Pushpdeep @pushpdeep.bsky.social · 12/09/2026Preprint Alert 🚨 The best thing about Bluesky and the AT Protocol is their openness and transparency 💓. We use this to study @moderation.bsky.app. Hoping it pushes other platforms toward more transparent moderation and allows better understanding of systemic risks. 📄 arxiv.org/abs/2609.11373 030
Reposted by PushpdeepInterdisciplinary Institute for Societal Computing @i2sc.net · 12/09/2025The lecture was followed by a hands-on activity where Abhisek Dash and @pushpdeep.bsky.social (MPI-SWS) guided the participants through the challenges of validating labels of content posted on Bluesky. 111
Reposted by PushpdeepChristopher Barrie @cbarrie.bsky.social · 25/03/2025📄NEW PAPER📄 Ever wondered content people actually pay *attention* to online? Our new research reveals that you likely pay attention to far more varied political content than your likes and shares suggest 812343
Reposted by Pushpdeeparxiv cs.CL @arxiv-cs-cl.bsky.social · 03/03/2025Matthias Orlikowski, Jiaxin Pei, Paul R\"ottger, Philipp Cimiano, David Jurgens, Dirk Hovy Beyond Demographics: Fine-tuning Large Language Models to Predict Individuals' Subjective Text Perceptions arxiv.org/abs/2502.20897 021
Reposted by Pushpdeeparxiv cs.CL @arxiv-cs-cl.bsky.social · 26/02/2025Artem Vazhentsev, Ivan Sviridov, Alvard Barseghyan, Gleb Kuzmin, Alexander Panchenko, Aleksandr Nesterov, Artem Shelmanov, Maxim Panov Uncertainty-aware abstention in medical diagnosis based on medical texts arxiv.org/abs/2502.18050 021
Reposted by Pushpdeeparxiv cs.CL @arxiv-cs-cl.bsky.social · 26/02/2025Tushar Aggarwal, Kumar Tanmay, Ayush Agrawal, Kumar Ayush, Hamid Palangi, Paul Pu Liang Language Models' Factuality Depends on the Language of Inquiry arxiv.org/abs/2502.17955 021
Reposted by PushpdeepVicki @vickiboykis.com · 26/02/2025Every day we get closer and closer to OG information retrieval , imagine spending billions of dollars to do the same thing that Tf-IDF did noperator.dev/posts/docume...noperator.devHard problems that reduce to document rankingThere are two claims I’d like to make: LLMs can be used effectively1 for listwise document ranking. Some complex problems can (surprisingly) be solved by transforming them into document ranking proble... 4759
Reposted by PushpdeepLinguistic Discovery @linguisticdiscovery.com · 23/02/2025A new book on mixed languages in South Asia has just been released! @PenguinBooks 171
Reposted by PushpdeepJessy Li @jessyjli.bsky.social · 21/02/2025Do you want to know what information LLMs prioritize in text synthesis tasks? Here's a short 🧵 about our new paper, led by Jan Trienes: an interpretable framework for salience analysis in LLMs. First of all, information salience is a fuzzy concept. So how can we even measure it? (1/6) 1156
Reposted by PushpdeepHiba Ahsan @hibaahsan.bsky.social · 22/02/2025LLMs are known to perpetuate social biases in clinical tasks. Can we locate and intervene upon LLM activations that encode patient demographics like gender and race? 🧵 Work w/ @arnabsensharma.bsky.social, @silvioamir.bsky.social, @davidbau.bsky.social, @byron.bsky.social arxiv.org/abs/2502.13319 3177
Reposted by PushpdeepThomas Wolf @thomwolf.bsky.social · 19/02/2025After 6+ months in the making and over a year of GPU compute, we're excited to release the "Ultra-Scale Playbook": hf.co/spaces/nanot... A book to learn all about 5D parallelism, ZeRO, CUDA kernels, how/why overlap compute & coms with theory, motivation, interactive plots and 4000+ experiments!hf.coThe Ultra-Scale Playbook - a Hugging Face Space by nanotronThe ultimate guide to training LLM on large GPU Clusters 217952
Reposted by PushpdeepAdi Simhi @adisimhi.bsky.social · 19/02/2025🚨New arXiv preprint!🚨 LLMs can hallucinate - but did you know they can do so with high certainty even when they know the correct answer? 🤯 We find those hallucinations in our latest work with @itay-itzhak.bsky.social, @fbarez.bsky.social, @gabistanovsky.bsky.social and Yonatan Belinkov 32110
Reposted by PushpdeepIshika Agarwal @wonderingishika.bsky.social · 17/02/2025🚀Very excited about my new paper! NN-CIFT slashes data valuation costs by 99% using tiny neural nets (205k params, just 0.0027% of 8B LLMs) while maintaining top-tier performance! 1114
Reposted by PushpdeepChristopher Barrie @cbarrie.bsky.social · 18/02/2025We have a new version of our 𝕡𝕣𝕠𝕞𝕡𝕥𝕤𝕥𝕒𝕓𝕚𝕝𝕚𝕥𝕪 paper now on arxiv! This is a significant update that test *a lot* more data, suggests post-processing techniques, outlines how to compare across models, and tests with new models... 13510
Reposted by PushpdeepLinguistic Discovery @linguisticdiscovery.com · 08/02/2025Read more in this issue of the Linguistic Discovery newsletter, where I explore the Trisolaran language from the Three-Body Problem and how it compares to human language! buff.ly/3ErakOl #Trisolarans #aliens #xenolinguistics #ThreeBodyProblem #linguistics #language #SciFi #review buff.lyWhat if we could hear each other's thoughts? The linguistics of The Three-Body ProblemImagine if every word you thought could be heard by everyone around you. In this world, thinking would be the same as communicating. What would language—and society—be like? 062
Reposted by PushpdeepQueer in AI @queerinai.com · 03/02/2025Our workshop has been extended till Feb 20. We are looking forward for your papers at NAACL's Queer in AI workshop. 01717
Reposted by PushpdeepShaily @shaily99.bsky.social · 01/02/2025This is a good start!! More *CL conferences should come 😍 @aaclmeeting.bsky.social 1151
Reposted by PushpdeepÓlafur Waage @olafurw.com · 31/01/2025Programming languages: "We are just a way to operate computers in a way that makes sense to humans." Programming languages [takes a big joint hit]: "What if there were 5 kinds of nothingness?" 2663386
Reposted by PushpdeepNathan Lambert @natolambert.bsky.social · 31/01/2025A useful oversimplification Instruction finetuning (IFT/SFT): imprinting features or shape in responses Preference finetuning (RLHF/DPO/etc): style Reinforcement finetuning (RFT/RLVR/etc): learning new behaviors 1182
Reposted by PushpdeepDustin Wright @dustinbwright.com · 31/01/2025📄 New preprint: "Collecting Cost-Effective, High-Quality Truthfulness Assessments with LLM Summarized Evidence" We show: fact checking w/ crowd workers is more efficient when using LLM summaries, quality doesn't suffer. arxiv.org/abs/2501.18265arxiv.orgCollecting Cost-Effective, High-Quality Truthfulness Assessments with LLM Summarized EvidenceWith the degradation of guardrails against mis- and disinformation online, it is more critical than ever to be able to effectively combat it. In this paper, we explore the efficiency and effectiveness... 1193
Reposted by PushpdeepMor Geva @megamor2.bsky.social · 28/01/2025How can we interpret LLM features at scale? 🤔 Current pipelines use activating inputs, which is costly and ignores how features causally affect model outputs! We propose efficient output-centric methods that better predict the steering effect of a feature. New preprint led by @yoav.ml 🧵1/ 1324
Reposted by PushpdeepNature @nature.com · 24/01/2025Roughly 6,000 readers answered our poll, with many declaring that Bluesky was nicer, kinder and less antagonistic to science than X go.nature.com/42tH8Aigo.nature.comBluesky’s science takeover: 70% of Nature poll respondents use platformRoughly 6,000 readers answered our poll, with many declaring that Bluesky was nicer, kinder and less antagonistic to science than X. 5452651142
Reposted by PushpdeepMilaNLP Lab @milanlp.bsky.social · 16/01/2025#ThrowbackThursday #NLProc "My Answer is C" by Wang et al. highlights that first-token evaluation does not accurately reflect LLM behavior in user interactions, urging against sole reliance on this method.buff.ly“My Answer is C”: First-Token Probabilities Do Not Match Text Answers in Instruction-Tuned Language ModelsXinpeng Wang, Bolei Ma, Chengzhi Hu, Leon Weber-Genzel, Paul Röttger, Frauke Kreuter, Dirk Hovy, Barbara Plank. Findings of the Association for Computational Linguistics: ACL 2024. 2024. 042
Reposted by PushpdeepMiriam Schirmer @miriamschirmer.bsky.social · 17/01/2025📕 My dissertation on #NLP for #Violence Studies has been published: mediatum.ub.tum.de?id=1751256 I've been looking at #abusive behavior online, as well as sharing of personal experiences with violence, incl. psychological #trauma. Excited to push this research forward and connect with others 🌐mediatum.ub.tum.demediaTUM - Medien- und Publikationsserver 071
Reposted by PushpdeepDr. Jordan Taylor @jordant.bsky.social · 15/01/2025Excited this work is finally published: dl.acm.org/doi/10.1145/...dl.acm.orgBuilding Solidarity Amid Hostility: Experiences of Fat People in Online Communities | Proceedings of the ACM on Human-Computer InteractionOnline communities are important spaces for members of marginalized groups to organize and support one another. To better understand the experiences of fat people - a group whose marginalization often... 161
Reposted by PushpdeepChris Potts @cgpotts.bsky.social · 13/01/2025I've posted the practice run of my LSA keynote. My core claim is that LLMs can be useful tools for doing close linguistic analysis. I illustrate with a detailed case study, drawing on corpus evidence, targeted syntactic evaluations, and causal intervention-based analyses: youtu.be/DBorepHuKDMyoutu.beFinding linguistic structure in large language modelsYouTube video by Chris Potts 17420
Reposted by PushpdeepNathan Lambert @natolambert.bsky.social · 08/01/2025The state of post-training in 2025: a tutorial on modern post-training A re-record of my NeurIPS tutorial on language modeling (plus some added content on the high level state of things) Blog + extra context: buff.ly/424VvLm YouTube: buff.ly/40808l5 Slides: buff.ly/404jGa9buff.lyThe state of post-training in 2025A re-record of my NeurIPS tutorial on language modeling (plus some added content). 48018
Reposted by PushpdeepMolly White @molly.wiki · 05/01/2025call me when you figure out how to make working out more like reading (in bed with a cup of tea) 1766462
Reposted by Pushpdeeppixelatedboat aka “mr bluesky” @pixelatedboat.bsky.social · 06/01/2025I can’t enjoy horror movies because the characters always act so irrationally. If I was in a horror movie situation I would simply kill myself straight away 3038415769
Reposted by Pushpdeepdanah boyd @zephoria.bsky.social · 04/01/2025My advisor warned me that academics trend towards bitterness. He encouraged me to intentionally resist this, remember where I came from, and never forget the privilege of getting to spend a life working with knowledge and ideas. He too said that bitterness and resentment is easy. 124937
Reposted by PushpdeepNilah "Nine Jobs" Magruder @nilahmagruder.com · 31/12/2024This comic gets shared every year with the date changed. It is by Brazilian cartoonist Glauco Villas Boas (1957-2010). 3677152395
Reposted by PushpdeepInternet Archive @archive.org · 27/12/2024Preservation is at the heart of what we do. This year, we released the Vanishing Culture report— detailing instances of cultural loss and emphasizing the role that libraries and archives play in preserving materials. Help us in saving these resources:https://archive.org/donate/?origin=blsky-eoy2024 121710599
Reposted by PushpdeepMeredith Whittaker @meredithmeredith.bsky.social · 26/12/2024This is a key point. If there's one thing I learned cutting my teeth on measurement, it's that if you control the benchmark--the terms by which progress is evaluated--you control the game board. 217828
Reposted by PushpdeepBluesky @bsky.app · 26/12/2024Merry Christmas from us to you 🎄🎁💙 We launched Trending Topics today, and you can find it by tapping the search icon on the bottom bar of the app or the right sidebar on desktop. 1507464125227
Reposted by PushpdeepMegan Figueroa @meganfigueroa.bsky.social · 19/12/2024My favorite print from liberaljane.store “English is a language, not a way to measure someone’s intelligence.” 515939
Reposted by PushpdeepBetsy Sneller @betsysneller.bsky.social · 18/12/2024Bill Labov died this morning. I'm not coherent enough to talk about how important and influential and brilliant he was. I am very sad. I was so lucky to know him, and I am grateful every day that he (and Gillian, and Walt, etc) built an academic field where kindness is expected. 24698120
Reposted by PushpdeepSandra González-Bailón @sgonzalezbailon.bsky.social · 11/12/2024New publication (4+ years in the making): “The Diffusion and Reach of (Mis)Information on Facebook”. shorturl.at/VE2fU We analyze the propagation of 1B+ posts across content moderation regimes, with @davidlazer.bsky.social @jatucker.bsky.social @taliastroud.bsky.social @annenbergpenn.bsky.social 925494
Reposted by PushpdeepBen Schmidt @bschmidt.bsky.social · 11/12/2024Google quietly updated their ngrams viewer again this year. The books used appear to be extremely different yet again--the rate of the words "she said" are about 60% what they were in the 20th century compared to the 2019 release, and just 20% compared to 2009. But there's a catch: 521579
Reposted by PushpdeepDr Abeba Birhane @abeba.blacksky.app · 11/12/2024what are the legal repercussions of genAI producing misleading, false or otherwise damaging information about individuals? this paper by @rdbinns.bsky.social & @lilianedwards.bsky.social is super informative and insightful 27321
Reposted by PushpdeepSarah Bull @sarahebull.bsky.social · 06/12/2024As a historian who focuses on historical texts and has done a lot of work with offensive historical texts in particular, this paper on a process to 'detoxify' historical texts so they can be used to train LLMs for contemporary applications was very interesting! + arxiv.org/abs/2410.22587arxiv.orgToxicity of the Commons: Curating Open-Source Pre-Training DataOpen-source large language models are becoming increasingly available and popular among researchers and practitioners. While significant progress has been made on open-weight models, open training dat... 45110
Reposted by PushpdeepDominic @dmncschmtz.bsky.social · 05/12/2024He's also putting the glamour back into grammar on YouTube now! 🎶 #linguistics www.youtube.com/watch?v=pSOl...youtube.comMeet Christian Uffmann! #phonologyYouTube video by HHU 39846
Reposted by PushpdeepLynn Cherny @arnicas.bsky.social · 05/12/2024Super work on the “poetic” style output of GPT models from - humans flout the rules, and vis of unstoppable limerick output #chr2024 @mellymeldubs.bsky.social et al 2024.computational-humanities-research.org/papers/paper... 23310
Reposted by PushpdeepBen Burtenshaw @benburtenshaw.bsky.social · 03/12/2024For anyone interested in fine-tuning or aligning LLMs, I’m running this free and open course called smol course. It’s not a big deal, it’s just smol. 🧵>> 932363
