QWEN plans

Keynote at #COLM2025: Nicholas Carlini from Anthropic "Are language models worth it?" Explains that the prior decade of his work on adversarial images, while it taught us a lot, isn't very applied; it's unlikely anyone is actually altering images of cats in scary ways.

Defining training document quality seems to still be an open problem #COLM

COLM word cloud. Yoav says it’s the year of reasoning, but evaluation is also huge.

Interesting q: in preliterate times would the desire and ability to memorize 27k lines of Homer have been a kind of neurodiversity?

This time we finish it

Three major drifts we found: 1️⃣ Licenses: from corporate to other types. We often see use restrictions mutate to permissive or copyleft (even when counter to upstream license terms) 2️⃣ Languages: from multilingual → English-only 3️⃣ Docs: from long & detailed → short & templated

I hear such good things about Rust, and then I actually look at it and its stuff like this:

“Which model will rid me of this turbulent priest?”

The UC Berkeley School of Information is hiring an assistant professor in the broad field of Information--including areas of info seeking/retrieval, digital humanities, cultural analytics, info viz, & philosophy of information (among others). Deadline Nov 1! aprecruit.berkeley.edu/JPF05014

Frankly going to implore progressives to follow better what is happening in China tech. It's not a VC bubble. It's a dozen top labs, it's most large co getting heavily into AI (at a pretraining level) and robotics, it's a deep bet on the next industrial revolution.

I hate to talk about Christmas at a time like this (it being only Sept.), but if you're thinking about commissioning a gift map for Christmas...'tis the season! I've done a number of gift maps over the last few years, and they take a while. So now's the time to get in line!

Does anyone know the purpose or typical use of the "next line" character, Unicode 0x85? How does it relate to CR and LF?

It’s not the Reichstag, it’s Horst Wessel

All of these tips are derived from the psychological protections developed for content moderators, and if you want more detail and citations, crowd.cs.vt.edu/wp-content/u... is the paper that has the best collection of citations that I've found .

There have been *some* studies on other games, but Tetris is the best-studied because it was the one in the original study. It does seem that the key elements are "fast/timed", "lots of concentration", and "lots of eye movement across the game board" if you can't do Tetris for whatever reason.

Job alert! 🌸🏔️ The UW iSchool is hiring for *two* TT faculty positions, with a focus on Artificial Intelligence, broadly defined. It's pretty here! Applications due November 15th 2025 (priority deadline). Link: apply.interfolio.com/171020

BLS revisions often correlated, so July prob revises down, right? But Yahoo finance reporting it as Fact: “In July, the economy created 73,000 new jobs, but the headline that emerged from that report was revisions to job gains in May and June, which wiped out some 258,000 previously reported gains.”

Please share! We're trying to crowd-source a dataset of post-1929 novels with maps.

We are trying to create a list of in-copyright novels that contain maps. If you know of some, drop them in the thread below! 🧵👇

All this about South Carolina deploying troops to DC is a reminder of what the 2nd amendment is really about: not whether Jebediah can have a musket, but whether Pennsylvania can have an army

NYC mayoral race has platonic ideals of campaign types: “pure entitled nepotism”, “placate the elites”, and “unmistakably love the place and the people who live there”

📢 Don’t miss the first articles from CHR’s themed issue, Expanding the Toolkit: Large Language Models in Humanities Research! Read the issue #openaccess here: 📚 cup.org/4lvH2Ow #computationalhumanities #CHR @comphumresearch.bsky.social @nolauren.bsky.social @folgertk.bsky.social

The only factor that really matters is that Putin reminds Trump of his cold, tough-guy father, and he desperately, desperately wants to impress him. I would bet Putin has studied Fred Trump's mannerisms.

Are you afraid of LLMs teaching people how to build bioweapons? Have you tried just... not teaching LLMs about bioweapons? @eleutherai.bsky.social‬ and the UK AISI joined forces to see what would happen, pretraining three 6.9B models for 500B tokens and producing 15 total models to study

Me in 2023: Just chill. All the AI doom scenarios involve giving models direct access to and control over real world systems, and no one would be that dumb In 2025: oh fuck

Me in 2023: Just chill. All the AI doom scenarios involve giving models direct access to and control over real world systems, and no one would be that dumb In 2025: oh fuck

It's true that "don't try to be clever, just buy bigger GPUs" has been the right choice for the past 10ish years, but investment on the scale of "revive the fossil fuel industry" and "greater than all consumer spending" is just asking to be crushed by DeepSeek-style improvements in efficiency.

The biggest shift in the EO to me is that it actively asserts that projects should advance the president’s agenda, not congress’s agenda.

The White House is taking new steps to undermine every international student at our universities. It has approved a new rule restricting student visas to 4 years max. That's not long enough for PhD students, or about half of undergrads, to finish their degree. Likely to enter force within days.

If we ever have a functioning BLS again, they should just not release jobs numbers for a month. Media have shown they cannot be trusted to make sense of preliminary data. Have any pointed out that the 73k for July is almost certainly wrong? This is a Lucy-with-the-football situation.

Mary Ellen Bute's abstract animation celebrates the parabola. She also set her elegant animations to classical music (prior to FANTASIA).

It's important to note here that we're not just talking about faculty & university staff. 100s of U.S. colleges are in towns or small cities where they form the major industry & lifeblood of the community. Loss of students -> loss of rent $, restaurant $, bars, bookstores, gaming stores...

New dataset on bestsellers from 40+ countries, with consistent coverage for France, Germany, Spain, Italy, and the U.S. Congrats to the authors @sdileonardi.bsky.social, @beccacohen.bsky.social, and @dan-sinnamon.bsky.social on this major contribution! 🎉 🔗: doi.org/10.18737/386...

#OnThisDay - 29 July - in AD 238, the Year of the Six Emperors, the short-lived reigns of the co-emperors Pupienus and Balbinus were brought to an end as the Praetorian Guard moved against them. #AncientHistory 🏺 Image: RIC IV Pupienus 9A; ANS 1944.100.13004. Link - numismatics.org/ocre/id/ric....

The future is here, and it is annoying. genAI "saves time" for one person by wasting time for lots of other people

So Overleaf is now owned by a publishing company connected with NPG? Still seems ok: "We don’t claim any ownership of your stuff. These Terms only grant us rights to use your stuff to fulfill a request you've made and otherwise provide the Services, and as explained below." But still...

I need help: decades ago Minicon was held in a Sheraton that used these pillows. We have two, only one pillowcase, all in bad shape. But they are the pillows my son wants. He’s autistic. Consistency really matters. And I can’t find them for sale anywhere. I’m hoping the internet can help!

Article I Section 8: "The Congress shall have Power To lay and collect Taxes, Duties, Imposts and Excises, unless the President should be in Disagreement with Judicial Cases, arising in Portuguese America, as he thinks proper"

SmolLM3: a highly detailed look into modern model training this is amazing. They go into great detail on just about every aspect. The number of stages, algorithms, optimizer settings, datasets, blueprints, recipes, open source training scripts, .. huggingface.co/blog/smollm3

“Look at Mr. Frog! He’s so grumpy.” Wait, why are so many frogs boys? @mellymeldubs.bsky.social explores how we gender animals in kids’ stories—and why it matters. pudding.cool/2025/07/kids...

To me this argues for a standard LLM prereview phase. The kind of basic wellformedness checks and procedural gotchas in checklists are exactly what I see in the ~60% AI-generated reviews out now. Why not cut out the middleman?

Even the Amish use power tools for woodworking. You cannot make a living with hand tools. What you're left with is higher level planning/design and a lot of safety training.

why topic model eval is hard! here's a set of words that make absolutely no sense *unless* you are reading today's Bluesky: "concentration alligator camp camps auschwitz alcatraz florida trump alligators death hurricane cages detention prison desantis nazi everglades built merch ice"

From this morning's topic model: "jimmy swaggart jesus hell god televangelist died dies dead church christian christ ministry scandals prostitution lord brothers whose rest twins"

I've been promoted to Full Professor. Thanks to all the students who have given me the opportunity to teach and mentor, to my department for offering me the opportunity to lead, and to my research community for valuing my work!

“Intellectual property law is the first aid you apply while waiting for the legislation to arrive”

Translation: For the benefit of Mr Kite…

Defunded so far: tax police, environmental police, financial services police Now firearms police