🔥 New Preprint! 🔥 How should LLMs handle ambiguous questions in text-to-SQL semantic parsing? 👉🏼 Disambiguate First, Parse Later! We propose a plug-and-play approach that explicitly disambiguates the question 💬 Paper: https://t.co/wtGBpGaElb

🎤 @serj_troshin will present this work November 9th at 09:55-10:30 (poster lightning talk) and 11:00 - 12:15 (poster), Room A207 📄 Paper: https://t.co/4MRhKuVjxs 💻 Code: https://t.co/vbMbhOl5oN 3/3

We compare these strategies on: ✔️ Quality & efficiency ✔️ Lexical diversity ✔️ Computational flow diversity (a new metric we propose that's more suitable for reasoning tasks like math) Key finding: non-independent sampling boosts diversity without sacrificing quality! ✨ 2/3

Excited to share our paper "Asking a Language Model for Diverse Responses" accepted at #UncertaiNLP workshop at #EMNLP2025 When generating diverse responses, most use parallel sampling. We study non-independent alternatives: enumeration & iterative sampling. w/ @serj_troshin 👇

At #EMNLP2025 to present the last chapter of my PhD 🐼 Let's talk #HateSpeech detection, generalisation and NLP safety at my poster: 📆tomorrow 🕟4.30pm Look for the circus-themed poster 🎪🤸🏻‍♀️ Work with @tomsherborne @bjoernross and @mlapata at @EdinburghNLP + @cohere

Excited to share my first work as a PhD student at @EdinburghNLP that I will be presenting at EMNLP! RQ1: Can we achieve scalable oversight across modalities via debate? Yes! We show that debating VLMs lead to better model quality of answers for reasoning tasks.

Represent! ✌️

At the @aclmeeting Panel on Generalization of NLP Models, Mirella @mlapata argues that the real problem isn’t generalization but how to get models that learn and adapt in real time. That’s a pretty good requirement for true intelligence!

Well deserved 👏

Interested in text-to-SQL or ambiguity? Curious how we can turn LLM overconfidence into an advantage? Let's talk! Come say hi 👋

Excited to present our Findings paper at #ACL2025 in Vienna next week! 📄Disambiguate First, Parse Later: Generating Interpretations for Ambiguity Resolution in Semantic Parsing 🗓️ Tue, July 29, 10:30–12:00 📍 Hall 4/5, Session 7

Preprint: Can we learn to reason for story generation (~100k tokens), without reward models? Yes! We introduce an RLVR-inspired reward paradigm VR-CLI that correlates with human judgements of quality on the 'novel' task of Next-Chapter Prediction. Paper: https://t.co/eO0nUHzRjG

(5/5) 💻 Check out the code:

(4/5) Experiments on AmbiQT & Ambrosia in both in-domain and out-of-domain settings show: 📌 SFT easily overfits 📌 Disambiguate first, parse later works better! 📌 Infilling improves performance 📌 Ambiguity remains a hard, low-resource problem!

(3/5) Advantages: ✅ Interpretations act as transparent, explainable reasoning steps ✅ Plug-and-play approach ✅ Handles both ambiguous & unambiguous queries

(2/5) LLMs struggle with ambiguity and default to a single preferred interpretation. We turn this bias into an advantage! Our approach: 1️⃣ Generate default NL interpretations using an LLM 2️⃣ Infill missing interpretations 3️⃣ Parse each into SQL (standard text-to-SQL)

Congratulations to Dr. Parag Jain @jparag123 🥳🎓 Well deserved!

LLMs can tackle math olympiad probs but... can they read a clock 🤔? 🕰️📆 Our experiments reveal surprising failures in temporal reasoning—MLLMs struggle with analogue clock reading & date inference! Lost in Time: Clock and Calendar Understanding Challenges in Multimodal LLMs

At #NeurIPS2024 with such amazing people!

See you tomorrow (Thursday) from 11 pm to 2 pm in West Ballroom A-D #5309 🚀