Clara Na
@claranahhh
PhD student at @LTIatCMU / @SCSatCMU she/her, prev. @UVA and intern @ai2_allennlp @/clara on http://sigmoid.social and @/clarana on http://bsky.app
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) 🧵
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
!!! I'm at #ICLR2025 to present 🧄Aioli🧄 a unified framework for data mixing on Thursday afternoon! 🔗 arxiv.org/abs/2411.05735 Message me to chat about pre/post training data (mixing, curriculum, understanding); test-time compute/verification; or to try new food 🇸🇬
coming to a NAACL 2025 near you! 🌞 Looking forward to discussing with folks in Albuquerque :) The camera-ready is on arxiv now, with more models, more tasks, and more compared settings-- including results comparing ICL to full finetuning! arxiv.org/abs/2405.00200
In-context learning provides an LLM with a few examples to improve accuracy. But with long-context LLMs, we can now use *thousands* of examples in-context. We find that this long-context ICL paradigm is surprisingly effective– and differs in behavior from short-context ICL! 🧵
Check out our work on improving LLM's ability to seek information through asking better questions! 💫
Asking the right questions can make or break decisions in high-stake fields like medicine, law, and beyond✴️ Our new framework ALFA—ALignment with Fine-grained Attributes—teaches LLMs to PROACTIVELY seek information through better questions🏥❓ (co-led with @jiminmun_) 👉🏻🧵
Did you know? Gestures to express universal concepts—like wishing for luck—vary WIDELY across cultures? 🤞means luck in US but deeply offensive in Vietnam 🚨 📣We introduce MC-SIGNS, a test bed to evaluate how LLMs/VLMs/T2I handle such nonverbal cues 📜: arxiv.org/abs/2502.17710
🚨😱Obligatory job market announcement post‼️🤯 I'm searching for faculty positions/postdocs in multimodal/multilingual NLP and generative AI! I'll be at #NeurIPS2024 presenting our work on meta-evaluation for text-to-image faithfulness! Let's chat! Website in bio, papers in🧵
My lab at Duke has multiple Ph.D. openings! Our mission is to augment human decision-making by advancing the reasoning, comprehension, and autonomy of modern AI systems. I am attending #emnlp2024, happy to chat about PhD applications, LLM agents, evaluation etc etc!
💬 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 🧵:
Thank you so much @emnlpmeeting for this wonderful recognition! I’m so honored and humbled 💕 Thanks @gneubig for your support throughout! We’ve been working on this for 1.5 years and everyone who has spoken with me in the recent past knows how passionately I feel about this…
#EMNLP2024 Best Paper 1/5: An image speaks a thousand words, but can everyone listen? On image transcreation for cultural relevance
I'm at EMNLP! Presenting the poster for this paper on Thursday morning (10:30-12), Session F Riverfront Hall, come say hi :)
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) 🧵
There are many algorithms for constructing pre-training data mixtures—which one should we use? Turns out: many of them fall under one framework, have similar issues, and can be improved with a straightforward modification. Introducing Aioli! 🧄 1/9
I’m thrilled to be at EMNLP this week presenting our paper, “The Empirical Variability of Narrative Perceptions of Social Media Texts” I’ll be giving an oral presentation during the CSS + Cultural Analytics Session 2 (Nov 14). Paper: aclanthology.org/2024.emnlp-mai… 🧵(1/12)
My last PhD project at CMU will be presented today and tomorrow at @corl_conf as one of the Demos: SIAT (SIAT: Stretch control with Immersive AR Teleoperation) Find @viddivj, Su Li and @ybisk who will be presenting this work in person. The original goal of me doing this…
Connecting our world 🇺🇲🌍 Proud to present ✨ Aya Expanse✨ Text Aya over SMS in the US: +18332746219 Or try it on WhatsApp worldwide: +14313028498
Do you want to select great LLM pretraining data but don’t have 1000 H100s for a ton of mixture experiments? What about a method that requires none of your own training, matches the best known existing method, and has some nice theory? New preprint: Perplexity Correlations
One of the most useful applications of model merging is approximating training data ablations. Check out this cool work led by @claranahhh, and talk to us at #EMNLP2024 if you are there.
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) 🧵