Sign in

Amanda Bertsch

@abertsch.bsky.social
2.2K followers 546 following 20 posts

PhD student @ CMU LTI. working on text generation + long context www.cs.cmu.edu/~abertsch

PostsRepliesMedia
Amanda Bertsch @abertsch.bsky.social · 10/11/2025
ooh, interesting! would the best xLSTM model to try be the xLSTM Large 7B ?
110
Reposted by Amanda Bertsch
Ari @ari-holtzman.bsky.social · 09/11/2025
LLMs don't accumulate information over the course of a text the way you'd hope! I think this is why LLMs often feel 'fixated on the wrong thing' or 'overly literal'—they are usually responding using the most relevant single thing they remember, not the aggregate of what was said
061
Amanda Bertsch @abertsch.bsky.social · 08/11/2025
Thank you so much!
010
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
We’re excited about Oolong as a challenging benchmark for information aggregation! Let us know which models we should benchmark next 👀 Paper: arxiv.org/abs/2511.02817 Dataset: huggingface.co/oolongbench Code: github.com/abertsch72/o... Leaderboard: oolongbench.github.io
arxiv.org
Oolong: Evaluating Long Context Reasoning and Aggregation Capabilities
As model context lengths continue to grow, concerns about whether models effectively use the full context length have persisted. While several carefully designed long-context evaluations have recently...
143
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
While long-context models can do many retrieval tasks impressively well, they have a long way to go to solve realistic information synthesis problems! Oolong is joint work with Adithya Pratapa, Teruko Mitamura, @gneubig.bsky.social , and Matt Gormley.
121
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
Models show varying error patterns. Claude and some GPT-family models underperform on tasks that require outputting dates; Gemini and Deepseek-R1 frequently over-reason and fail to return an answer at all on Oolong-synth, although Gemini is the best model on Oolong-real.
Score by answer type and task type for Oolong-synth. The month+year and date types are the hardest for many models, corresponding with the difficulty of the timeline tasks.
101
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
Why is this so hard? Models must identify relevant sections of input, label or categorize these sections, and then accumulate information to make distributional-level decisions. Adding labels in-context or specifying more reasoning effort has limited benefit.
Graph showing that the performance with labels in-context for Oolong synth is only slightly better.
Graph showing that increasing reasoning effort only helps marginally, and only in contexts shorter than 64K.
121
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
Oolong has a synthetic setting that poses distributional questions over sets of classification examples and their metadata and a realistic setting using conversational data from game transcripts. Both splits require counting, temporal reasoning, and multi-step entity resolution.
A figure demonstrating the two splits of Oolong: the left side has the question “Were there more news articles about the economy in September or August?”, from Oolong-synth; the right side has the question “How many times in this set of episodes does the character Jester cast Healing Word?”, from Oolong-real. Both questions require the model to label sections of input, identify the important segments, and aggregate across these to answer the question.
100
Amanda Bertsch @abertsch.bsky.social · 07/11/2025
Can LLMs accurately aggregate information over long, information-dense texts? Not yet… We introduce Oolong, a dataset of simple-to-verify information aggregation questions over long inputs. No model achieves >50% accuracy at 128K on Oolong!
Performance of a sweep of models on Oolong-synth and Oolong-real. Performance decreases with increasing context length, sometimes steeply.
35020
Reposted by Amanda Bertsch
Kyle Lo @ COLM2026 @kylelo.bsky.social · 05/11/2025
why intern at Ai2? 🐟interns own major parts of our model development, sometimes even leading whole projects 🐡we're committed to open science & actively help our interns publish their work reach out if u wanna build open language models together 🤝 links 👇
2278
Reposted by Amanda Bertsch
Sung Kim @sungkim.bsky.social · 05/11/2025
DeltaNet Explained by Sonlin Yang A gentle and comprehensive introduction to the DeltaNet Part 1: sustcsonglin.github.io/blog/2024/de... Part 2: sustcsonglin.github.io/blog/2024/de... Part 3: sustcsonglin.github.io/blog/2024/de...
0152
Reposted by Amanda Bertsch
Natasha Johnson @natashamarie330.bsky.social · 05/11/2025
I’ll be presenting this work in **2 hours** at EMNLP’s Gather Session 3. Come by to chat about fanfiction, literary notions of similarity, long-context modeling, and consent-focused data collection!
071
Reposted by Amanda Bertsch
Natasha Johnson @natashamarie330.bsky.social · 05/11/2025
Digital humanities researchers often care about fine-grained similarity based on narrative elements like plot or tone, which don’t necessarily correlate with surface-level textual features. Can embedding models capture this? We study this in the context of fanfiction!
Figure showing a similarity comparison between three stories. Story A and story B have the same author, and story A and story C have the same tone. A human might care about which stories are tonally the most similar, but a language model's notion of similarity is strongly informed by surface-level features like small differences in writing style across authors.
13913
Amanda Bertsch @abertsch.bsky.social · 12/09/2025
We'll be posting course content for anyone who would like to follow along! The first four lecture videos are available now: youtube.com/playlist?lis...
010
Amanda Bertsch @abertsch.bsky.social · 12/09/2025
. @gneubig.bsky.social and I are co-teaching a new class on LM inference this fall! We designed this class to give a broad view on the space, from more classical decoding algorithms to recent methods for LLMs, plus a wide range of efficiency-focused work. website: phontron.com/class/lminfe...
phontron.com
11-664/763 LM Inference
A class at Carnegie Mellon University on language model inference algorithms.
140
Reposted by Amanda Bertsch
Lindia Tjuatja @lindiatjuatja.bsky.social · 09/06/2025
When it comes to text prediction, where does one LM outperform another? If you've ever worked on LM evals, you know this question is a lot more complex than it seems. In our new #acl2025 paper, we developed a method to find fine-grained differences between LMs: 🧵1/9
27020
Amanda Bertsch @abertsch.bsky.social · 30/04/2025
we also have a followup work, and ‪@emilyxiao.bsky.social will also be around the conference to discuss! bsky.app/profile/emil...
000
Amanda Bertsch @abertsch.bsky.social · 30/04/2025
our paper (arxiv.org/abs/2405.00200) studies properties + tradeoffs of using long-context models for ICL, and we're very excited that it won the Language Modeling SAC award this year!
arxiv.org
In-Context Learning with Long-Context Models: An In-Depth Exploration
As model context lengths continue to increase, the number of demonstrations that can be provided in-context approaches the size of entire training datasets. We study the behavior of in-context learnin...
161
Amanda Bertsch @abertsch.bsky.social · 30/04/2025
super excited to see folks at #NAACL25 this week! I'll be presenting our work on long-context ICL Wednesday in the 2pm poster session in Hall 3-- would love to chat with folks there or at the rest of the conference about long context data, ICL, inference time methods, New Mexican food, etc :)
1111
Reposted by Amanda Bertsch
Alicia DeVrio @uhleeeeeeeshuh.bsky.social · 06/03/2025
How can we better think and talk about human-like qualities attributed to language technologies like LLMs? In our #CHI2025 paper, we taxonomize how text outputs from cases of user interactions with language technologies can contribute to anthropomorphism. arxiv.org/abs/2502.09870 1/n
Image of the first page of the CHI 2025 paper titled "A Taxonomy of Linguistic Expressions That Contribute To Anthropomorphism of Language Technologies" by authors Alicia DeVrio, Myra Cheng, Lisa Egede, Alexandra Olteanu, & Su Lin Blodgett
24311
Reposted by Amanda Bertsch
Max Müller-Eberstein @mxij.me · 23/12/2024
9.6 million seconds = 1 PhD 🔥 Finally analyzed my PhD time tracking data so you can plan your own research journey more effectively: mxij.me/x/phd-learning-dynamics For current students: I hope this helps put your journey into perspective. Wishing you all the best!
mxij.me
The Learning Dynamics of a PhD
This is what a PhD looks like: 9.6 million seconds of research.
0367
Reposted by Amanda Bertsch
Sireesh Gururaja @siree.sh · 17/12/2024
When I started on ARL project that funds my PhD, the thing we were supposed to build was a "MaterialsGPT". What is a MaterialsGPT? Where does that idea come from? I got to spend a lot of time thinking about that second question with @davidthewid.bsky.social and Lucy Suchman (!) working on this:
The abstract of a paper titled "Basic Research, Lethal Effects: Military AI Research Funding as Enlistment".

In the context of unprecedented U.S. Department of Defense (DoD) budgets, this paper examines the recent history of DoD funding for academic research in algorithmically based warfighting. We draw from a corpus of DoD grant solicitations from 2007 to 2023, focusing on those addressed to researchers in the field of artificial intelligence (AI). Considering the implications of DoD funding for academic research, the paper proceeds through three analytic sections. In the first, we offer a critical examination of the distinction between basic and applied research, showing how funding calls framed as basic research nonetheless enlist researchers in a war fighting agenda. In the second, we offer a diachronic analysis of the corpus, showing how a 'one small problem' caveat, in which affirmation of progress in military technologies is qualified by acknowledgement of outstanding problems, becomes justification for additional investments in research. We close with an analysis of DoD aspirations based on a subset of Defense Advanced Research Projects Agency (DARPA) grant solicitations for the use of AI in battlefield applications. Taken together, we argue that grant solicitations work as a vehicle for the mutual enlistment of DoD funding agencies and the academic AI research community in setting research agendas. The trope of basic research in this context offers shelter from significant moral questions that military applications of one's research would raise, by obscuring the connections that implicate researchers in U.S. militarism.
1207
Reposted by Amanda Bertsch
🌶 David Gray Widder @davidthewid.bsky.social · 09/12/2024
📢 NEW Paper! @siree.sh, Lucy Suchman, and I examine a corpus of 7,000 US Military grant solicitations to ask what the world’s largest military wants with to do with AI, by looking at what it seeks to fund. #STS 📄: arxiv.org/pdf/2411.17840 We find…
Basic Research, Lethal Effects: Military AI Research Funding as Enlistment David Gray Widder Digital Life Initiative, Cornell University Sireesh Gururaja School of Computer Science, Carnegie Mellon University Lucy Suchman Department of Sociology, Lancaster University Abstract In the context of unprecedented U.S. Department of Defense (DoD) budgets, this paper examines the recent history of DoD funding for academic research in algorithmically based warfighting. We draw from a corpus of DoD grant solicitations from 2007 to 2023, focusing on those addressed to researchers in the field of artificial intelligence (AI). Considering the implications of DoD funding for academic research, the paper proceeds through three analytic sections. In the first, we offer a critical examination of the distinction between basic and applied research, showing how funding calls framed as basic research nonetheless enlist researchers in a war fighting agenda. In the second, we offer a diachronic analysis of the corpus, showing how a ‘one small problem’ caveat, in which affirmation of progress in military technologies is qualified by acknowledgement of outstanding problems, becomes justification for additional investments in research. We close with an analysis of DoD aspirations based on a subset of Defense Advanced Research Projects Agency (DARPA) grant solicitations for the use of AI in battlefield applications. Taken together, we argue that grant solicitations work as a vehicle for the mutual enlistment of DoD funding agencies and the academic AI research community in setting research agendas. The trope of basic research in this context offers shelter from significant moral questions that military applications of one’s research would raise, by obscuring the connections that implicate researchers in U.S. militarism. Keywords: artificial intelligence; US Department of Defense; military; funding; investment, war
46632
Amanda Bertsch @abertsch.bsky.social · 25/11/2024
I think @siree.sh was also looking at this! No marker of arxiv category in the url, unfortunately :/
130
Amanda Bertsch @abertsch.bsky.social · 25/11/2024
and just realized this post is a full two weeks old but! bsky showed it to me now 🥲
110
Amanda Bertsch @abertsch.bsky.social · 25/11/2024
if the j*b m*rket takes you anywhere in a mining or old industrial area (incl most of the east coast), soil test before planting a garden in the ground! most universities will let you mail them like $30 and a bag of dirt to check it for lead and arsenic
250
Reposted by Amanda Bertsch
Vicki @vickiboykis.com · 20/11/2024
when you try to convert your text into smaller pieces but all it gives you is Elvish, that’s a tolkienizer
341006103
Reposted by Amanda Bertsch
Charles Sutton @randomlywalking.bsky.social · 21/11/2024
That’s right. You might think that all successful CS academics are good at running. But that’s only because the ones who weren’t, have been eaten by bears.
5804
Amanda Bertsch @abertsch.bsky.social · 21/11/2024
if we put our models through training runs, I think it's only fair that we do the same
1180
Reposted by Amanda Bertsch
Lindia Tjuatja @lindiatjuatja.bsky.social · 20/11/2024
💬 Have you or a loved one compared LM probabilities to human linguistic acceptability judgments? You may be overcompensating for the effect of frequency and length! 🌟 In our new paper, we rethink how we should be controlling for these factors 🧵:
Screenshot of the paper title "What Goes Into a LM Acceptability Judgment? Rethinking the Impact of Frequency and Length"
18519
Reposted by Amanda Bertsch
Sireesh Gururaja @siree.sh · 12/11/2024
I'm keeping track of people at the CMU Language Technologies Institute here: go.bsky.app/NhTwCVb. Follow along!
073
Reposted by Amanda Bertsch
Shaily @shaily99.bsky.social · 14/11/2024
Today is the day!! Find me at 2 PM in the Jasmine Hall (the one on the floor near food).
063
Amanda Bertsch @abertsch.bsky.social · 14/11/2024
Tim Dettmers has a great blog about this! There's a training+inference comparison in the "raw performance ranking" section: timdettmers.com/2023/01/30/w... (Though iirc many of these numbers are expected performance calculated from specs, not benchmarked performance)
timdettmers.com
The Best GPUs for Deep Learning in 2023 — An In-depth Analysis
Here, I provide an in-depth analysis of GPUs for deep learning/machine learning and explain what is the best GPU for your use-case and budget.
140
Reposted by Amanda Bertsch
Lindia Tjuatja @lindiatjuatja.bsky.social · 08/11/2024
(Hehe first bsky post!) I'll be at #EMNLP2024 💃🌴! Happy to chat about (among other things): ✨linguistically+cognitively motivated evaluation ✨NLP for low-resource+endangered languages ✨figuring out what features of language data LMs are *actually* learning I'll be presenting two posters 🧵:
1296
Reposted by Amanda Bertsch
Naomi Saphra @nsaphra.bsky.social · 08/11/2024
Taking a stand that we aren’t doing the #nlproc tag here. It’s #nlp. We used #nlproc because a decade ago the #nlp tag was full of sleazy scammers selling guides for hypnotizing women into sleeping with you. But guess what? We won. All the sleazy scammers are doing natural language processing now.
1118129
Amanda Bertsch @abertsch.bsky.social · 08/11/2024
glad it's useful! :) we have some additional results lying around w full fine-tuning; I'll update the arxiv version with those soon (nothing hugely surprising in those-- full FT is a bit better than LoRA, but long-context ICL is still generally more sample-efficient, especially with <1k samples)
010
Amanda Bertsch @abertsch.bsky.social · 07/11/2024
thank you for putting this together! can you add me to the list?
010
Reposted by Amanda Bertsch
Clara Na @clarana.bsky.social · 05/11/2024
Building/customizing your own LLM? You'll want to curate training data for it, but how do you know what makes the data good? You can try out recipes👩‍🍳 iterate on ✨vibes✨ but we can't actually test all possible combos of tweaks,,, right?? 🙅‍♂️WRONG! arxiv.org/abs/2410.15661 (1/n) 🧵
1498
Reposted by Amanda Bertsch
Clara Na @clarana.bsky.social · 12/10/2023
bsky.app/profile/sire... it was so interesting to see participants' perceptions of "paradigm shifts" paralleled across eras and cycles of NLP, and at the same time nothing until recent years had reached quite the level of *47%* of ACL papers in 2021 citing BERT
021
Reposted by Amanda Bertsch
Sireesh Gururaja @siree.sh · 12/10/2023
We all know that “recently large language models have”, “large language models are”, and “large language models can.” But *why* LLMs? How did we get here? (where is “here”?) What forces are shaping NLP, and how recent are they, actually? To appear at EMNLP 2023: arxiv.org/abs/2310.07715
Screenshot of paper title: "To Build Our Future, We Must Know Our Past: Contextualizing Paradigm Shifts in Natural Language Processing"
2174
Reposted by Amanda Bertsch
Gretchen McCulloch @gretchenmcculloch.com · 11/10/2023
Talking to the youths in 2023: did you know that "podcast" comes from a pun on "broadcast" plus the Apple iPod, a precursor to the iPhone that only played music Talking to the youths in 2043: did you know that "tweet" comes from a pun on Twitter, a precursor to various shortform social media
1226047