🚨New paper led by @danielsc4.it & @saracandussio.bsky.social studying answer commitment across CoT for math & logic reasoning! Turns out, models often converge to final answers early in the CoT, then tend to "fake" hedging & re-checks - and we can tell apart mid/final guesses quite robustly! See 👇

🐸 New position paper on compositionality! 🐸 I synthesize my thoughts about the proper role/interpretation of behavior and mechanism in asking the question "Is this system (mind or machine) exhibiting compositionality?"

folks, don't use claude for slides. the style's immediately recognizable, it looks good, but it slides right over people's brains. in an era where sounding professional doesn't cost nothing to nobody, just some tokens, we're entirely free to be non-professional . embrace dreamcore wordart or smth

End of the month nl post has a bunch of bangers on AI and fiction, two by @mariaa.bsky.social in fact 😀. Also the usual web toys and games, some AI video contest winner links, Krea 2 LoRAs, and some fun games things. open.substack.com/pub/arnicas/...

not wrong! bambamramfan.github.io/ai-compass/

I know a bunch of bsky employees follow me, please, just set For You to be the default or in the set of defaults. The atmosphere is definitely the end-goal but in the short term having bluesky be usable benefits all of that

📣 #EACL2027 updates: the Call for Papers is live, and our keynote speakers are confirmed! Main conference: 9–14 March 2027. Special theme: The Human in Language. 🧵👇 2027.eacl.org/calls/papers/

even if it was good policy, this shit is completely ancillary to the goal of actually limiting the power of big tech. Elon Musk has a trillion dollars and is trying to start a global race war, your response to that can’t be “kids should be on their phones less”

2) Characterizing Narrative Content in Web-scale LLM Pretraining Data, by @teagrjohnson.bsky.social, @elliottash.bsky.social, @andrewpiper.bsky.social, @mariaa.bsky.social arxiv.org/abs/2606.19468 Why: annotation and analysis of narrative features across the pretraining data (Dolma)!

Some thoughts on how to collaborate on reproducible paper writing with AI agents, without them introducing hallucinated references or other issues into your paper. babieslearninglanguage.blogspot.com/2026/06/help...

Boulder NLP has 19 papers accepted to #ACL2026 across the main conference, findings and workshops! 🚀 We'll be presenting work on morphology, machine translation, narratives, low-resource languages, CSS, HCI, LLMs, fairness/bias, and more. See you in San Diego! 🌊 ☀️ 🌮

a nice readable survey/introduction post

I added two new sections to my tenure-line academic job market guide: "Are You Ready to Apply?", with suggestions for how to self-assess a portfolio, and "Re-Entering Academia", for people with PhDs who have been away from universities or research for a long time. shomir.net/tt_job_guide...

📣 I'm hiring! Two positions (PhD student & Postdoc) in my ERC project Macroevolution of European Literature. Let's study the cultural evolution of literature using massive data 📚📈 📍Frankfurt | Apply by August 15 Details: www.aesthetics.mpg.de/en/career/jo... Please repost to help spread the word!

Not great that the richest man in the world agrees with this post. Note the giant gun in the image. 👀 1/

One of the things that AI now gets thrown into is qualitative interviews. I'm very lucky to work with real social scientists on the pros and cons of that. The first outcome of our @villumfonden.bsky.social grant with @hjalmarcarlsen.bsky.social is ACL'26 demo: aclanthology.org/2026.acl-dem... /1

This is such a fantastic paper! We need more of this kind of empirical work on what people really do with AI. The idea of the "solipsistic reader-writer" seems important. Also, the paper is so well-written!

Recent earthquakes in northern Venezuela have caused an emergency the country is not equipped to handle. The state has lagged behind and citizens have been self-organizing. @caracaschronicles.tierra-de-gracia.com collected a list of trustworthy organizations to donate to if you are wanting to help:

I haven't seen anyone share this post here, but it's (1) a very nice guide to interviewing for ML/NLP positions these days and (2) what everyone in my NLP circle wants to talk about in person for the last week, partly out of surprise that coding interviews remain so central.

📣 Applications are now open! Flourishing in Florence is a writing workshop for health care professionals focused on storytelling, narrative medicine, + reflection. 📅 Oct. 11–16, 2026 📍 Florence, Italy 🎓 CME credit available Apply: www.writeyourlifeletter.com/florence-wri...

PSA Anisota is no longer a Bluesky client, it is an Atmosphere client. You can publish directly to @grain.social + @leaflet.pub / @standard.site among many other things. You can also cast, craft, & share Spells. There's a new way to engage called Post Poetry. + alt text generation. So much more!

Har delt om denne klient Lea før. Men hvis man er nysgerrig hvad man kan kreere her på ATproto (protokollen Bluesky/Eurosky er bygget på) er præsentationen af Lea her et godt eksempel. En klient af to forskere, for forskere til at kunne lave en fokuseret brug af skyen.

Two ICE agents tracked down a woman *inside a polling place* yesterday. Why? They demanded that she take down an Instagram post in which she named Jonathan Ross as the man who shot & killed Renee Good (after the Star Tribune identified Ross). www.syracuse.com/news/2026/06...

Love to see this work on Wildchat, one of the most bonkers datasets of our time

Very interesting paper! Looking forward to seeing whether my speculations match what people do and how the future plays out.

This is so wildly exciting - I'm just blown away by how interesting this project is, how special it is. Congrats to you all!

reminded of www.chinatalk.media/p/ken-liu-on... "AI is a desire-fulfilling machine, but it’s only able to do that for you, and only you would find it interesting. [...] doesn’t mean people who love this will stop appreciating fiction written by humans that’s not meant to fulfill desires."

Very important updates from our #FAccT2027 Program Chairs. Our next reviewing cycle will run on a different schedule from previous years. ⚠️ Abstract Deadline: Oct 27, 2026 ⚠️ Paper Deadline: Nov 3, 2026 Please read this blog post from the PCs for more info: facct-blog.github.io/2026-06-24/s...

“Hill and Morris received 50-year sentences for conspiracy to riot and ambush a law enforcement officer, even though they were not present when the shots were fired.”

Have been telling friends about this since the talk version blew my mind. See e.g. this recurring story about Meredith Lexington, an editor who wants to go back in time to change history. Appropriately enough the prompt writer goes back (in time??) to edit Meredith's prompt — more than 30 times.

What are people really doing with chatbots? 👀 Turns out that a lot of them are writing/reading various kinds of fiction, and in particular, they're generating a ton of erotica! What does this mean for the future of reading? Paper: arxiv.org/abs/2606.22748 Web Explorer: ai-fiction-wild.com

Excited to share this. @neel2112.bsky.social, @mariaa.bsky.social, and I analyzed 500K anonymous ChatGPT convos (shared w/ consent from WildChat) to see if people were generating fiction. We found tons of stories, fanfiction & erotica. Many users iterated on the same stories for days and weeks.

Preprint with @mariaa.bsky.social and @mellymeldubs.bsky.social on how real people are writing fiction with LLMs based on the Wildchat dataset here arxiv.org/abs/2606.22748.

Happy to share that our work “Text Region Detection in Historical Astronomical Diagrams” is accepted to #ICDAR2026! 🎉 An open-access dataset of 948 historical astronomical diagrams spanning 10 centuries and 7 linguistic traditions to tackle text detection in complex and diverse settings. 🌌 1/4

About clothes, it’s awful. Everything makes you want pretty clothes like hell.

📢 New paper where we evaluate the process of how researchers edit AI-generated scientific text by analyzing the entire editing trajectories instead of just the final edit. we study: what kind of edits they make and when, how does it change the quality of text, and a lot more 👇

story shapes, but make it interpretability

Unlimited-OCR 🔥New OCR from Baidu huggingface.co/baidu/Unlimi... It can parse hundreds of pages in a single pass while maintaining stable speed. The key is R-SWA (Reference Sliding Window Attention), which keeps KV cache constant during decoding. 🏆 93% on OmniDocBench 📈 +6% over DeepSeek-OCR

immediately ordered

This feed built with love by your local #FAccT2026 publicity chair and #1 fan of @graze.social. If you're interested in fairness and ethics of tech, then like, subscribe, and enjoy all the paper threads and discussion!

Ran the same 650M model across 7 languages: German, Polish, Finnish, Russian, Serbian, Estonian, Swedish via @europeana.bsky.social newspapers. Still some challenging layouts, but pretty mind-blowing to me that a 650M model is doing this when a year ago 72B VLMs failed very often.

Super Cool !! We need more such fun papers looking into Pre-training data !!

Very welcome cautious take on what historical language models actually do, from @tpoibeau.bsky.social. Section 4, on the ladder of validity, is particularly worth a look: philpapers.org/archive/POIW...

If you like writing long threads on Bluesky, wait until you discover "blog posts." Start blogging with @pckt.blog, @leaflet.pub or @offprint.app today and get 25% off your first year, on us.

I'm co-chairing the social impact track at AAAI this year, with Andrew Perrault. Send us your best society-facing work! Personally, I'm especially hoping to see more work speaking to mediators of why and when AI has social impact (or not), like how AI fits into human organizations and decisions.

please apply! come do something weird in the Media Archaeology Lab!

I have been watching the dialogue on this one closely. To everyone who is saying "OP is special, no one could use AI this way" well, I have spent 18 months doing significant amounts of scicomm in patient communities and they are ALL using AI this way. wrote this to help: github.com/DrCatHicks/i...

Nice work led by @teknology.bsky.social where we develop a memory system to support a heterogeneous population of independently acting agents by sharing experience through retrieval/personalization, contrasting it with memory confined to a single agent or cooperative team. arxiv.org/abs/2606.19911

Some new work on AI & science: A reporting checklist for LLMs in behavioral science Thanks to @sfeuerriegel.bsky.social for spearheading. As LLMs become more common in research, transparency is essential. We developed a reporting checklist for reporting how LLMs were used - please share! 1/4

This vis got me thinking about ways to show a quantitative variable on a text block. Here's a pure CSS library that provides some options (fill, like this; marks in the margin; angled texture; an underline "bar plot"; etc) and picks up a value from a <div> element. mimno.github.io/TextVariable...