🔥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

adapter-transformers is now ✨Adapters✨ All adapter implementations, composition features, training & hub tooling have found a new home in the new adapters package, an add-on to the great Transformers lib How to enable adapter magic 🧙‍♂️ in a Transformer with 1 line of code ⬇️

Pointwise V-usable information (PVI) excels in many #NLProc tasks. But fine-tuning #LLMs with it is very time-consuming 🐌 Is in-context PVI the necessary next step? Yes! 🎉Check out our empirical analysis accepted at #EMNLP2023 – and this 🧵 (1/7) 📄 arxiv.org/abs/2310.12300

This is going to be a busy month. To start off, 🚨accepted as poster at ICLR 2026: arxiv.org/abs/2310.02832!🚨 Great work by Fran Jelenić & Josip Jukic proposing a novel OOD detection method suitable for local LLMs (only req. access to params). Thread & CR coming soon.

Really excited that our work on model merging has been accepted as a poster to ICLR 2024! There have been many discussions recently on model merging on here and many people are questioning why it should work. We try to answer exactly this, as briefly summarized in this thread.

I am at the #EACL2024 today and welcome anyone who wants to talk or is interested in employment opportunities with us. Follow our profile UKP Lab for more info on our contributions to this year's eaclmeeting.

We are proud to announce that the contribution »Sensitivity, Performance, Robustness: Deconstructing the Effect of Sociodemographic Prompting« by Tilman Beck, Hendrik Schuff, Anne Lauscher (she/her) (Universität Hamburg) and Iryna Gurevych (UKP Lab) has just been awarded the #EACL2024 Social Impact Award!

Five papers authored or co-authored by UKP members have been accepted for publication at this year's @NAACLmeeting, the Conference of the North American Chapter of the ACL (@NAACL). Congratulations to everyone involved! 💐 #NAACL2024 informatik.tu-darmstadt.de/ukp/ukp_home/u…

If you're interested in chatting about model merging, unlearning, etc. please stop by our poster tomorrow at ICLR in poster session 1 (10:45-12:45)!

Stop complaining about the bad review quality. Join forces and start research on #NLProc for #PeerReview! 🚨 A new white paper by over 20 top AI and NLP researchers provides a thorough discussion of AI assistance for scientific quality control. (1/🧵) 📑 arxiv.org/abs/2405.06563

Happening tomorrow 9AM at the MWE-UD Workshop (MWE Workshop) @ LREC-COLING! Harish will deliver the first keynote on semantics and reasoning of LLMs 🤖📖

Andreas Waldis HSLU Hochschule Luzern Yufang Hou IBM Research Iryna Gurevych Md. Imbesat Hassan Rizvi Xiaodan Zhu ECE Queens Ingenuity Labs Indraneil Paul Goran Glavaš Universität Würzburg #UniWürzburg Jonibek Mansurov Jinyan Su Artem Shelmanov Akim Tsvigun Osama Afzal Alham Fikri Aji @nyhabash Preslav Nakov MBZUAI TU Darmstadt @IstiCnr NYU Abu Dhabi Furkan Şahinuç Ilia Kuznetsov »Are Emergent Abilities in Large Language Models just In-Context Learning?« by Sheng Lu (UKP Lab), Irina Bigoulaeva, @Rachneet4 Singh Sachdeva, @Harish Tayyar Madabushi (BathNLP) and Iryna Gurevych (8/🧵) #ACL2024NLP (arXiv coming soon) x.com/UKPLab/status/…

If you’re at ACL and interested in learning more about emergent abilities in LLMs, please meet my amazing colleagues: Bob Irina Bigoulaeva Harish Tayyar Madabushi

Happening now

Indeed :)

Re: “Every major breakthrough in AI has been American”: America does itself no favors when it overestimates its specialness. Yes, the center of the AI industry is the US (California!), but many of the breakthroughs of (neural, gradient-based) AI happened elsewhere: • LSTMs,

This has got to be my fav part of the claude 3.7 release. 😃😆😇 Best eval ever. 🤣 Congrats to Anthropic and my friends there for the cool model.

A bit of a mess around the conflict of COLM with the ARR (and to lesser degree ICML) reviews release. We feel this is creating a lot of pressure and uncertainty. So, we are pushing our deadlines: Abstracts due March 22 AoE (+48hr) Full papers due March 28 AoE (+24hr) Plz RT 🙏

The hardest part about finetuning LLMs is that people generally don't have high-quality labeled data. Today, Databricks introduced TAO, a new finetuning method that only needs inputs, no labels necessary. Best of all, it actually beats supervised finetuning on labeled data.

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 🔍

What a finish! Gemini 2.5 Pro just completed Pokémon Blue!  Special thanks to Joel Z for creating and running the livestream, and to everyone who cheered Gem on along the way.