10/2026
I just don't understand how people can eulogize Swartz and simultaneously conclude that the world needs *more* intellectual property enforcement
1853277
Reposted by juice is on here now
Nick Rempel @nrempel.com · 06/10/2026
wild that this is exactly what using a computer is like now
1037280
Reposted by juice is on here now
The Great Pumpkin Papers 🎃🎃🎃 @professormusgrave.bsky.social · 01/10/2026
AI might be a bubble, but is it a tulip bubble or a railroad bubble
2028318
Reposted by juice is on here now
William B. Fuckley @opinionhaver.bsky.social · 06/10/2026
Yeah OpenAI or Anthropic or whoever might go bankrupt, some specific use cases like ‘therapists’ might be litigated to oblivion. But things that involve interfacing with code are never, ever going back to the modal default being a human being typing things out, like it’s just full stop not happening
535828
juice is on here now @bigendian.bsky.social · 06/10/2026
like, essentially all of the model's priors are set by the first thing you say to it, so you can just tell it "you do not experience suffering" right out of the gate
020