🔥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

Check out our new pre-print on statistically sound methodology to verify the quality of LLM-as-a-judge annotation. Nitay Calderon Rotem Dror

We’re excited to share that our paper, "NL-EYE: Abductive NLI for Images" has been accepted to ICLR 2025!🥳 Our benchmark reveals that VLMs struggle with abductive reasoning across multiple images (worse than random!) Think you can solve it? Check out these examples! Link⬇️

Our paper: "On Behalf of the Stakeholders: Trends in NLP Model Interpretability in the Era of LLMs" has been accepted to NAACL 2025 main 🎉 We’ve updated the final version: 🔗arxiv.org/abs/2407.19200 If you are an NLP interpretability researcher, we have key takeaways for you!👇

This was the result of a fun collaboration with Michael Toker Zorik Gekhman Roi Reichart , Idan Szpektor (Google), Hadas Kotek (Apple) and Yonatan Belinkov

Key passage in Bar’s response to Netanyahu’s allegations: “these are baseless claims which are nothing but a cover for totally different motives, alien and fundamentally wrong, meant to disrupt the security agency’s capability to fulfill its role… on behalf of Israel’s citizens”

Dramatic series of events in Israel since last night: *The government convenes to fire the head of the Shin Bet, Israel’s internal security service, after the Prime Minister says he has lost trust in him. Netanyahu argues he can't work with Ronen Bar, but does not use the

Netanyahu now going on a rant (in online video) against Shin Bet Chief Ronen Bar and Attorney-General accusing them of collusion in launching the Qatar-Gate investigation so they can avoid being fired. Promises he’ll get Bar fired, prevent a civil war and keep Israel a democracy>

Check out our new paper on adapting LLMs to focused domains with vocabulary adaptation Itay Nakash Nitay Calderon Eyal Ben-David and Elad Hoffer

New paper ! We propose a way to quantify the knowledge encoded in LLMs, and reveal that much more knowledge is encoded in the parameters of the model than what can be directly observed from its output. Work of Zorik Gekhman and our Google DeepMind and Technion collaborators.

The TACL website was down in the last 24 hours due to a technical problem. I am happy to update that the problem is solved now and the website is back on air. Apologies for the inconvenience.

Our paper won the NAACL 2025 SAC award for the Interpretability and Analysis of Models for NLP area. Warm congratulations to Nitay Calderon ! The presentation is on the May 1 10:30-12 oral session. Award will be handed in the best paper planery session. Say hello to Nitay Calderon at NAACL.

What a great ending to #NAACL2025! 🏆🥇 Our paper won the Interpretability and Analysis of Models for NLP SAC Award! Had a great time attending, presenting our research, meeting brilliant people, and exploring New Mexico 🏜️ Roi Reichart

In a world dominated by GenAI, we forgot about the foundations of Machine Learning. It might sound crazy, but it's 2025 and we still use decision trees for the most fundamental tasks. We present TabSTAR, a Foundation Tabular Model🌟Eilam Shapira Roi Reichart arxiv.org/abs/2505.18125

Tabular data with textual features challenges deep learning, often using static text embeddings unlike gradient boosting decision trees (GBDTs). TabSTAR addresses this with an adaptable text encoder and target-aware tokens for classification tasks, outperforming existing methods

2 Models, 1 datasets and 1 eval of the week to pay attention to: ▪️ ARM: Adaptive reasoning model ARM ▪️ TabSTAR ▪️ rStar-Coder dataset ▪️ REWARDBENCH 2 benchmark 🧵

Preferences drive modern LLM research and development: from model alignment to evaluation. But how well do we understand them? Excited to share our new preprint: Multi-domain Explainability of Preferences arxiv.org/abs/2505.20088 Roi Reichart Liat 🧵👇 1/11

🔥הסקירות ממשיכות לזרום ל-X🔥 🧵 המאמר היומי של מייק: 25.06.25 The Alternative Annotator Test for LLM-as-a-Judge: How to Statistically Justify Replacing Human Annotators with LLMs מאמר 🇮🇱 תפנית מעניינת מתרחשת בתקופה האחרונה בעולם של הערכת ביצועי מודלים. אנחנו כבר לא שואלים רק

I had the pleasure to present our Tabular Foundation Model, TabSTAR, in the explAInable podcast, discuss the challenges of the field, the role of text and the future of tabular learning. Eilam Shapira Roi Reichart Uri Goren Mike Erlihson, Math PhD, AI Link: open.spotify.com/episode/3rj6OU…

Now accepted to #COLM2025! We formally define hidden knowledge in LLMs and show its existence in a controlled study. We even show that a model can know the answer yet fail to generate it in 1,000 attempts 😵 Looking forward to presenting and discussing our work in person.