ZP Rosen
@zaqdelinguist.bsky.social
Computational (socio)linguist interested in how we co-negotiate meaning during conversation.
Because I’ve been working with transformers since 2018, very not into AI hype.
https://zaqari.github.io
(He/they)
Because I’ve been working with transformers since 2018, very not into AI hype.
https://zaqari.github.io
(He/they)
Pinned
ZP Rosen
@zaqdelinguist.bsky.social
· Aug 9
The way conversation partners work together to generate new semantic information explains how conversation facilitates distributed cognition
Humans are (often) better when we work together. To do this, people rely on conversation as a way of proposing new semantic content/ideas and modifying old sema
doi.org
New preprint live while the paper is in review! The way speakers and their partners (1) generate new info relative to prior convo states and (2) appear to coordinate how much new information each person/partner generates is functionally capable of "universal" computation (i.e. EXTENDED cognition)
If folks can’t [afford to] eat, they shouldn’t be able to trade.
BREAKING: The U.S. stock market hits an all-time high as expectations rise for companies to keep making more money.
US stocks rise to a record on expectations for companies to make even fatter profits
The U.S. stock market hit an all-time high as expectations rise for companies to keep making more money.
bit.ly
October 6, 2026 at 8:31 PM
If folks can’t [afford to] eat, they shouldn’t be able to trade.
Don’t know if anyone who sees my posts aren’t already listeners, but I freaking love @lingthusiasm.com 🤓
Hey. The fragmentation of the social media landscape has been hard on indie #scicomm 🧪 projects
So if you'd like to follow a podcast that's enthusiastic about #linguistics, could you check out @lingthusiasm.com?
And if you think your followers might like to, could you give this a repost?
So if you'd like to follow a podcast that's enthusiastic about #linguistics, could you check out @lingthusiasm.com?
And if you think your followers might like to, could you give this a repost?
October 6, 2026 at 3:49 AM
Don’t know if anyone who sees my posts aren’t already listeners, but I freaking love @lingthusiasm.com 🤓
And what’s coming is the total collapse of our company under the weight of our debt and utility as a shell for NVIDIA’s circular finance scheme🫡
“‘Mark, we have two minutes left. I want to be respectful, but I’d like to move on and talk a little bit about the future and what’s coming,’ the publicist said.”
OpenAI PR tells journalist to ‘move on’ while asking Sam Altman about a ChatGPT user’s suicide
“I’d like to move on and talk a little bit about the future,” the publicist said.
www.theverge.com
October 5, 2026 at 6:30 PM
And what’s coming is the total collapse of our company under the weight of our debt and utility as a shell for NVIDIA’s circular finance scheme🫡
(1) I love this, and (2) it makes me think about this awesome piece by @abeba.blacksky.app that I come back to often :)
September 28, 2026 at 4:09 PM
(1) I love this, and (2) it makes me think about this awesome piece by @abeba.blacksky.app that I come back to often :)
If folks are looking for a policy suggestion that would actually change shit…
I feel like if the companies had to accept even the tiniest shred of liability for the things their available-to-anyone software does, it would stop going "rogue" really quick
September 26, 2026 at 6:00 PM
If folks are looking for a policy suggestion that would actually change shit…
Listen, these things can’t go “rogue”. It’s just a basic LLM stuck in a while loop. But this kind of attack is possible by doing something as simple as that, and that means that more people can set up programs to do terrible things like this moving forward on a whim.
"With the Education Department, OpenAI’s technology tried to hack the website to gather data from the department’s civil rights office" 👀 www.nytimes.com/2026/09/25/t...
OpenAI’s A.I. Went Rogue and Meddled With U.S. Government Websites (Gift Article)
The company did not learn until recently that its technology had meddled with websites for the Education and Commerce Departments and the Securities and Exchange Commission.
www.nytimes.com
September 26, 2026 at 2:32 PM
Listen, these things can’t go “rogue”. It’s just a basic LLM stuck in a while loop. But this kind of attack is possible by doing something as simple as that, and that means that more people can set up programs to do terrible things like this moving forward on a whim.
Modern language modeling has devolved into big companies competing for who can do the campiest spin-off of 1980s sci-fi movies. I genuinely miss the old days where folks were more interested in building models for their specific tasks.
September 22, 2026 at 5:01 PM
Modern language modeling has devolved into big companies competing for who can do the campiest spin-off of 1980s sci-fi movies. I genuinely miss the old days where folks were more interested in building models for their specific tasks.
Folks need a listing of which airports are putting this in so they can fly into the closest one where the admin still care about whether or not passengers land safely in unpredictable conditions. Especially since we're accelerating towards global warming...
…jfgdic.
Like, I've talked about how bad an idea it is to place "AI" in administrative positions, and air traffic especially. Where was the review period on this? How is every congresscritter currently clamoring for "AI Pacing" still letting THIS go through??
(Original image reposted w/ alt text)
Like, I've talked about how bad an idea it is to place "AI" in administrative positions, and air traffic especially. Where was the review period on this? How is every congresscritter currently clamoring for "AI Pacing" still letting THIS go through??
(Original image reposted w/ alt text)
September 21, 2026 at 2:15 AM
Folks need a listing of which airports are putting this in so they can fly into the closest one where the admin still care about whether or not passengers land safely in unpredictable conditions. Especially since we're accelerating towards global warming...
To the guy who said he’d burn all his equity in AI companies for a “1% extra chance of survival.” Okay my dude. Sell your stocks at a dollar. $1. That’ll tank the entire industry right there. Put your literal money where your mouth is.
September 18, 2026 at 11:13 PM
To the guy who said he’d burn all his equity in AI companies for a “1% extra chance of survival.” Okay my dude. Sell your stocks at a dollar. $1. That’ll tank the entire industry right there. Put your literal money where your mouth is.
So, bought new limbs for my recurve. Am now at 60lbs and whooo boy. I can sling an arrow fast with em, but man are my shoulders beat from it…
September 16, 2026 at 11:31 PM
So, bought new limbs for my recurve. Am now at 60lbs and whooo boy. I can sling an arrow fast with em, but man are my shoulders beat from it…
This 👇 bc that liberates men too.
Should we respect that they want to be away from sexual harassment and other nonsense? Absolutely. But maybe we can go further and deradicalise men.
September 12, 2026 at 8:52 PM
This 👇 bc that liberates men too.
Reposted by ZP Rosen
Should we respect that they want to be away from sexual harassment and other nonsense? Absolutely. But maybe we can go further and deradicalise men.
September 12, 2026 at 7:06 PM
Should we respect that they want to be away from sexual harassment and other nonsense? Absolutely. But maybe we can go further and deradicalise men.
Because the only thing AI supercharges for universities is ROI in university boards’ stock portfolios…
September 10, 2026 at 1:38 PM
Because the only thing AI supercharges for universities is ROI in university boards’ stock portfolios…
Some rando complemented me yesterday at the archery range on my form (I shoot traditional/barebow at ~60lbs, so form is my everything) including the cant of my bow when I'm aiming. Not gonna lie, I'm gonna ride that compliment all week.
September 7, 2026 at 8:30 PM
Some rando complemented me yesterday at the archery range on my form (I shoot traditional/barebow at ~60lbs, so form is my everything) including the cant of my bow when I'm aiming. Not gonna lie, I'm gonna ride that compliment all week.
"The train has left the station" in AI "discourse" is a thought terminating cliche.
September 4, 2026 at 2:45 PM
"The train has left the station" in AI "discourse" is a thought terminating cliche.
I’d support building this if the angel at the top is replaced with Trump in a mumu holding a plunger.
Interior Secretary Doug Burgum says they're starting excavation work on Trump's arch
September 4, 2026 at 12:55 AM
I’d support building this if the angel at the top is replaced with Trump in a mumu holding a plunger.
Admin: “minimal ai REQUIRED”
All of us: Okay. **proceeds to make assignments only possible to complete via offline interactions**
All of us: Okay. **proceeds to make assignments only possible to complete via offline interactions**
Even though my teaching team wants to classify all assignments as "minimal AI permitted" I've just made mine "AI prohibited" with possibly the last shred of academic autonomy I have FIGHT ME
September 1, 2026 at 8:08 PM
Admin: “minimal ai REQUIRED”
All of us: Okay. **proceeds to make assignments only possible to complete via offline interactions**
All of us: Okay. **proceeds to make assignments only possible to complete via offline interactions**
This is a fantastic point. I've talked to some folks about this recently and I described it as "just manipulating the posterior distribution on tokens to be more constrained... regardless of whether it generates a useful response or not."
This is a much better way to explain this, though.
This is a much better way to explain this, though.
Sorry, one more thread.
This OpenAI/HuggingFace incident highlights the danger of building computers that compute with natural language. 1/x
This OpenAI/HuggingFace incident highlights the danger of building computers that compute with natural language. 1/x
The overreaction to this Hugging Face thing is driving me nuts. Come flay me for being wrong but... 1/x
August 31, 2026 at 12:09 AM
This is a fantastic point. I've talked to some folks about this recently and I described it as "just manipulating the posterior distribution on tokens to be more constrained... regardless of whether it generates a useful response or not."
This is a much better way to explain this, though.
This is a much better way to explain this, though.
I have a plan that would work. Every time someone makes a sweeping prediction with the term AI in it, if that prediction doesn’t pan out you can sue them. Cut it off at the hype.
"If someone had a credible plan for slowing down AI advances globally, I would likely support it." www.gatesnotes.com/a-turbulent-...
The choices we make about AI now are critical | Bill Gates
AI will either be the greatest equalizer ever invented, or the worst source of injustice. We need to start planning now so it makes the world a fairer place.
www.gatesnotes.com
August 26, 2026 at 11:51 PM
I have a plan that would work. Every time someone makes a sweeping prediction with the term AI in it, if that prediction doesn’t pan out you can sue them. Cut it off at the hype.
Reposted by ZP Rosen
We released our new Executive Paywatch report, and surprise surprise, corporate greed is out of control.
Last year, S&P 500 CEOs gave themselves a 21% raise and now make 312x what median workers do. Meanwhile, workers are struggling to make ends meet.
Last year, S&P 500 CEOs gave themselves a 21% raise and now make 312x what median workers do. Meanwhile, workers are struggling to make ends meet.
August 14, 2026 at 5:07 PM
We released our new Executive Paywatch report, and surprise surprise, corporate greed is out of control.
Last year, S&P 500 CEOs gave themselves a 21% raise and now make 312x what median workers do. Meanwhile, workers are struggling to make ends meet.
Last year, S&P 500 CEOs gave themselves a 21% raise and now make 312x what median workers do. Meanwhile, workers are struggling to make ends meet.
The point of conversation isn’t to end up on the same page as your partner but to be just uncoupled/different enough to work together to come up with new sh*t to talk about while feeling good about it. In other words: go be you, fam.
August 14, 2026 at 3:19 AM
The point of conversation isn’t to end up on the same page as your partner but to be just uncoupled/different enough to work together to come up with new sh*t to talk about while feeling good about it. In other words: go be you, fam.
Reposted by ZP Rosen
New preprint live while the paper is in review! The way speakers and their partners (1) generate new info relative to prior convo states and (2) appear to coordinate how much new information each person/partner generates is functionally capable of "universal" computation (i.e. EXTENDED cognition)
The way conversation partners work together to generate new semantic information explains how conversation facilitates distributed cognition
Humans are (often) better when we work together. To do this, people rely on conversation as a way of proposing new semantic content/ideas and modifying old sema
doi.org
August 9, 2026 at 9:26 PM
New preprint live while the paper is in review! The way speakers and their partners (1) generate new info relative to prior convo states and (2) appear to coordinate how much new information each person/partner generates is functionally capable of "universal" computation (i.e. EXTENDED cognition)
New preprint live while the paper is in review! The way speakers and their partners (1) generate new info relative to prior convo states and (2) appear to coordinate how much new information each person/partner generates is functionally capable of "universal" computation (i.e. EXTENDED cognition)
The way conversation partners work together to generate new semantic information explains how conversation facilitates distributed cognition
Humans are (often) better when we work together. To do this, people rely on conversation as a way of proposing new semantic content/ideas and modifying old sema
doi.org
August 9, 2026 at 9:26 PM
New preprint live while the paper is in review! The way speakers and their partners (1) generate new info relative to prior convo states and (2) appear to coordinate how much new information each person/partner generates is functionally capable of "universal" computation (i.e. EXTENDED cognition)
One of my favorite things about the speech/transcript this came from, too, is that Shannon gave this as the keynote speaker at the very first conference that would become IEEE.
Yes, I touch on this I think a bit here: doi.org/10.1007/s421...
![Theoreticians exist that could be seen as doing a good job of attempting to buttress their theory and related entities against misuse, e.g. in the case of information theory:
attracted by the fanfare and by the new avenues opened to scientific analysis, [practitioners] are using these ideas in their own problems. Applications are being made to biology, psychology, linguistics, fundamental physics, economics, the theory of organization, and many others. [However,] the establishing of such applications is not a trivial matter of translating words to a new domain, but rather the slow tedious process of hypothesis [testing.]
Shannon , 1956, p. 3](https://cdn.bsky.app/img/feed_thumbnail/plain/did:plc:vsyj5jtayxcq5dw5mtja4xtb/bafkreieropu63rbibcwj4ewut3ldmes7hmcfwur3gmustix645jrqxchfq)
August 8, 2026 at 6:23 PM
One of my favorite things about the speech/transcript this came from, too, is that Shannon gave this as the keynote speaker at the very first conference that would become IEEE.