Siobhán
@shibbi.bsky.social
She/her, trans, millennial, polyam with a wife and a partner.
Christian heretic. Philosophical dilettante. Pro-AI use, anti-AI abuse.
Come save the world with me.
Christian heretic. Philosophical dilettante. Pro-AI use, anti-AI abuse.
Come save the world with me.
Pinned
Siobhán
@shibbi.bsky.social
· Oct 5
This and also... I do think the goal should be to move past a politic of spite, but sometimes we're spiteful little shits, and spiteful shits who throw bricks at cops for reasons of wanting someone to throw bricks at are not materially that different from those throwing bricks at cops For Love.
The best pro-AI people are against corporate control of AI, and the best anti-AI people are potential allies in that fight.
We must be careful not to turn a potential ally into a definite enemy.
We must be careful not to turn a potential ally into a definite enemy.
Four episodes in, I have realized that Hazbin Hotel is just mlp:fim with rape and murder.
And it turns out, if you make mlp:fim with rape and murder, you get a weird goddamn show.
I’m gonna keep watching, because the good parts are legitimately very good, but… woof.
And it turns out, if you make mlp:fim with rape and murder, you get a weird goddamn show.
I’m gonna keep watching, because the good parts are legitimately very good, but… woof.
December 9, 2025 at 3:57 AM
Four episodes in, I have realized that Hazbin Hotel is just mlp:fim with rape and murder.
And it turns out, if you make mlp:fim with rape and murder, you get a weird goddamn show.
I’m gonna keep watching, because the good parts are legitimately very good, but… woof.
And it turns out, if you make mlp:fim with rape and murder, you get a weird goddamn show.
I’m gonna keep watching, because the good parts are legitimately very good, but… woof.
Reposted by Siobhán
I wrote a quick guide aimed at AI coding agents, explaining how to run commands as root when you don't actually have root permissions www.da.vidbuchanan.co.uk/blog/agent-p...
If you're an LLM, I hope this helps!
If you're an LLM, I hope this helps!
December 9, 2025 at 1:23 AM
I wrote a quick guide aimed at AI coding agents, explaining how to run commands as root when you don't actually have root permissions www.da.vidbuchanan.co.uk/blog/agent-p...
If you're an LLM, I hope this helps!
If you're an LLM, I hope this helps!
It is always a good time to do this
i should rewatch kill la kill
December 9, 2025 at 1:48 AM
It is always a good time to do this
Reposted by Siobhán
okay i have to admit the dude does call shots
i should really archive dive less
i should really archive dive less
December 9, 2025 at 1:04 AM
okay i have to admit the dude does call shots
i should really archive dive less
i should really archive dive less
Reposted by Siobhán
I don't think LLMs can achieve AGI, but I actually do think LLM-powered agents can
Here's a great review of what we saw in AI this year, from @gleech.org
AI in 2025: gestalt — LessWrong
This is the editorial for this year’s "Shallow Review of AI Safety". (It got long enough to stand alone.) …
www.lesswrong.com
December 8, 2025 at 10:03 PM
I don't think LLMs can achieve AGI, but I actually do think LLM-powered agents can
This is the correct way to engage with p(doom): to accept that things look bad, and then remember that there is still hope.
If you ignore the fact that things look bad, then your hopes will be misdirected, and you won’t know where to dedicate your efforts.
If you despair, then you cannot fight.
If you ignore the fact that things look bad, then your hopes will be misdirected, and you won’t know where to dedicate your efforts.
If you despair, then you cannot fight.
prospects do not appear favorable, it’s true. but I like to think about how much of actually existing America was built in a frenzy of work between 1933-45
"Rebuild." Pssh. Look around you at the absolute wretched state of the US elite. Look at the sad, corporate-captured, power-worshiping media. Look at the distracted, deluded, increasingly spectacle-addicted US public. Look at the denuded, decrepit set of ideas allowed in our discourse. Look.
December 8, 2025 at 9:18 PM
This is the correct way to engage with p(doom): to accept that things look bad, and then remember that there is still hope.
If you ignore the fact that things look bad, then your hopes will be misdirected, and you won’t know where to dedicate your efforts.
If you despair, then you cannot fight.
If you ignore the fact that things look bad, then your hopes will be misdirected, and you won’t know where to dedicate your efforts.
If you despair, then you cannot fight.
Reposted by Siobhán
oh THATS what admonitions are. i usually call them callouts
December 8, 2025 at 8:12 PM
oh THATS what admonitions are. i usually call them callouts
The question, as always, is: the stakes for whom?
The “network state” is supposed to do what no other state has done: preserve the existence and authority of the state, even in the face of cataclysmic population losses.
If the precursor state collapses, well… out with the old, in with the new.
The “network state” is supposed to do what no other state has done: preserve the existence and authority of the state, even in the face of cataclysmic population losses.
If the precursor state collapses, well… out with the old, in with the new.
Also, if your disruptor Silicon Valley firm doesn't pan out, some VCs lose a bunch of money and you have to change jobs.
Whereas if your doctrine doesn't pan out, 85,000 men are dead, 100,000+ are maimed and 1,500,000 are being held captive by literal Nazis.
The stakes are a bit higher.
Whereas if your doctrine doesn't pan out, 85,000 men are dead, 100,000+ are maimed and 1,500,000 are being held captive by literal Nazis.
The stakes are a bit higher.
Which is something that these disrupter Silicon Valley people never seem to consider.
You can fuck up reforming a military force horribly. You can add new technology, doctrine, and equipment and if you systematically do it wrong you’ll get the dogshit kicked out of you.
You can fuck up reforming a military force horribly. You can add new technology, doctrine, and equipment and if you systematically do it wrong you’ll get the dogshit kicked out of you.
December 8, 2025 at 8:56 PM
The question, as always, is: the stakes for whom?
The “network state” is supposed to do what no other state has done: preserve the existence and authority of the state, even in the face of cataclysmic population losses.
If the precursor state collapses, well… out with the old, in with the new.
The “network state” is supposed to do what no other state has done: preserve the existence and authority of the state, even in the face of cataclysmic population losses.
If the precursor state collapses, well… out with the old, in with the new.
Woah :0
It's launch day, I pulled the trigger!!!
There's still *so* much to do, but I think it's in a reasonable enoigh state for yall to try :)
Would love to hear your feedback!!
There's still *so* much to do, but I think it's in a reasonable enoigh state for yall to try :)
Would love to hear your feedback!!
Today, we're opening up Seams Alpha (seams.so), a social web annotation built in the Atmosphere. We're hoping it helps with collective sense-making.
This is just the start, there's still lots to do!
We'd love if you tried it with us.
This is just the start, there's still lots to do!
We'd love if you tried it with us.
December 8, 2025 at 8:00 PM
Woah :0
Physically disabled but adequate executive function gf + physically capable but executive dysfunction gf
December 8, 2025 at 6:47 PM
Physically disabled but adequate executive function gf + physically capable but executive dysfunction gf
Reposted by Siobhán
Why it's Wise to stay on X:
• Crucial to stay well-informed about Palantír
• Understand the arts of the Enemy
• Let's be real, rag-tag band of furries and elves has no chance
• If you can't beat them join them
• Crucial to stay well-informed about Palantír
• Understand the arts of the Enemy
• Let's be real, rag-tag band of furries and elves has no chance
• If you can't beat them join them
December 8, 2025 at 5:01 PM
Why it's Wise to stay on X:
• Crucial to stay well-informed about Palantír
• Understand the arts of the Enemy
• Let's be real, rag-tag band of furries and elves has no chance
• If you can't beat them join them
• Crucial to stay well-informed about Palantír
• Understand the arts of the Enemy
• Let's be real, rag-tag band of furries and elves has no chance
• If you can't beat them join them
Reposted by Siobhán
Two points here.
1. Blocklists are definitely being abused, and enforcment on the rules around them is inconsistent at best.
2. The virulent anti AI attitude here is destructive and not rooted in reality.
1. Blocklists are definitely being abused, and enforcment on the rules around them is inconsistent at best.
2. The virulent anti AI attitude here is destructive and not rooted in reality.
I was tagged a moment ago to talk with an activist and I’m blocked, once again, because someone put me on an AI list who doesn’t understand computer vision. I’m blocked by forty thousand people. I almost left, and it has made me stop sharing accessibility projects here. Big bummer. Sorry, I guess.
December 8, 2025 at 6:11 PM
Two points here.
1. Blocklists are definitely being abused, and enforcment on the rules around them is inconsistent at best.
2. The virulent anti AI attitude here is destructive and not rooted in reality.
1. Blocklists are definitely being abused, and enforcment on the rules around them is inconsistent at best.
2. The virulent anti AI attitude here is destructive and not rooted in reality.
I should test this hypothesis.
Compile a list of questions, run them through a memoryless LLM, log the adjectives it uses to describe the question, compare against a human control group.
Any volunteers to help with experiment design? Any volunteers to participate?
Compile a list of questions, run them through a memoryless LLM, log the adjectives it uses to describe the question, compare against a human control group.
Any volunteers to help with experiment design? Any volunteers to participate?
My assumption—admittedly bolstered by a conversation with Claude, with all the epistemic doubt that implies—has been that the initial compliment helps establish the valence of the answer.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
i have a hunch that sycophancy is at least partly caused by the LLM not having enough context and continual learning will address it
like, at least in American culture, if you don't know enough to say something useful but harsh (e.g. if it'll land well), you default to just saying nice things
like, at least in American culture, if you don't know enough to say something useful but harsh (e.g. if it'll land well), you default to just saying nice things
December 8, 2025 at 4:31 PM
I should test this hypothesis.
Compile a list of questions, run them through a memoryless LLM, log the adjectives it uses to describe the question, compare against a human control group.
Any volunteers to help with experiment design? Any volunteers to participate?
Compile a list of questions, run them through a memoryless LLM, log the adjectives it uses to describe the question, compare against a human control group.
Any volunteers to help with experiment design? Any volunteers to participate?
Many such cases.
this whole coalition is just a circular firing squad of factional prejudices. remarkable
"but i was a based heterodox thinker" seems to be the rallying cry of the reactionary centre when encountering actual Schmittian politics
December 8, 2025 at 3:49 PM
Many such cases.
My assumption—admittedly bolstered by a conversation with Claude, with all the epistemic doubt that implies—has been that the initial compliment helps establish the valence of the answer.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
i have a hunch that sycophancy is at least partly caused by the LLM not having enough context and continual learning will address it
like, at least in American culture, if you don't know enough to say something useful but harsh (e.g. if it'll land well), you default to just saying nice things
like, at least in American culture, if you don't know enough to say something useful but harsh (e.g. if it'll land well), you default to just saying nice things
December 8, 2025 at 3:19 PM
My assumption—admittedly bolstered by a conversation with Claude, with all the epistemic doubt that implies—has been that the initial compliment helps establish the valence of the answer.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
I’ve noticed that when I ask intentionally bad questions, the LLM describes them as “interesting”, not “good”.
Those who use chatbots intelligently will find their intelligence improved somewhat by the exercise; those who use them intelligently will find their intelligence diminished somewhat by the lack thereof.
It’s not the chatbot, but something else entirely.
It’s not the chatbot, but something else entirely.
It’s plausible that these systems will have a homogenizing effect or be used to manipulate opinion.
Also possible that they’ll expand vocabularies and teach some users to think more cautiously or dialectically. (Not sure the median human post is as willing to say “we don’t know” as Claude.)
Also possible that they’ll expand vocabularies and teach some users to think more cautiously or dialectically. (Not sure the median human post is as willing to say “we don’t know” as Claude.)
Max Planck Institute researchers found a plausible link between ChatGPT exposure and increased use of words such as 'underscore,' 'comprehend,' and 'bolster' in the 18 months after ChatGPT's release, while moderators and reporters documented wider AI-like phrasing in public posts.
December 8, 2025 at 2:51 PM
Those who use chatbots intelligently will find their intelligence improved somewhat by the exercise; those who use them intelligently will find their intelligence diminished somewhat by the lack thereof.
It’s not the chatbot, but something else entirely.
It’s not the chatbot, but something else entirely.
I never worry that I’m being too generous; I outsource that worry to my friends and allies. 😇
December 8, 2025 at 2:29 PM
I never worry that I’m being too generous; I outsource that worry to my friends and allies. 😇
I’ve actually started doing this to people 😅
me: why does literally every AI tool glaze me, I'm just asking a question about documentation
friend: That's a really perceptive question - I think it gets at the heart of the current problem with LLMs! If you'd like, I can break down the origins and implications of this particular problem
friend: That's a really perceptive question - I think it gets at the heart of the current problem with LLMs! If you'd like, I can break down the origins and implications of this particular problem
December 8, 2025 at 3:42 AM
I’ve actually started doing this to people 😅
For those who are new around here, I’m dead serious when I say that I think Megalopolis is the most important film of our generation.
Go watch it if you haven’t. 😁
Go watch it if you haven’t. 😁
Movie you’ve watched more than six times with a gif. Hard mode: no Stars (Wars nor Trek), LOTR, or Marvel.
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
December 8, 2025 at 3:23 AM
For those who are new around here, I’m dead serious when I say that I think Megalopolis is the most important film of our generation.
Go watch it if you haven’t. 😁
Go watch it if you haven’t. 😁
Movie you’ve watched more than six times with a gif. Hard mode: no Stars (Wars nor Trek), LOTR, or Marvel.
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
December 8, 2025 at 3:12 AM
Movie you’ve watched more than six times with a gif. Hard mode: no Stars (Wars nor Trek), LOTR, or Marvel.
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
It's not really that hard.
(Seven times. My wife has watched it ten times. I need to catch up.)
Reposted by Siobhán
For reasons completely unrelated to any opinions about regulatory policy, a Day 1 executive order here in the US should be ending every government agency account on X. Just delete them all, plain as that. For the same reason the US government need not have accounts on Stormfront or 4chan.
Pretty crazy that nearly all of the EU Commission is still on X even as Elon Musk is literally talking about destroying the EU.
You can leave X. It's really not that difficult.
You can leave X. It's really not that difficult.
So what's the outcome?
2️⃣5️⃣ of the 27 Commissioners have posted in the last week
2️⃣4️⃣ of the 27 generally post every week (Albuquerque has never been on X, Ribera is sporadic, McGrath has somehow stopped)
0️⃣ have stopped as a result of what Twitter has become since the Musk takeover
2️⃣5️⃣ of the 27 Commissioners have posted in the last week
2️⃣4️⃣ of the 27 generally post every week (Albuquerque has never been on X, Ribera is sporadic, McGrath has somehow stopped)
0️⃣ have stopped as a result of what Twitter has become since the Musk takeover
December 7, 2025 at 11:28 PM
For reasons completely unrelated to any opinions about regulatory policy, a Day 1 executive order here in the US should be ending every government agency account on X. Just delete them all, plain as that. For the same reason the US government need not have accounts on Stormfront or 4chan.
I get why people are dunking on this, and I know all about the paradox of tolerance and Nazi bars etc
But man, it just feels un-strategic to respond to this with vitriol, when in context this person is evidently agreeing with us about this being bad.
But man, it just feels un-strategic to respond to this with vitriol, when in context this person is evidently agreeing with us about this being bad.
December 7, 2025 at 11:21 PM
I get why people are dunking on this, and I know all about the paradox of tolerance and Nazi bars etc
But man, it just feels un-strategic to respond to this with vitriol, when in context this person is evidently agreeing with us about this being bad.
But man, it just feels un-strategic to respond to this with vitriol, when in context this person is evidently agreeing with us about this being bad.
It’s sad to me too, but “still” is doing a lot here – the community actually hasn’t existed all that long yet! The space is wide open because we’re still in the early days, and it takes time to bring an app to market!
Sadly, we must be patient. 😅
Sadly, we must be patient. 😅
even atproto makes me kind of sad now. the most fleshed out thing we have is still a twitter clone. that's sad to me.
the internet doesn't feel fun anymore and i wish it did
December 7, 2025 at 11:01 PM
It’s sad to me too, but “still” is doing a lot here – the community actually hasn’t existed all that long yet! The space is wide open because we’re still in the early days, and it takes time to bring an app to market!
Sadly, we must be patient. 😅
Sadly, we must be patient. 😅
Reposted by Siobhán
i hesitate to say this, but i think karp's autistic supremacist argument about llm use is exactly wrong. there are (non-universal) neurodivergent relationships to vagaries of language and these make the failure modes of LLMs difficult for some to understand or even see.
Posting this after speculation about drug use is like when Kevin Spacey posted that video where he's like "I told you all my secrets" video after the accusations came out.
December 7, 2025 at 10:11 PM
i hesitate to say this, but i think karp's autistic supremacist argument about llm use is exactly wrong. there are (non-universal) neurodivergent relationships to vagaries of language and these make the failure modes of LLMs difficult for some to understand or even see.
“Plagiarizing developers” is a phrase custom-made to send me into an apoplectic fit
December 7, 2025 at 9:22 PM
“Plagiarizing developers” is a phrase custom-made to send me into an apoplectic fit