I find it so wild (and cruel) that #NeurIPS would ask for location preference for presentations before announcing results. If this is for planning, more useful to ask at submission time, if not, this could wait for a ~week. We need to be kinder to the authors, especially the first time ones.

Anthropic will invisibly watermark all of Claude's text and file output from 2 August to meet EU AI Act transparency rules, with the marking applied globally, whether or not the user is anywhere near Brussels.

EU rules force Anthropic to expose AI writing worldwide

Anthropic will invisibly watermark all of Claude's text and file output from 2 August to meet EU AI Act transparency rules, with the marking applied globally, whether or not the user is anywhere near Brussels.

l.euronews.com

New achievement unlocked — organically came across an article I was quoted in 🥳 A couple of years ago, we were trying a human-centered way to analyze LLMs. Though LLMs continue to get better on tasks these are based on LLM-as-a-judge setting which I am not a 100% sold on.

MIT Technology Review@technologyreview.com · 2mo ago

If Dr. ChatGPT is an improvement over Dr. Google—and early evidence suggests it may be—it could potentially lessen the enormous burden of medical misinformation and unnecessary health anxiety that the internet has created.

100% true. Expertise associated with women is systematically undervalued. Knitting, weaving, and sewing each involve extensive programming, material science, and engineering design. Yet, these are not seen from these perspectives.

What puts women off from academic careers is sexism and cryptogyny. Do you as editor fight that too? olivia.science/cryptogyny/

Cryptogyny

On the systematic obfuscation of women's contributions.

olivia.science

Finally, an attempt at some much needed clarification!

I am genuinely surprised how a number of things can be *load bearing* these days ✨🤖

How easy is it to tamper LLMs? Our latest work (set to appear at #KDD 2026 D&B) stress-tests 21 open-weight LLMs and against SOTA attacks and defenses, and makes it easy for the community to benchmark. TL;DR All LLMs are vulnerable. More info in this thread 🔥 1/9

Don’t ever judge a book by its cover 📚

Writing cards for accepted cohort is special as well as a reminder that I have not written in a long time. Hope they understand my cursive, and it does not deter them from joining 🙂 #Uwaterloo #AcademicLife

Hello blue sky!

💡Okay, I am so excited to announce the preprint of my lab’s work on memory and time-efficient training of #LLMs, especially with #DeepSeek’s success bringing this to the centerstage! arxiv.org/abs/2502.01586 #CriticalML #AI #ML

SubTrack your Grad: Gradient Subspace Tracking for Memory and Time Efficient Full-Parameter LLM Training

Training Large Language Models (LLMs) demand significant time and computational resources due to their large model sizes and optimizer states. To overcome these challenges, recent methods, such as BAd...

arxiv.org

University of Chicago Press has a pretty good sale with deeply discounted overstock from the recent back catalog (stuff published ~2-10 years ago). Got hardcovers of these two books for $11 each. press.uchicago.edu/resource/Sal...

Been listening to this whole day today youtu.be/OgU_UDYd9lY?...

jazzy but not too jazzy

YouTube video by take your time

youtu.be

It’s been a tough few weeks. My 10yo daughter was diagnosed with a very rare, aggressive cancer called interdigitating dendritic cell sarcoma (IDCS). I’m reaching out to identify clinicians/patients who have encountered pediatric IDCS or other (non-LCH) dendritic or histiocytic sarcomas cases.

🧑‍🎓Recently I had a student tell me that they will consider a Ph.D. if they got a particular award since it will demonstrate that they are good.

🌱 Watching the exponential rise in the attendance at Women in Engineering (WiE) socials over the past few months is so heartwarming, and a testament to the strong #WiE community at the University of Waterloo

🤖 I often talk to companies and organizations who want to use #AI/#ML in their workflows. But before they can leverage these powerful tools, they need to become #data-centric, and what I call — #AIReady

💭Okay, I have to say this. NYT (and I am sure some other news outlets) coming out NOW with articles and opinion pieces titled “How does tyranny begin?” is so jarring and infuriating. #Canadian media — I am also looking at you.

🌍This week has been indicative of the changing world order. While the looming tariffs are being seen as some end-of-times, yoking #Canada’s fate to a country trying to become insular and anti-science is also a recipe for disaster!

A professor of African American studies accidentally starting an entire trend on #tiktok where educators are teaching courses, and thousands of people (“students”) are “enrolling” is exactly the type of heart-warming news I needed. #hillmantok #hillmantokuniversity

Kind of wild to see #NVIDIA’s stock plunge so much because of #DeepSeek — we will still need GPUs — at least till the time we come up with more efficient hardware alternatives. Yes, orgs will not have to be compute ultra-rich, which is very exciting!

Okay, I am excited and relieved that with #DeepSeek the focus will move towards efficiently using training and compute. Just throwing compute at a problem is a problem in itself. Also, hello, it is open source?! That is going to spur a new wave of cool insights! 👀

We are looking for volunteers for reviewing and AC roles! Please sign up here: forms.gle/rZp67YvMn1hn...

COLM 2025 Program Committee Volunteer Form

This form is to volunteer to serve on the COLM 2025 (https://colmweb.org/) program committee as a reviewer or an area chair.

forms.gle

Weak people with power will say whatever they need to to retain power & be validated, including putting others down. Don’t waste your time on these theatrics. If you feel excluded, remember: These people are wherever they are BECAUSE of their identity & your are here DESPITE it — keep going!

Casey Newton@caseynewton.bsky.social · 2y ago

My favorite part of Zuckerberg's Joe Rogan interview is when he suggests there is not enough masculine energy at his company, where two out of every three employees is a man.

Have not used my #facebook account in years, and never had an #instagram (tried it to shop once and it did poorly even at that) BUT when I used #TikTok, it was a different story — I found niche communities, learned so much, and it became my go-to for art, books, articles, recipes, and so on.

Found this fantastic tool — arxiv2notion to add papers to a #notion database from #arxiv and #openreview using a chrome extension 🔥 Tip: Use the manual option, the chrome store option does not work (for me at least) github.com/denkiwakame/...

GitHub - denkiwakame/arxiv2notion: Chrome extension for clipping arXiv articles to Notion.

Chrome extension for clipping arXiv articles to Notion. - denkiwakame/arxiv2notion

github.com

I am so flattered that 99% of people mistook me for a PhD student at #neurips I will share my hair, skin, and fitness routine soon — stay tuned! #wiml #AI #ML

If we don't solve the alignment and control problems, AI risks will continue to grow as we give systems more and more agency. We need major progress in both science and policy beforehand. Webb Wright's article in @sciam.bsky.social to which I contributed: www.scientificamerican.com/article/what...

AI Agents with More Autonomy Than Chatbots Are Coming. Some Safety Experts Are Worried

Systems that operate on behalf of people or corporations are the latest product from the AI boom, but these “agents” may present new and unpredictable risks

scientificamerican.com

Now that @mkearnsphilly.bsky.social is here, this starter pack just became the real deal. Better refollow it. go.bsky.app/21nFz12