Excited to share that our paper 'Identifying Linear Relational Concepts in LLMs' with Oana-Maria Camburu and Anthony Hunter has been accepted to #NAACL2024 ! Details 👇
Paper: arxiv.org/abs/2311.08968
Code: github.com/chanind/linear…
See you in Mexico! 🇲🇽
#XAI #MechanisticInterpretability
1. 🧵🎉 Excited to share that our paper 'Sequential Compositional Generalization in Multimodal Models' is accepted as a long paper at #NAACL2024 ! 🌟 We'll be presenting our findings in Mexico City this June (NAACL HLT 2024). Dive into the full details here 👇
- Paper:
📢Our work X-posing Free Speech: Examining the Impact of Moderation Relaxation on Online Social Networks studying Elon Musk's takeover of #Twitter #X accepted Workshop on Online Abuse and Harms NAACL HLT 2024 #NAACL2024 w/ Arvindh Arun Saurav Chhatani Jisun An
📜arxiv.org/pdf/2404.11465…
Findings 🧵👇
Very excited and equally proud of this work! Spent a lot of time on data collection & systematic eval of several model paradigms (sft/n-shot), with the most surprising result: complementary perf bet human & AI. Go Sanchaita Hazra!! We'll be in Mexico City, hit us up! #NAACL2024
Checkout our #NAACL2024 paper on multilingual text editing 👇
We present results on Grammatical Error Correction, Text Simplification, and Paraphrasing on seven languages (Arabic, Chinese, English, German, Japanese, Korean, and Spanish)
🌟 Exciting News Alert! 🌟 Thrilled to announce our lineup of confirmed keynote speakers for the Mexican NLP Summer School 2024! 🎉
#MexicanNLPSummerSchool #NAACL2024
AMPLN
Red TTL
Performance as a function of tuning effort is called *the tuning curve*. If you haven't yet, you can learn more from the #NAACL2024 paper by He He, Kyunghyun Cho, and me: 'Show Your Work with Confidence'!
arxiv: arxiv.org/abs/2311.09480
og tweet: x.com/NickLourie/sta….
4/4
🤔 Diãcritics: Translation's Friend ôr Foe?
🚀 We are happy to share our work “Interplay of Machine Translation, Diacritics and Diacritization”, which will appear at #NAACL2024 main conference.
Diacritics are markers attached to characters (e.g. the circle on å, the hat on ê).
Our paper “Towards Automated Document Revision: Grammatical Error Correction, Fluency Edits, and Beyond” got accepted in BEA 2024 (co-located in #NAACL2024 )! w/ Keisuke Sakaguchi, Masato Hagiwara, みずもともや, Jun Suzuki, Kentaro Inui / 乾健太郎
Excited to share that latest paper was accepted at #NAACL2024 Industry Track!
In this work, we tackled the challenge of hallucination for the task of text to workflow generation.
arxiv.org/abs/2404.08189🧵
Paper from The Sanghani Center at Virginia Tech accepted to the #NAACL2024 which will be held in Mexico City in June.
Authors are Mandar Rutuja Taware Pravesh Koirala Nikhil Muralidhar Naren Ramakrishnan Read about their research here: arxiv.org/pdf/2404.01536…
Haize Labs Cool stuff! I believe Maksym Andriushchenko 🇺🇦 also explored this llama3 vulnerability concurrently in the below thread and the same vulnerability for Claude models in his earlier paper about pre-filling x.com/maksym_andr/st…
Happy to share that our paper on open-domain health question answering was accepted to Findings of #NAACL2024 ! 🇲🇽
We show that QA performance improves when retrieving fewer evidence documents and filtering them by publication year and reputation.🧮
📜: arxiv.org/abs/2404.08359
Ashwinee Panda Haize Labs Maksym Andriushchenko 🇺🇦 Ah, but I guess I didn't look closely enough, perhaps he didn't explore pre-filling in the latest attack on llama3 -- I think it mostly just reminded me of the pre-filling attack on Claude from Maksym's earlier paper arxiv.org/abs/2404.02151