Generative AI changes what things mean because it focuses on words, not the meaning. It adds words, changes meaning, and makes "plain language" that is not accessible. We know, because we tested it. Generative AI has no place in plain language. autisticadvocacy.org/2025/07/asan...

as a girl with a PhD in natural language processing and machine learning it's actually offensive to me when you say "we don't know how LLMs work so they might be conscious" I didn't spend 10 years in mines of academia to be told ignorance is morally equal knowledge. We know exactly how LLMs work.

"If it is inaccessible to the [marginalized/minoritized/oppressed], it is neither radical nor revolutionary."

!!!! @annleckie.com’s Ancillary series id one of my favorite SciFi series, period. Heck. I’ve loved all of their books :)))

I’m so fed up of seeing Ai slapped on everything! I spent 15 years building and running an Ai fraud detection company LLMs are NOT Ai 🤖 not even remotely related and as for Sam Altmans claim that they don’t fully understand how their system works… I’m happy to explain it to him using crayons 🖍️

I’m curious: I have a hunch about the information theoretic reasons behind why LLM outputs might trigger the positive assessment that it does in some users/what it is about the model that exhibits “sycophantic” behavior. Anyone we know have access to the transcripts openAI released in court?

Remember kids, the next time he forgets his login credentials it’s CENSORSHIP BY thE TERRible LeFt

I’m not convinced the brain worm is dead.

This. Not all peer reviewers treat authors with respect, but a majority of them do. And I will 100% own up to the fact that peer review has not only shored up my writing and occasionally my methods, but even got me to think about extensions and improvements for subsequent work.

I made the mistake of devoting myself to AI4SG early in my career because I wanted to believe I could use my skills to make the world a better place. Instead, I found exactly what Dr. Birhane describes - particularly with respect to accountability laundering. I'll share one example🧵

Exactly. You show up to these things and it’s basically a sales pitch/product demo.

Freaking wish the UC board of regents would read this . . .

Every time I see the word heteroscedasticity I have the urge to shout "hetero-SKIBIDI TOILET". As a computational linguist who studies discourse on social media, I feel like this is both a major problem, and probably explanatory. All my data is hetero-skibidi toilet all of the time.

I said it before and I’ll say it again. As someone who has coached national champs at debate and who competed in it myself, I think this is an absolutely shit assessment of student communication that will have strong classist and ableist biases. Khan is not a communication expert. Period.

A little spot of sunshine. I've been giving little stashes of dog treats to a crow that I know is the bottom of the pecking order for the crows in my neighborhood. It's now hopping around following me and my dog when we go for walks. We're a little trio of misfits, the three of us.

For those wondering what the hell mitochondria have to do with anything…they are currently the focus of A Lot of quackery, because they are the focus of a lot of Really Interesting Science Research. Quacks sound like experts to non-experts because the quacks are deliberately copying the experts.

When I think about the fact that universities are ALSO supposed to be repositories of knowledge and expertise you can't get elsewhere (and not just job training or even educational enterprises) I immediately go to ancient languages. I don't think ppl realize you can just lose them

I upgraded to @pycharm.dev 2025.2.0.1 and y’all… it runs like it was vibe coded. Ever wondered what it feels like to run big data research 3x slower because you’re constantly trouble shooting the IDE? Well now you can experience it yourself! Trying out 2025.2.1 but I’m about to utterly abandon it.

I really wanted to do other types of research, but we really have to fix all of AI's damage to reasoning and practice first, I guess, and then fix sexism and whatever else before we can even do new work on critical computationalism in peace because white supremacy and capitalism had to end the world

"Don't look away. Don't stay silent. If you do, don't be surprised if it's your city next." -- FUCK YOU 1000 times over.

You know, as someone who leans a lot on Reddit data for research, that’s actually a pretty cool idea 🤔

Offsetting pain for who? Working class people are not seeing the cash that corporations are spending on AI

i have a lot of thoughts about AI and LLMs and they are quite complex but a real fucking loud one is it is not my job to make a market for this technology or contribute to its optimisation barring an eight figure check delivered ditectly to my credit union account

CHICAGO! Yidden especially! MONDAY! TURN OUT! PLEASE SPREAD THE WORD!! Please get this out to all your networks!!

What kills me about this is that not only is this a very “businessman” way of dealing with numbers you don’t like, but heads up! Ignoring bad numbers that people don’t like is why the insurance industry is crashing too. So this is just more of the same stupid decision making process.

Extremely important and interesting study.

In the airport en route to CogSci 2025 (@cogscisociety.bsky.social). Looking forward to meeting up with folks if you’re around and wanna say hi! I’d be stoked to chat with folks 🤓

Legitimate class they offer at Orange Coast College: history of rock and roll 🎸🤘

Today!

Hidaya Al-Motawaq's son is a year and a half old and weighs less than 10 pounds. Doctors warn of permanent damage to children's health due to chronic malnutrition from Israel's earlier blockade.

Uncritical adoption of AI “undermines our basic pedagogical values and principles of scientific integrity. It prevents us from maintaining our standards of independence & transparency. And most concerning, AI use … hinder[s] learning and deskill critical thought.” openletter.earth/open-letter-...

This level of barbarity is unconscionable. Nobody should be comfortable with children and families starving to death. It is morally indefensible to allow the manufactured and man-made starvation to continue. The United States must stop funding and enabling this genocide.

Sharing this as someone who cares deeply about #pedagogy. Pedagogy is my protest in a world where human rights abuses rein supreme, and I’ll be f*cked before I give that over to Silicon Valley to be squashed in lieu of often inaccurate and always uncritical bullsh*t. #academia

There is no such thing as an "AI" (or any other technology) without biases and values in it, but there very definitely is such a thing as presenting one's own biases and values as "neutral" and then forcing people to build "AI" which conforms to *those.*

Academics have got to figure out a new way of organizing academic review and publication that isn’t the academic press. Even before LLMs, they were clearly not on the side of academics.

“Authors have until the end of August to opt out of the licensing agreement. If they do not, their work will be used to help train AI models.”

I just sent this out. The Jewish people have moral frameworks for this moment. It is a choice not to heed them. Please let us meet this moment, let us see the shattered Temple around us, right now, and our responsibility to it.

Every sentence of this story is heartbreaking. "After four months of a near-total Israeli siege, Gaza’s few remaining hospitals now have wards for the growing number of malnourished children whose tiny bodies are just the width of their bones." www.washingtonpost.com/world/2025/0...

So I actually really enjoy target archery. Just cause it’s low stakes and relaxing. I just bought new fletchings for my beat to hell arrows, and I’m so looking forward to working on them just because maintenance of my gear is also a focus intensive, low-stakes task :). Things my broken 🧠 needs.

On a lark I applied to an engineering role at a startup. One question they had on the app was “how has AI changed your field?” Answer: I’m a computational linguist. It f*cking hasn’t. People have been trying to treat all problems as seq-2-seq translation since Bahdanau attention. It’s the same idea.

So I had a (meager) speech and debate scholarship in college. It kept Ramen in the microwave. I then got my first professor gig partially teaching debate. Debate always devolves into a kind of structure/procedural game…

so, when I was doing trust and safety work, way back in 2023, there was already data from pre-LLM chatbots that men use AI to practice abusive behavior and construct authoritarian identities that they then inflict on human women (and children)

I hope you didn’t post on social media that a family member passed away recently, because Delta’s cookies and their new AI pricing are gonna scalp you buying tickets home to say goodbye.

Going to give a guest talk to for an Ed class, and that second paragraph sums up the whole theme of what I'll be talking about. Thanks for the good intro blurb.

Trying to get people to understand this feels like screaming into the void lately 😭

Continuing a thread with @vortexegg.com I have some really cool lit from a mix of cognitive linguistics and communication studies on how we language affects the way we align on shared understanding. Because, of course linguists have gone down this rabbit hole at some point!

Here’s something that kills me about education in the US and stats. This statement right here? This applies to all statistical models. All of them. And this is the first thing they should be teaching you in school 🧵

Sure. But quitting /companies obsessed with LLMs/ is like a carpenter quitting a job site because /the manager owns/ the table saw and requires all the carpenters to put their fingers directly on the blade when they use it.

Straight, I worked at a company building chatbots for other companies for a bit. And I sat in meetings where leadership for insurance companies would talk about it being too easy for a customer to reach an actual service person. They wanted it to be hard…