Why I study vector fields

Back from Seoul. My first paper, "An Isotropic Approach to Efficient UQ with Gradient Norms", got a poster and an oral at ProbML, and came away with the Best Paper Award. Still a bit stunned. arxiv.org/abs/2603.29466

🏆The Computational Linguistics High Impact Paper Award recognises papers published in Computational Linguistics within the past 3–5 years that have already demonstrated exceptional scholarly impact. Congratulations to the authors of doi.org/10.1162/coli... #NLProc #CLjournal @chrmanning.bsky.social

if you think students will be routinely collaborating with LLMs after graduation, #1 priority is to study rhetoric because you need to recognize BS and critique the rigor of an argument that *looks* good. the elite users are going to be like, philosophy, classics, and english majors.

The NEMI crew are entering the boat parade at Sail Boston while we wait for the mech interp abstracts to sail in on August 1

NEMI is exactly one month away! Get your abstracts in by August 1st

This interview with him is quite interesting, and it describes his own accommodations as a disabled academic. I feel genuinely bad for him, because it's clear that he wanted his students safe and supported, and then this happens.

OH MY GOD EVERYBODY I just learned about reciprocard.com , which lets you enter your public library and find out all the places you can get a card on Libby. I had not known my Minuteman card actually gave me *every network in the state*, including the flawlessly-named CLAMS?!?!

My coauthor Jay Falk's postdoctoral fellowship was terminated by the Trump administration for documenting the hummingbird example in this paper.

That's a wrap! We hope you had a fun, productive, and idea-filled Gangnam-styled #ICML2026! It's been a pleasure posting with you -- stay tuned for more content. We wish you a safe trip home, and see you next year!

feels like a real landmark of impunity that major corporations are making federal crimes part of the interview process

Quick appreciation thread for Eugenides’ The Virgin Suicides. Theres a reason weird girls love this book. Eugenides wrote a book all about what’s it’s like to be a girl, and he did it without directly writing about what it’s like to be a girl.

this is wild — the openai math proof this week could’ve been solved in the 1980s but it goes against human sensibilities, all it took was just more grit on dumb solutions

In the early stages of drafting a manuscript, it can be helpful to draw a little mascot (mousecot?) to cheer you on.

Our work got featured in Science News! What a great end to an amazing #ICML2026🇰🇷 Peer review is at a crossroads. How we handle the submission and review crisis over the next years will shape the future of scientific publishing. True impact will come from people writing fewer, more meaningful papers!

ChatGPT just told me helpfully in an explanation that the ceiling for something was lower than its floor and it instantly changed my belief that LLMs can apply systemic metaphors well without grounding.

We appreciate the love ❤️ But limited engagement and few followers on Bsky makes some (not us!) question whether it's worth it Be sure to follow (and share with your ML friends!) if you want to see more updates from ICML!

I will literally follow any and all ML PhD students on this website, where you at

Day 4 of #ICML2026 brings a full slate of #KempnerInstitute workshop presentations, featuring research on learning dynamics, mechanistic interpretability, weight-space symmetries, and more. Check out Friday's lineup! #AI

Are students embarrassed by AI cheating? Like, there has always been rare but unstigmatized dishonesty (memorizing a frat’s archive of finals questions) and common but stigmatized (saying you’ve started when you definitely have not started it). AI cheating should be the most stigmatized. Is it?

Day 2 of #ICML2026 is packed with #KempnerInstitute presentations on diffusion models, LLM reasoning, world models for robotics, and much more. Don't miss Wednesday's sessions! #AI

Kicking off Day 1 of #ICML2026 with presentations from #KempnerInstitute researchers on emergent capabilities, actionable #interpretability, and the importance of learning dynamics for a science of #AI. Check out the lineup!

I have always believed that ideas from basic systems neuroscience could eventually help people. But for years that felt more like hope than reality. But look! This is one of the most personally satisfying papers my lab has produced because we’re getting there. www.biorxiv.org/content/10.6... 🧪🧵 1/

🏆Outstanding Paper is awarded to the following research papers: The Flexibility Trap: Rethinking the Value of Arbitrary Order in Diffusion Language Models High-Accuracy Sampling for Diffusion Models and Log-Concave Distributions

I'd just like to interject for a moment. What you're referring to as Bluesky, is in fact, For You/Bluesky, or as I've recently taken to calling it, For You plus Bluesky. Bluesky is not a social media site unto itself, but rather another free component of a fully functioning recommendation algorith-

I was in the Bay Area over the weekend, and Alyosha Efros twisted my arm to give a talk in his weekly research group meeting and to "keep it entertaining." I took this as an opportunity to think of five philosophy papers everyone working on AI/ML should read. realizable.substack.com/p/five-philo...

There are lots of service roles that people want to be assigned by knowledge qualifications, but are CORRECTLY about who you know. ACs‘ main job is convincing buddies to do emergency reviews. Workshop organizers’ main job is recruiting speakers and sponsors.

A large part of CoNLL reviewers volunteered to do so without submitting a paper. A large sign of confidence. I'll be live tweeting from #acl and #icml so forgive the possibilty high volume

📢 Just announced! Join us for the #KempnerInstitute workshop “Learning Dynamics in Natural and Artificial Intelligence: Evolution, Adaptation, and the Foundations of Efficient Learning.” Learn more, register, or submit an abstract 👉 bit.ly/3QwJlHR

"Pokemon" is short for "pocket monsters". "Digimon" is short for "digital monsters". Therefore, "salmon" is short for "saltwater monsters". (You might object: salmon spend a big part of their life in freshwater. But pokemon spend a big part of their life outside pockets too!)

Congratulations to outgoing #KempnerInstitute research fellows Naomi Saphra (@nsaphra.bsky.social) and Alex Damian on their faculty appointments at Boston University and MIT! 🎉 Read more about Naomi and Alex's research 👇 bit.ly/4whMn1A @bucds.bsky.social @mitmath.bsky.social

ACL needs to adopt the expectation from ML conferences that workshops be exciting. If you come from NLP you know ACL workshops are mostly terminal venues for abandoned/unambitious work, but it's clear that the correct approach is to host cutting-edge WiP.

If you're heading to #ICML2026, you should definitely go meet my collaborators and check out our work! Highlight 🧵 for 2 main track papers, 2 position papers, and 2 workshop preprints!

the fact that AI judges prefer sloppy AI writing makes the total death of good human-readable prose almost inevitable in scientific publishing and writing competitions. not sure what we can do about that.

New paper! w/ @cocoscilab.bsky.social🧵Can large language models reason flexibly, or have they learned what reasoning looks like? We introduce a new paradigm to test this question—the riddle riddle—and find that humans and LLMs show opposite patterns of performance. 📜

I'm so happy when other people write papers on nondeterministic factors in training. embrace the chaos

When I started my PhD I thought it was fine for images but wouldn’t work for discrete structures data like language 😂

Random inquiry: does anyone here know of good research on euphemisms in 19th-century American print culture? I can find *a lot* about Victorian euphemism, but my searches for American equivalents are not yielding as much fruit

📣 Announcing the BlackboxNLP 2026 Reproducibility Challenge! A new track dedicated to rigorous robustness checks of NLP interpretability work - stress-testing baselines, ablations, generalizability, and evaluation.

before a model's optimizer settles into a smooth basin, gradient updates can explore in seemingly any direction, taking nearly orthogonal steps. This is known as the Hoe Phase

I printed a custom t-shirt that's an ode to @simonwillison.net's Pelican benchmark. The super chill highly saturated cruising pelican came from this Gemini 3 launch in Feb: simonwillison.net/2026/Feb/12/... (And the caption was inspired by the reaction on social media to the blog post.)

People don’t talk about how great it is to be monolingual when you listen to great music with really stupid lyrics

Why did RNNs fail to learn long-term dependencies? What if we added one more modification, maybe now? It turns out we can give a pretty broad, analytical answer! See the attached paper for a rigorous treatment using centre manifolds, low-rank RNNs, and dynamical systems theory! go.aps.org/4fXWEeF

What are your favorite scientific writing guides?

How do you make LLMs actually good at explaining new math? It's like reading a badly written reference for people who already know the subject. When I ask a question, it never matches my level. If I try to rephrase to test my understanding, it just sycophantically agrees.

We took a few LLMs, had them play 800 games against themselves in different instruction languages while keeping the board and actions the same, balanced across player roles, and measured win-loss margins If skills were language-invariant every matchup should be near 0.0 wl margin But they aren't!

I spent the last month obsessed with finding colors that can't be displayed on a conventional screen. This is what I found. moultano.wordpress.com/2026/06/19/w...

The interactive data storytelling in this blog post is so gorgeous I don't know how to take the screenshots I want to share :/

Random variation in training and generation legit is the most underrated thing