Evidence Against Syntactic Encapsulation in Large Language Models Transformer-based large language models (LLMs) have recently demonstrated exceptional performance in a variety of linguistic tasks. LLMs primarily combine information across words in a sentence using...

Evidence against syntactic encapsulation in large language models

onlinelibrary.wiley.com/doi/10.1111/...

How Transformers Work: A Detailed, Conceptual Explanation (No Coding / Math) YouTube video by IbanDlank

Idan Blank (UCLA, psych) makes the complex intuitive
if you want to learn how LLMs work, watch👇
newly posted to YouTube (no ads)
www.youtube.com/watch?v=cGMn...

Day before month before year??? Blasphemy! I rename my students’ files to end with yyyymmdd so that the sorting makes sense.

Looking forward to #CogSci2025 ! Find us throughout the conference

📣 I'm looking for a postdoc to join my lab at NYU! Come work with me on a principled, theory-driven approach to studying language, learning, and reasoning, in humans and AI agents.
Apply here: apply.interfolio.com/170656
And come chat with me at #CogSci2025 if interested!

Yes, and I replied to the specific part that doesn’t resonate with me.

I don’t think banning a super helpful technology in classrooms is useful. For me, it’s better to create assignments that disincentivize its use (under critical views on AI, this should be easy) and actively teach students to use it critically (when & why it’s useful vs. bad).

Thrilled to announce our symposium, Cognitively Inspired Interpretability inn Large Neural Networks, at #CogSci2025 featuring @taylorwwebb.bsky.social, Ellie Pavlick, Jiahai Feng, Gustaw Opielka, ‪‪@claires012345.bsky.social‬, and Idan Blank!

I think there is an audience for cranky. So many good stand up comedians are cranky! And maybe having a beer or two at a bar makes people more receptive to that?

You have such interesting data drama stories! They’re like detective novels.

Yes, in class would probably be even better! For me that would mean a flipped classroom and I don’t trust my students to watch lectures at home 😳 but if you can make that happen, that would be awesome! Happy to share my materials for setting up pods, etc.

In some of the meetings they are tasked with using AI in structured ways (e.g., as tutors) and reflect on strengths and limitations. Hopefully they develop some critical thinking about this tech.

Change the assignment structure! I have students discuss readings in “Learning Pods” and submit video recordings of those meetings (faces + screen sharing) - it discourages using AI to do your work because I can detect that so easily in the videos. And it teaches collaboration / communication.

Explaining “viewpoint diversity”, a senior faculty told me—a gay person—that if we don’t hire people who oppose living near gay people, we’ll never hire from countries where most people oppose gay rights, and that’s discrimination. How am I supposed to handle that?

Event Knowledge in Large Language Models: The Gap Between the Impossible and the Unlikely Word co-occurrence patterns in language corpora contain a surprising amount of conceptual knowledge. Large language models (LLMs), trained to predict words in context, leverage these patterns to achi....

We only looked at hidden representation, not perplexities, so I don’t know! (btw we didn’t use models trained on RLHF because that, to us, is non-linguistic training, which was outside the domain of our research question). But take a look at this:

He also imputed every lexical feature in the literature for ~every word in English. Which is absolutely amazing. But, again, I need someone who can figure out the pipeline (if I can even find all his code) to write it up. 2/2

He wrote a beautiful mega-paper on it that I can’t finish on my own because it was his area of expertise. If you know anyone who’s interested in carrying it to completion, I’d LOVE to have Bryor’s creation out in the world! 1/2

Thank you, Jamie. Yes, he was so kind to everyone, and I think he would have become a leader in comp psycholing. He passed in 2023. Still hard to believe. I wouldn’t have a lab if it weren’t for him.

Do Large Language Models know who did what to whom? Large Language Models (LLMs) are commonly criticized for not understanding language. However, many critiques focus on cognitive abilities that, in humans, are distinct from language processing. Here, ...

Does training LLMs on word prediction result in human-like representation of thematic roles?

Work by @josephdenning.bsky.social, Hannah Guo, and Bryor Snefjella who is no longer with us

Indeed shocking, and infuriating. I’m sorry. One of your papers is an assigned reading in my class. We’ll make the youths understand the value of this work!

1/x My NIH grant that would collect data to help *prevent suicidal thoughts and behaviors* among LGBTQ+ people was terminated today, along with many other LGBTQ+ focused grants. These terminations make no sense scientifically. Here's why:

Excited to introduce funROI: A Python package for functional ROI analyses of fMRI data!

funroi.readthedocs.io/en/latest/

#fMRI #Neuroimaging #Python #OpenScience

Work w @neuranna.bsky.social

🧵👇

New preprint “Monkey See, Model Knew: LLMs accurately predict visual responses in humans AND NHPs”
Led by Colin Conwell with @emaliemcmahon.bsky.social Akshay Jagadeesh, Kasper Vinken @amrahs-inolas.bsky.social @jacob-prince.bsky.social George Alvarez @taliakonkle.bsky.social & Marge Livingstone 1/n

People talk a lot about objects, but what about the softness of a cushion, the greenness of an emerald, or the viscosity of oil? In our work just published @pnas.org, we shed light on how we make sense of the hundreds of materials around us.
www.pnas.org/doi/10.1073/...

It's the *anti-DEI* drive now that has outsiders acting "like many faculty members' direct supervisor and telling them how and what to teach."

DEI officials never reviewed syllabi or challenged course titles or targeted entire fields of scholarship. That's what the anti-DEI forces do.

Submit your discourse studies to this SI of Cortex. I’m co-editing with the fabulous @briellestark.bsky.social and @andreaeyleen.bsky.social. May 1 deadline.

Come work with me! We're hiring a 3-year Visiting Assistant Prof in Semantics, bonus if you can teach courses in other areas too. We're a small liberal arts college just outside Philly with a robust undergrad Ling dept. Full info at
apply.interfolio.com/160957

Feel free to ping me with any Qs!
🐦🐦
🧪

Somewhat related is this review piece by @coryshain.bsky.social & @evfedorenko.bsky.social, and a more recent fMRI paper distinguishing “linguistic” WM from domain-general WM (the latter includes “verbal” WM).

journals.sagepub.com/doi/full/10....

coryshain.github.io/assets/pdf/s...

A logo of the National Institute of Mental Health

Interested in a researcher position studying the neural and computational bases of adaptive and maladaptive decision-making, and their relationship to mental health? Our colleague Silvia Lopez-Guzman is hiring a Data Scientist/Data Analyst, read more at cdn-lab.github.io .

aerial view of the Madison isthmus - evolang fifteen - Madison, Wisconsin May 18-21, 2024

Registration for #Evolang2024 is now open! And in addition to a great regular program, we have 7 amazing workshops! evolang2024.github.io