🔥The 9th Round of Easy Loan, Earn $40 Reward is in progress❗️ ⏰ Promotion Period: January 15th - Feburary 15th, 2025 👉 Register now and check more details at gate.io/campaigns/358

Hi, I gathered a ton of XAI papers (>5k) and wrote up an empirical trends report. check it out here!! Arxiv: arxiv.org/abs/2301.05433 Blog: medium.com/@alonjacovi/tr… Data: github.com/alonjacovi/XAI…

Finally, in "Multilingual Sequence-to-Sequence Models for Hebrew NLP" (arxiv.org/pdf/2212.09682…) we show that multilingual models like mT5 can go a long way for Hebrew when modeled right, with lots of strong results. With Matan Eyal Hila Noga هيله نوغا Reut Tsarfaty and Idan Szpektor

A very neat idea to a very important problem

"semantic embeddings" are becoming increasingly popular, but "semantics" is really ill-defined. sometimes you want to search for text given a description of its content. current embedders suck at this. in this work we introduce a new embedder. Shauli Ravfogel Valentina Pyatkin ➡️ ICML Avshalom Manevich

Happy to share ZeroSCROLLS, a zero-shot benchmark for long text understanding! 📜Paper arxiv.org/pdf/2305.14196… 📜 Leaderboard zero.scrolls-benchmark.com 📜 Data (inputs only) huggingface.co/datasets/tau/z… Maor Ivgi Avia Efrat Jonathan Berant Omer Levy 1/5

Multilingual Instruction Tuning With Just a Pinch of Multilinguality paper page: huggingface.co/papers/2401.01… As instruction-tuned large language models (LLMs) gain global adoption, their ability to follow instructions in multiple languages becomes increasingly crucial. One

Coming into this project we had a lot of open questions when ITing in multilingual settings Can LLMs Xligual-Transfer in IT❓ How many multilingual examples should we curate❓ How many languages should we curate❓ And so on... Great work from Uri giving great insights on these!

so what’s up with tokenization? why and how does it work? welcome to a🚨new paper🚨 thread! "Unpacking Tokenization: Evaluating Text Compression and its Correlation with Model Performance" w/ Avi Caciularu Matan Eyal Kris Cao idan szpektor and Reut Tsarfaty arxiv.org/abs/2403.06265 1/🧵

Does Fine-Tuning LLMs on New Knowledge Encourage Hallucinations? New preprint!📣 - LLMs struggle to integrate new factual knowledge through fine-tuning - As the model eventually learns new knowledge, it becomes more prone to hallucinations😵‍💫 📜arxiv.org/pdf/2405.05904 🧵1/12👇

Glad to announce that these cool papers were accepted at #ICLR2025!! 💫 BOND: Aligning LLMs with Best-of-N Distillation arxiv.org/abs/2407.14622 Diversity-Rewarded Classifier-Free Guidance Distillation arxiv.org/abs/2410.06084 🧵

So proud to have been part of the team bringing Gemma 3 to life, working on its multilingual capabilities! 🌎 It's amazing to see the impact of everyone's hard work #Gemma3 #Multilingual

Meet Ai2 Paper Finder, an LLM-powered literature search system. Searching for relevant work is a multi-step process that requires iteration. Paper Finder mimics this workflow — and helps researchers find more papers than ever 🔍

I really enjoyed working on this paper! Towards more challenging multilingual research questions!

Introducing ECLeKTic, a new benchmark for Evaluating Cross-Lingual Knowledge Transfer in LLMs. It uses a closed-book QA task, where models must rely on internal knowledge to answer questions based on information captured only in a single language. More →goo.gle/3Y5TqvZ

our research featured in a Google Research blog post! Uri Shaham Matan Eyal

Are you in ACL? Don't skip on having a chat with Omer about our work!