Rui Chaves

ICE detains UB researcher after his attempted return from Canada A University at Buffalo medical researcher was taken into custody after attempting to re-enter the United States earlier this month.

Can we improve the road signs near border accesses? People make a wrong turn too often, and now there's a steeper price to pay:
www.wgrz.com/article/news...

All code and more plots and regressions at: github.com/RuiPChaves/L...

Number of Linguistics job postings have dropped significantly in the Linguist List. Please note that the numbers reported below are inflated because some job postings mention multiple areas of study (e.g. phonology + phonetics, or computational linguistics + semantics).

Putting my Asimov hat on, maybe one day tutor systems will be very good, at least for some people. But I can’t imagine being a kid and enjoying learning from a screen rather than from a person, in a spontaneous classroom environment. Most of us are social creatures.

The deeper the model goes into a topic the worse it gets (plain wrong or hallucinates refs), and I suspect it depends a lot on the field. Including when it is asked to create exercises. For basic things probably fine, but how can a student tell when the model strays from the truth?

420k malicious tokens (0.00016% of the total training tokens) were enough to ruin the model, even after fine-tuning. This means the public can attack these models by creating and uploading bad data to Github, Stackoverflow, blogs, etc...

Yet another example of how LLMs are not robust (a key argument against symbolic AI): a small fraction of 'bad' data can compromise models regardless of dataset or model size. Data poisoning does not scale with model size. arxiv.org/pdf/2510.07192

YouTube video by minutephysics The Physics Of Dissonance

www.youtube.com/watch?v=tCsl...

At Starbucks I order a Short coffee. I suspect most people don’t even know that exists…

Mechanisms vs. Outcomes: Probing for Syntax Fails to Explain Performance on Targeted Syntactic Evaluations Large Language Models (LLMs) exhibit a robust mastery of syntax when processing and generating text. While this suggests internalized understanding of hierarchical syntax and dependency relations, the...

There is mounting evidence that LLMs rely heavily on brute-force and shallow heuristics to achieve high accuracy, despite some studies (though not all) finding evidence for sophisticated latent representations. Add another to the bunch:
www.arxiv.org/abs/2506.16678

Vehicles These imaginative thought experiments are the inventions of one of the world's eminent brain researchers. They are "vehicles," a series of hypothetical, self...

Coolest? It's a tie between Braitenberg's "Vehicles"
and Hofstadter "GEB".

He has never read a book in his life.

I feel sick.

Two HIV vaccine trials show proof of concept for pathway to broadly neutralizing antibodies Learn more about North American and African clinical trials led by Scripps Research, IAVI and collaborators that show early immune activation using mRNA vaccine technology—marking progress toward an e...

We even got real close to curing HIV: www.iavi.org/press-releas.... But this uses mrna, and that has come to be a dirty word for these people. It shouldn't. It's a triumph of modern molecular biology.

Wow, yes. An amazing glimpse of the LLM Lie Machine at work. Anyone whose students are bring tempted by LLMs, sharing this could be very helpful showing how it lies, and how it fails.

I'm so sorry. Beyond words.

Science Is Fraught Ignorance can't govern and reality isn't negotiable

"So here’s one place my compass is pointing me to: pushing back against bullshit. Frequently. Publicly. This frightens me because I have enough to deal with...I’m a people pleaser. I love being loved. But we are in terrifying times and all I have is my compass." - @jimmycoan.bsky.social

Bresnan (1991) journals.linguisticsociety.org/proceedings/...

Register now for the LSA Linguistic Institute in Eugene, OR: over 80 courses over two 2.5-week terms (July 7-22::July 24-August 8). I'm teaching Constructions & the Grammar of Context with E Francis in session 1! center.uoregon.edu/LSA/2025/

Interesting! I've no problem with that. This is crafted as a cartoon, so it is read top to bottom.

Thank you for doing this work, it is very interesting! Challenge: I suspect if you asked o1 to create an X-bar grammar sufficiently large to parse a couple of pages of English it would not do a good job
.

I've annotated the o1 trees here: drive.google.com/file/d/1we0S... I think these systems *are* amazing, and getting better and better, but have a long way to go. Major issues: inconsistent analyses (suggesting not a single grammar exists), wrong analysis, partial trees, and gibberish trees (only 2).

O1's results are impressive, but they are not as good as a professional human annotator, at least for these sentences.

This is a good one but the other outputs form o1, are not good, and often mutually inconsistent. E.g. in the "This town's only road's potholes' depths are driving me insane" example the VP is too flat, the "only road" NP is wrong, inconsistent with tree for "a medium-sized sturdy lamp" in next S

Maybe they are assuming AI will bring about those breakthroughs? What happens when that bubble eventually pops?

ps - Should have used NDL instead of HOL.

The LL mostly consists of US posts. Here are the 2024 posts for "Professor":

That's a great idea. Maybe mine ProQuest, LingBuzz, and similar venues?

Keep in mind that there are hardly any CL posts that don't mention another discipline, so that blue bar mostly consists of posts from a few of the other bars. Similarly, some posts mention both phonetics and phonology, or syntax and semantics, etc. The raw post count is even smaller.