ZP Rosen

i need the “llms are might be conscious” folx to read this

My thoughts every time Sam Altman speaks: f*ck you Sam. No one needs more SAS bloatware. Least of all from you.

Because if you know how these systems work under the hood all the woo woo business-idiot slop just disappears. Heck, here are all the papers that went into developing them.

This is gross on so many levels, and turning it off for good probably means putting pressure on Grammarly and whatever journal publisher Grammarly bought publication rights from to build this.

Made it up until 1300 without having to relive learning Old English in college. And my OE brain is clearly toast past 1100. Super, super cool!

You rock,thank you!

I'm gonna be in Tucson for #ICLASP19 and if folks know cool hikes out there while I'm in town please show me!

I’d never thought of it like this but oh my god now I can’t unsee it.

Justice is if that head gets revived inside of a soccer ball.

The thought of an eternal Peter Thiel: 🤮

Oooh! Which “book”?

Exams weren't meant to function as gatekeeping devices in the way they do. The intention of an exam is supposed to be to direct student and educator attention to clear gaps in student knowledge. Not to gate keep, but to fill those gaps. And I wonder what other things we can do to accomplish that.

Honestly, I wish Labov was still with us to help us figure out how to do that. We could use some creativity in the same vein as his famous bunny experiment. Maybe some of the answers are actually tucked away in old (countercultural) work like this.
🧵

In any case, I think we're going to need to pick tools from a very different toolkit than the one most of us probably grew up with. Not tech-based tools, but new ways of thinking about how to evaluate student knowledge naturalistically.
🧵

I'm kicking around a number of ideas. Some of them are more workable than others--no more exams + activities that require real world interaction? Students giving oral updates on term-long projects? The possibilities are endless. But tailoring them to any given classroom context is difficult.
🧵

And no. More AI isn't the answer. GTFO-I am not buying your shit subscription with zero accountability and no customer service after the fact.

Exams aren't sustainable... but that might not be all bad. I think we as educators need to think deeply about alternative mechanisms for evaluation.
🧵

year it'll be 160. Up and up and up. And it's not like teacher pay is going to go up commensurate with that.

Honestly, this isn't going to be sustainable. Because the way schools are treated like another economic market makes it unsustainable. More work for less pay = profit.
🧵

I actually think exams might be dead.

I'm grading about 150 bluebook exams right now. And it's striking to me that the reason I'm doing this is because we're now afraid of folks cheating via AI in any other medium.

And... y'all class sizes are only increasing. This quarter its 150 exams. Next 🧵

Model to Meaning: How to Interpret Statistical Models with R and Python Published in Journal of the American Statistical Association (Just accepted, 2026)

"Model to Meaning" just received a super generous review in JASA.

Check it out! (I'm blushing 😊)

Reminder: You can get the paper copy from CRC, and the full **free** version will remain online forever at marginaleffects.com

www.tandfonline.com/doi/full/10....

Chip Berlet, who worked to take on everything from COINTELPRO to neo-Nazis, will be remembered for helping to inform, inspire, challenge and strengthen generations of anti-racist movements on the Left.

We were lucky to speak with him in 2020. itsgoingdown.org/from-souther...

Science is currently deciphering whale language, once we learn enough of it the first words out of our mouths should be something along the lines of; "we're sorry".

Again, just throwing monkey wrenches here. And the Piantadosi/Fedorenko paper is chef’s kiss good if you like formal reasoning papers. :)

LLMs though don’t produce language based on thought or communicative need. The /algorithm/ samples likely continuations based on the prior input. It /samples/ those. It doesn’t “think” about it. It doesn’t choose. It’s cool how well that works, but it’s opposite of how ppl (thinkers) communicate.

Infinitely productive language can arise from chance under communicative pressure Abstract. Human communication is unparalleled in the animal kingdom. The key distinctive feature of our language is productivity: we are able to express an

Here’s a cool monkey wrench for this logic :). Sure. The p of a short sequence existing in L is high (seq length is powerlaw distributed in Lang—see Zipf). But in /humans/ thought must have evolutionarily proceeded Lang.

Washington Post is shuttering its book section, part of the ongoing war against literacy & knowledge.

We at Current Affairs want to pick up the slack by expanding book reviews. Writers: pitch us anytime! And if you subscribe/donate we'll use the funding to commission book coverage.

Gonna do the exact opposite. @kirbyconrod.bsky.social is a fucking badass linguist and if you want to read some cool work on pronoun usage and gender you should go grab any one of these articles and give 'em a read. scholar.google.com/citations?us...

get to have that same level of support and belief as well.

(1) exceptionally lucky, and (2) that I'm going to be living with a debt to pay back every ounce of supportiveness I get from my folks for the rest of my life. I'm only here because I have a community that believes in me, and it's absolutely my duty to do whatever the hell I can to make sure others

I defended my prospectus yesterday and am now officially ABD :).

I don't deserve that karma I have, but I recognize that I am extremely lucky to have advisors and collaborators who are incredibly supportive. Not every PhD student has that, and it's not lost on me at all that I'm

Call your reps today.

Nullius addictus iurare in verba magistri
~ Horace

Seems pertinent to remember (and rally around) at this point in time.