Papers by Michael Zhang

RWKV: Reinventing RNNs for the Transformer Era (2023.findings-emnlp)

Challenge:	Existing fashion recommendation systems struggle with the unique challenges of the fashion domain.
Approach:	They propose a sequential fashion recommendation framework that leverages a pre-trained large language model enhanced with recommendation-specific prompts.
Outcome:	The proposed framework significantly improves fashion recommendation performance on Amazon fashion.

Copied to clipboard

Bo Peng, Eric Alcaide, Quentin Anthony, Alon Albalak, Samuel Arcadinho, Stella Biderman, Huanqi Cao, Xin Cheng, Michael Chung, Leon Derczynski, Xingjian Du, Matteo Grella, Kranthi Gv, Xuzheng He, Haowen Hou, Przemyslaw Kazienko, Jan Kocon, Jiaming Kong, Bartłomiej Koptyra, Hayden Lau, Jiaju Lin, Krishna Sri Ipsit Mantri, Ferdinand Mom, Atsushi Saito, Guangyu Song, Xiangru Tang, Johan Wind, Stanisław Woźniak, Zhenyuan Zhang, Qinghua Zhou, Jian Zhu, Rui-Jie Zhu

Mitigating Temporal Misalignment by Discarding Outdated Facts (2023.emnlp-main)

Challenge:	recurrent neural networks struggle to match the performance of Transformers due to limitations in parallelization and scalability.
Approach:	They propose a model architecture that combines the efficient parallelizable training of transformers with the efficient inference of RNNs.
Outcome:	The proposed model performs on par with similarly sized RNNs, suggesting future work can leverage this architecture to create more efficient models.

Copied to clipboard

Michael Zhang, Eunsol Choi

Question Answering

Challenge:	Temporal misalignment is a problem for knowledge-intensive tasks where models must rely on data from the past to make predictions.
Approach:	They propose a temporal misalignment task to predict how long a given fact will remain true.
Outcome:	The proposed task improves calibration for knowledge-intensive tasks under temporal misalignment by discarding volatile facts.

The TechQA Dataset (2020.acl-main)

Copied to clipboard

Vittorio Castelli, Rishav Chakravarti, Saswati Dana, Anthony Ferritto, Radu Florian, Martin Franz, Dinesh Garg, Dinesh Khandelwal, Scott McCarley, Michael McCawley, Mohamed Nasr, Lin Pan, Cezar Pendus, John Pitrelli, Saurabh Pujar, Salim Roukos, Andrzej Sakrajda, Avi Sil, Rosario Uceda-Sosa, Todd Ward, Rong Zhang

Low-Resource NLP

Challenge:	TECHQA is a domain-adaptation question answering dataset for the technical support domain.
Approach:	They propose a domain-adaptation question-answering dataset for the technical support domain that contains actual questions posed by users on a technical forum .
Outcome:	The TECHQA dataset highlights two real-world issues from the automated customer support domain.

The Death and Life of Great Prompts: Analyzing the Evolution of LLM Prompts from the Structural Perspective (2024.emnlp-main)

Copied to clipboard

Yihan Ma, Xinyue Shen, Yixin Wu, Boyang Zhang, Michael Backes, Yang Zhang

Breaking Agents: Compromising Autonomous LLM Agents Through Malfunction Amplification (2025.emnlp-main)

Challenge:	Recent research has shown that high-quality prompts are essential for LLMs to produce accurate and relevant responses.
Approach:	They analyze 10,538 in-the-wild prompts collected from various platforms and develop a framework that decomposes the prompts into eight key components.
Outcome:	The proposed framework decomposes 10,538 in-the-wild prompts into eight components.

Copied to clipboard

Boyang Zhang, Yicong Tan, Yun Shen, Ahmed Salem, Michael Backes, Savvas Zannettou, Yang Zhang

CommonLID: Re-evaluating State-of-the-Art Language Identification Performance on Web Data (2026.acl-long)

Challenge:	Recent advances in large language models (LLMs) have increased the vulnerability of LLMs, but they can cause more severe damage than standalone systems if compromised.
Approach:	They propose a new type of attack that induces malfunctions by misleading the agent into executing repetitive or irrelevant actions.
Outcome:	The proposed attacks induce failure rates exceeding 80% in multiple scenarios, highlighting the substantial risks associated with this vulnerability.

Copied to clipboard

Pedro Ortiz Suarez, Laurie Burchell, Catherine Arnett, Rafael Mosquera, Sara Hincapié Monsalve, Thom Vaughan, Damian Stewart, Malte Ostendorff, Idris Abdulmumin, Vukosi Marivate, Shamsuddeen Hassan Muhammad, Atnafu Lambebo Tonja, Hend Al-Khalifa, Nadia Ghezaiel Hammouda, Verrah Akinyi Otiende, Tack Hwa Wong, Jakhongir Saydaliev, Melika Nobakhtian, Muhammad Ravi Shulthan Habibi, Chalamalasetti Kranti, Carol Muchemi, Khang Nguyen, Faisal Muhammad Adam, Luis Frentzen Salim, Reem Alqifari, Cynthia Jayne Amol, Joseph Marvin Imperial, Ilker Kesen, Ahmad Mustafid, Pavel Stepachev, Leshem Choshen, David Anugraha, Hamada Nayel, Seid Muhie Yimam, Vallerie Alexandra Putra, My Chiffon Nguyen, Azmine Toushik Wasi, Gouthami Vadithya, Rob Van Der Goot, Lanwenn ar C’horr, Karan Dua, Andrew Yates, Mithil Bangera, Yeshil Bangera, Hitesh Laxmichand Patel, Shu Okabe, Fenal Ashokbhai Ilasariya, Dmitry Gaynullin, Genta Indra Winata, Yiyuan Li, Juan Pablo Martínez, Amit Agarwal, Ikhlasul Akmal Hanif, Raia Abu Ahmad, Esther Adenuga, Filbert Aurelian Tjiaranata, Weerayut Buaphet, Michael Anugraha, Sowmya Vajjala, Benjamin L Rice, Azril Hafizi Amirudin, Jesujoba Oluwadara Alabi, Srikant Panda, Yassine Toughrai, Bruhan Kyomuhendo, Daniel Ruffinelli, null Akshata, Manuel Goulão, Ej Zhou, Ingrid Gabriela Franco Ramirez, Cristina Aggazzotti, Konstantin Dobler, Jun Kevin, Quentin Pagès, Nicholas Andrews, Nuhu Ibrahim, Mattes Ruckdeschel, Amr Keleg, Mike Zhang, Casper Rufaro Muziri, Saron Samuel, Sotaro Takeshita, Kun Kerdthaisong, Luca Foppiano, Rasul Dent, Tommaso Green, Ahmad Mustapha Wali, Kamohelo Makaaka, Vicky Feliren, Inshirah Idris, Hande Celikkanat, Abdulhamid Abubakar, Jean Maillard, Benoît Sagot, Thibault Clérice, Kenton Murray, Sarah K. K. Luger

Text Classification

Challenge:	Language identification (LID) is a fundamental step in curating multilingual corpora.
Approach:	They introduce CommonLID, a community-driven, human-annotated LID benchmark for the web domain, covering 109 languages.
Outcome:	The proposed benchmark covers 109 languages and shows that existing evaluations overestimate accuracy for many languages in the web domain.

PRBench: Large-Scale Expert Rubrics for Evaluating High-Stakes Professional Reasoning (2026.acl-long)

Copied to clipboard

Afra Feyza Akyürek, Advait Gosai, Chen Bo Calvin Zhang, Vipul Gupta, Jaehwan Jeong, Anisha Gunjal, Tahseen Rabbani, Maria Mazzone, David Randolph IV, Mohammad Mahmoudi Meymand, Gurshaan Chattha, Paula Rodriguez, Diego A. Mares Buendia, Pavit Singh, Michael Liu, Subodh Chawla, Peter Cline, Lucy Ogaz, Ernesto Gabriel Hernández Montoya, Zihao Wang, Pavi Bhatter, Marcos Ayestaran, Bing Liu, Yunzhong He

Reasoning

Challenge:	Frontier models often lack a view of performance on open-ended, economically consequential tasks in high-stakes professional domains where practical returns matter most.
Approach:	They introduce a professional reasoning benchmark that recruits 182 qualified professionals to contribute questions inspired by their workflows.
Outcome:	The proposed model outperforms other models in 114 countries and 47 US jurisdictions on hard subsets.

A Computational Method for Measuring Open Codes in Qualitative Analysis (2026.findings-acl)

Copied to clipboard

John Chen, Alexandros Nikolaos Lotsos, Sihan Cheng, Lexie Zhao, Yanjia Zhang, Jessica Hullman, Bruce Sherin, Uri Wilensky, Michael Horn

REIC: RAG-Enhanced Intent Classification at Scale (2025.emnlp-industry)

Challenge:	Qualitative analysis is widely adopted across many social science disciplines.
Approach:	They propose a theory-informed computational method for measuring inductive coding results from humans and GAI.
Outcome:	The proposed method captures breadth, consensus, unique contribution, and systematic deviation without assuming ground truth.

Copied to clipboard

Ziji Zhang, Michael Yang, Zhiyu Chen, Yingying Zhuang, Shu-Ting Pi, Qun Liu, Rajashekar Maragoud, Vy Nguyen, Anurag Beniwal

JailbreakRadar: Comprehensive Assessment of Jailbreak Attacks Against LLMs (2025.acl-long)

Challenge:	Accurate intent classification is critical for efficient routing in customer service . however, as companies expand their product lines, intent classification faces scalability challenges .
Approach:	They propose a retrieval-augmented generation Enhanced Intent Classification approach which leverages retrieval augmented generation to integrate relevant knowledge into a model.
Outcome:	The proposed approach outperforms fine-tuning, zero-shot, and few-shot methods on real-world datasets.

Copied to clipboard

Junjie Chu, Yugeng Liu, Ziqing Yang, Xinyue Shen, Michael Backes, Yang Zhang

Composite Backdoor Attacks Against Large Language Models (2024.findings-naacl)

Challenge:	Large language models (LLMs) have been used to mitigate misuse and to align with human values.
Approach:	They propose to use large-scale evaluations of various jailbreak attacks to identify key patterns and test them under eight advanced defenses.
Outcome:	The proposed attacks achieve high success rates but are easy to mitigate by defenses.

Copied to clipboard

Hai Huang, Zhengyu Zhao, Michael Backes, Yun Shen, Yang Zhang

Are We in the AI-Generated Text World Already? Quantifying and Monitoring AIGT on Social Media (2025.acl-long)

Challenge:	Large language models (LLMs) have demonstrated superior performance on various tasks, but untrustworthy third-party LLMs may covertly introduce vulnerabilities for downstream tasks.
Approach:	They propose a composite backdoor attack that scatters multiple trigger keys in different prompt components.
Outcome:	The proposed attack achieves 100% Attack Success Rate (ASR) with a False Triggered Rate (FTR) below 2.06% and negligible model accuracy degradation.

Copied to clipboard

Zhen Sun, Zongmin Zhang, Xinyue Shen, Ziyi Zhang, Yule Liu, Michael Backes, Yang Zhang, Xinlei He

Text Generation

Challenge:	Social media platforms are experiencing a growing presence of AI-Generated Texts (AIGTs) however, the misuse of AIGTs could have profound implications for public opinion .
Approach:	They collect a dataset with 2.4M posts from 3 major social media platforms . they then construct a diverse dataset to train and evaluate AIGT detectors .
Outcome:	The proposed dataset analyzes 2.4M posts from 3 major social media platforms from 2022 to 2024 . it finds that Medium and Quora show marked increases in AAR .

Commonsense Inference in Natural Language Processing (COIN) - Shared Task Report (D19-60)

Copied to clipboard

Simon Ostermann, Sheng Zhang, Michael Roth, Peter Clark

Reasoning

Challenge:	The workshop on Commonsense Inference in NLP (COIN) evaluated text understanding systems' ability to draw inferences about facts that are not mentioned in the text, but that are assumed to be common ground.
Approach:	They propose to use commonsense knowledge to evaluate systems' ability to answer questions/queries about a text.
Outcome:	The proposed tasks evaluated systems in two contexts: Commonsense Inference and Commonsensible Inference.

ProPy: Building Interactive Prompt Pyramids upon CLIP for Partially Relevant Video Retrieval (2025.findings-emnlp)

Copied to clipboard

Yi Pan, Yujia Zhang, Michael Kampffmeyer, Xiaoguang Zhao

Low-Resource NLP

Challenge:	Existing models for PRVR use unimodal features, but powerful pretrained vision-language models like CLIP are underexplored.
Approach:	ProPy is a model with systematic architectural adaptation of CLIP specifically designed for PRVR.
Outcome:	ProPy outperforms existing models on three public datasets in terms of performance on the datasets.

BinaryBERT: Pushing the Limit of BERT Quantization (2021.acl-long)

Copied to clipboard

Haoli Bai, Wei Zhang, Lu Hou, Lifeng Shang, Jin Jin, Xin Jiang, Qun Liu, Michael Lyu, Irwin King

RESIN-11: Schema-guided Event Prediction for 11 Newsworthy Scenarios (2022.naacl-demo)

Challenge:	Recent pre-trained language models have achieved remarkable performance improvement in various tasks, but the improvement generally comes at the cost of increasing model size and computation.
Approach:	They propose a binary quantization technique which initializes binaryBERT by splitting from a ternary network.
Outcome:	The proposed model achieves state-of-the-art performance on the GLUE and SQUAD benchmarks while being 24x smaller.

Copied to clipboard

Xinya Du, Zixuan Zhang, Sha Li, Pengfei Yu, Hongwei Wang, Tuan Lai, Xudong Lin, Ziqi Wang, Iris Liu, Ben Zhou, Haoyang Wen, Manling Li, Darryl Hannan, Jie Lei, Hyounghun Kim, Rotem Dror, Haoyu Wang, Michael Regan, Qi Zeng, Qing Lyu, Charles Yu, Carl Edwards, Xiaomeng Jin, Yizhu Jiao, Ghazaleh Kazeminejad, Zhenhailong Wang, Chris Callison-Burch, Mohit Bansal, Carl Vondrick, Jiawei Han, Dan Roth, Shih-Fu Chang, Martha Palmer, Heng Ji

Information Extraction & Text Mining

Challenge:	Existing methods for event prediction are incomplete and noisy.
Approach:	They propose to use news-related event schemas to extract newsworthy events . they build a demo website and include a video demonstrating the framework .
Outcome:	The proposed framework can be applied to a wide variety of newsworthy scenarios.

Graph-Guided Textual Explanation Generation Framework (2025.emnlp-main)

Copied to clipboard

Shuzhou Yuan, Jingyi Sun, Ran Zhang, Michael Färber, Steffen Eger, Pepa Atanasova, Isabelle Augenstein

DE-CLIP: Few-Shot Anomaly Detection via Difference-Guided Embedding Editing (2026.acl-long)

Challenge:	Existing work has questioned their faithfulness, as they may not accurately reflect the model’s internal reasoning process regarding its predicted answer.
Approach:	They propose a Graph-Guided Textual Explanation Generation framework that generates a graph neural network layer that guides the NLE generation and generates explanations with greater semantic and lexical similarity to human-written ones.
Outcome:	The proposed framework improves NLE faithfulness by up to 12.12% compared to baseline methods on encoder-decoder and decoder-only models.

Copied to clipboard

Yage Zhang, Yukun Jiang, Michael Backes, Yang Zhang

InferPilot: Autonomous Inference Attacks Against ML Services With LLM-Based Agents (2026.findings-acl)

Challenge:	Existing approaches to detect anomalies are limited due to the lack of anomalous samples .
Approach:	They propose a framework that edits text embeddings based on the differences between normal and anomalous samples.
Outcome:	The proposed framework achieves 96.6% and 96.99% AUROC on MVTec datasets.

Copied to clipboard

Yixin Wu, Rui Wen, Chi Cui, Michael Backes, Yang Zhang

A Generalizable Rhetorical Strategy Annotation Model Using LLM-based Debate Simulation and Labelling (2025.findings-emnlp)

Challenge:	Inference attacks are important for assessing model's robustness, but their implementation and parameters are challenging for non-experts.
Approach:	They propose an autonomous agent capable of conducting inference attacks without human intervention.
Outcome:	The proposed agent achieves a 100.0% task completion rate and near-expert attack performance with an average token cost of only 0.627 per run.

Copied to clipboard

Shiyu Ji, Farnoosh Hashemi, Joice Chen, Juanwen Pan, Weicheng Ma, Hefan Zhang, Sophia Pan, Ming Cheng, Shubham Mohole, Saeed Hassanpour, Soroush Vosoughi, Michael Macy

Challenge:	Rhetorical strategies are important to persuasive communication, but their analysis relies on human annotation, which is costly, inconsistent and difficult to scale.
Approach:	They propose a framework that leverages large language models to generate and label debate data . they fine-tune transformer-based classifiers on this dataset and validate it against human data a .
Outcome:	The proposed model achieves high performance and strong generalization across topical domains.

UniSumm and SummZoo: Unified Model and Diverse Benchmark for Few-Shot Summarization (2023.acl-long)

Copied to clipboard

Yulong Chen, Yang Liu, Ruochen Xu, Ziyi Yang, Chenguang Zhu, Michael Zeng, Yue Zhang

Enhancing LLM-Based Persuasion Simulations with Cultural and Speaker-Specific Information (2025.findings-emnlp)

Challenge:	a new benchmark summarization model is being developed to train few-shot summarizers . a large number of summarizing tasks are required to perform well in heterogeneous datasets.
Approach:	They propose a few-shot summarization model pre-trained with multiple summarizing tasks . they propose 'uniSumm' to be prefix-tuned to excel at any few-shot summarisation task .
Outcome:	The proposed model outperforms baseline models under automatic and human evaluations and achieves comparable results in human evaluation.

Copied to clipboard

Weicheng Ma, Hefan Zhang, Shiyu Ji, Farnoosh Hashemi, Qichao Wang, Ivory Yang, Joice Chen, Juanwen Pan, Michael Macy, Saeed Hassanpour, Soroush Vosoughi

Dialogue Systems & Conversational Agents

Challenge:	Existing approaches to persuasive dialogue generation suffer from stance oscillation and low informativeness.
Approach:	They propose reinforced instructional prompting, a method that ensures speaker characteristics consistently guide all stages of dialogue generation.
Outcome:	The proposed method ensures speaker characteristics guide all stages of dialogue generation and aligns language use with speakers’ native languages to better capture cultural nuances.

WorldCuisines: A Massive-Scale Benchmark for Multilingual and Multicultural Visual Question Answering on Global Cuisines (2025.naacl-long)

Copied to clipboard

Genta Indra Winata, Frederikus Hudi, Patrick Amadeus Irawan, David Anugraha, Rifki Afina Putri, Wang Yutong, Adam Nohejl, Ubaidillah Ariq Prathama, Nedjma Ousidhoum, Afifa Amriani, Anar Rzayev, Anirban Das, Ashmari Pramodya, Aulia Adila, Bryan Wilie, Candy Olivia Mawalim, Cheng Ching Lam, Daud Abolade, Emmanuele Chersoni, Enrico Santus, Fariz Ikhwantri, Garry Kuwanto, Hanyang Zhao, Haryo Akbarianto Wibowo, Holy Lovenia, Jan Christian Blaise Cruz, Jan Wira Gotama Putra, Junho Myung, Lucky Susanto, Maria Angelica Riera Machin, Marina Zhukova, Michael Anugraha, Muhammad Farid Adilazuarda, Natasha Christabelle Santosa, Peerat Limkonchotiwat, Raj Dabre, Rio Alexander Audino, Samuel Cahyawijaya, Shi-Xiong Zhang, Stephanie Yulia Salim, Yi Zhou, Yinxuan Gui, David Ifeoluwa Adelani, En-Shiun Annie Lee, Shogo Okada, Ayu Purwarianti, Alham Fikri Aji, Taro Watanabe, Derry Tanti Wijaya, Alice Oh, Chong-Wah Ngo

Visual Data in NLP

Challenge:	Vision Language Models struggle with cultural-specific knowledge, especially in languages other than English and in underrepresented cultural contexts.
Approach:	They propose a visual question answering (VQA) dataset with text-image pairs across 30 languages and dialects and a training dataset.
Outcome:	The proposed model performs better with correct location context, but struggles with adversarial contexts and predicting specific regional cuisines and languages.

One Document, Many Revisions: A Dataset for Classification and Description of Edit Intents (2022.lrec-1)

Copied to clipboard

Dheeraj Rajagopal, Xuchao Zhang, Michael Gamon, Sujay Kumar Jauhar, Diyi Yang, Eduard Hovy

ModSCAN: Measuring Stereotypical Bias in Large Vision-Language Models from Vision and Language Modalities (2024.emnlp-main)

Challenge:	Existing methods to understand revisions have failed to provide a deeper understanding of the nature of these edits.
Approach:	They propose to use a Wikipedia revision history dataset to train a classifier that achieves a 90% accuracy in identifying edit intent and a distantly-supervised model that generates .
Outcome:	The proposed model achieves 90% accuracy in identifying edit intent and a best score of 28 ROUGE.

Copied to clipboard

Yukun Jiang, Zheng Li, Xinyue Shen, Yugeng Liu, Michael Backes, Yang Zhang

ConvLab-3: A Flexible Dialogue System Toolkit Based on a Unified Data Format (2023.emnlp-demo)

Challenge:	Large vision-language models have been widely used but stereotypical biases are unexplored.
Approach:	They propose a framework to SCAN stereotypical bias within large vision-language models . they examine stereotype biases with respect to gender and race in three scenarios .
Outcome:	The proposed framework can reduce stereotypical biases in large vision-language models . the currently popular models show significant stereotype biase .

Copied to clipboard

Qi Zhu, Christian Geishauser, Hsien-chin Lin, Carel van Niekerk, Baolin Peng, Zheng Zhang, Shutong Feng, Michael Heck, Nurul Lubis, Dazhen Wan, Xiaochen Zhu, Jianfeng Gao, Milica Gasic, Minlie Huang

MACSum: Controllable Summarization with Mixed Attributes (2023.tacl-1)

Challenge:	Existing tools for building TOD systems often lack a user-friendly interface . a toolkit with advanced, easily integrable modules is needed to bridge this gap .
Approach:	They propose a multifaceted dialogue system toolkit that integrates diverse datasets and models with a streamlined training process and in-depth evaluation tools.
Outcome:	The proposed toolkit combines RL and transfer learning to support the rapid development and evaluation of robust dialogue policies.

Copied to clipboard

Yusen Zhang, Yang Liu, Ziyi Yang, Yuwei Fang, Yulong Chen, Dragomir Radev, Chenguang Zhu, Michael Zeng, Rui Zhang

Information Extraction & Text Mining

Challenge:	Existing work on controllable summarization with mixed attributes lacks designated annotations.
Approach:	They propose a human-annotated summarization benchmark for controllable summarizing with mixed attributes based on news and dialogue sources .
Outcome:	The proposed dataset contains human-annotated summarization datasets with mixed attributes . hard prompt models yield the best performance on most metrics and human evaluations . mixed-attribute control is still challenging for summarizing tasks .

GEMv2: Multilingual NLG Benchmarking in a Single Line of Code (2022.emnlp-demos)

Copied to clipboard

Sebastian Gehrmann, Abhik Bhattacharjee, Abinaya Mahendiran, Alex Wang, Alexandros Papangelis, Aman Madaan, Angelina Mcmillan-major, Anna Shvets, Ashish Upadhyay, Bernd Bohnet, Bingsheng Yao, Bryan Wilie, Chandra Bhagavatula, Chaobin You, Craig Thomson, Cristina Garbacea, Dakuo Wang, Daniel Deutsch, Deyi Xiong, Di Jin, Dimitra Gkatzia, Dragomir Radev, Elizabeth Clark, Esin Durmus, Faisal Ladhak, Filip Ginter, Genta Indra Winata, Hendrik Strobelt, Hiroaki Hayashi, Jekaterina Novikova, Jenna Kanerva, Jenny Chim, Jiawei Zhou, Jordan Clive, Joshua Maynez, João Sedoc, Juraj Juraska, Kaustubh Dhole, Khyathi Raghavi Chandu, Laura Perez Beltrachini, Leonardo F . R. Ribeiro, Lewis Tunstall, Li Zhang, Mahim Pushkarna, Mathias Creutz, Michael White, Mihir Sanjay Kale, Moussa Kamal Eddine, Nico Daheim, Nishant Subramani, Ondrej Dusek, Paul Pu Liang, Pawan Sasanka Ammanamanchi, Qi Zhu, Ratish Puduppully, Reno Kriz, Rifat Shahriyar, Ronald Cardenas, Saad Mahamood, Salomey Osei, Samuel Cahyawijaya, Sanja Štajner, Sebastien Montella, Shailza Jolly, Simon Mille, Tahmid Hasan, Tianhao Shen, Tosin Adewumi, Vikas Raunak, Vipul Raheja, Vitaly Nikolaev, Vivian Tsai, Yacine Jernite, Ying Xu, Yisi Sang, Yixin Liu, Yufang Hou

Text Generation

Challenge:	Evaluations in machine learning rarely use the latest metrics, datasets, or human evaluation in favor of remaining compatible with prior work.
Approach:	They propose to use the Generation, Evaluation, and Metrics Benchmark to integrate new evaluation methods into existing evaluations.
Outcome:	The proposed evaluation infrastructure bridges the gap between the advantages of leaderboards and in-depth and evolving evaluations by allowing model developers to benefit from each other's work.

Crowdsource, Crawl, or Generate? Creating SEA-VL, a Multicultural Vision-Language Dataset for Southeast Asia (2025.acl-long)

Copied to clipboard

Samuel Cahyawijaya, Holy Lovenia, Joel Ruben Antony Moniz, Tack Hwa Wong, Mohammad Rifqi Farhansyah, Thant Thiri Maung, Frederikus Hudi, David Anugraha, Muhammad Ravi Shulthan Habibi, Muhammad Reza Qorib, Amit Agarwal, Joseph Marvin Imperial, Hitesh Laxmichand Patel, Vicky Feliren, Bahrul Ilmi Nasution, Manuel Antonio Rufino, Genta Indra Winata, Rian Adam Rajagede, Carlos Rafael Catalan, Mohamed Fazli Mohamed Imam, Priyaranjan Pattnayak, Salsabila Zahirah Pranida, Kevin Pratama, Yeshil Bangera, Adisai Na-Thalang, Patricia Nicole Monderin, Yueqi Song, Christian Simon, Lynnette Hui Xian Ng, Richardy Lobo Sapan, Taki Hasan Rafi, Bin Wang, null Supryadi, Kanyakorn Veerakanjana, Piyalitt Ittichaiwong, Matthew Theodore Roque, Karissa Vincentio, Takdanai Kreangphet, Phakphum Artkaew, Kadek Hendrawan Palgunadi, Yanzhi Yu, Rochana Prih Hastuti, William Nixon, Mithil Bangera, Adrian Xuan Wei Lim, Aye Hninn Khine, Hanif Muhammad Zhafran, Teddy Ferdinan, Audra Aurora Izzani, Ayushman Singh, Evan Evan, Jauza Akbar Krito, Michael Anugraha, Fenal Ashokbhai Ilasariya, Haochen Li, John Amadeo Daniswara, Filbert Aurelian Tjiaranata, Eryawan Presma Yulianrifat, Can Udomcharoenchaikit, Fadil Risdian Ansori, Mahardika Krisna Ihsani, Giang Nguyen, Anab Maulana Barik, Dan John Velasco, Rifo Ahmad Genadi, Saptarshi Saha, Chengwei Wei, Isaiah Edri W. Flores, Kenneth Chen Ko Han, Anjela Gail D. Santos, Wan Shen Lim, Kaung Si Phyo, Tim Santos, Meisyarah Dwiastuti, Jiayun Luo, Jan Christian Blaise Cruz, Ming Shan Hee, Ikhlasul Akmal Hanif, M.Alif Al Hakim, Muhammad Rizky Sya’ban, Kun Kerdthaisong, Lester James Validad Miranda, Fajri Koto, Tirana Noor Fatyanosa, Alham Fikri Aji, Jostin Jerico Rosal, Jun Kevin, Robert Wijaya, Onno P. Kampman, Ruochen Zhang, Börje F. Karlsson, Peerat Limkonchotiwat

Multimodality

Challenge:	Southeast Asia is underrepresented in vision-language research . SEA-VL is an open-source initiative dedicated to developing culturally relevant datasets for SEA languages.
Approach:	They propose to use crowdsourced, automated image crawling and synthetic image generation to develop culturally relevant datasets for SEA languages.
Outcome:	The proposed datasets capture SEA cultural nuances and contexts better than existing datasets.

SituatedQA: Incorporating Extra-Linguistic Contexts into QA (2021.emnlp-main)

Copied to clipboard

Michael Zhang, Eunsol Choi

Question Answering

Challenge:	Existing models struggle with producing answers that are frequently updated or from uncommon locations.
Approach:	They propose an open-retrieval QA dataset where systems must produce the correct answer given the context.
Outcome:	The proposed dataset shows that existing models struggle with producing answers that are frequently updated or from uncommon locations.

Modeling the Relationship between User Comments and Edits in Document Revision (D19-1)

Copied to clipboard

Xuchao Zhang, Dheeraj Rajagopal, Michael Gamon, Sujay Kumar Jauhar, ChangTien Lu

Challenge:	Managing collaborative documents can be difficult due to the profusion of edits and comments that multiple authors make during a document’s evolution.
Approach:	They propose a hierarchical multi-layer deep neural network to model the relationship between edits and comments by encoding specific edit actions such as additions and deletions while accounting for document context.
Outcome:	The proposed model outperforms baselines in a number of evaluation settings and achieves a precision@1 of 71.0% and precision@3 of 94.4% for Comment Ranking while achieving 74.4% accuracy on Edit Anchoring.

The Art of (Mis)alignment: How Fine-Tuning Methods Effectively Misalign and Realign LLMs in Post-Training (2026.findings-acl)

Copied to clipboard

Rui Zhang, Hongwei Li, Yun Shen, Xinyue Shen, Wenbo Jiang, Guowen Xu, Yang Liu, Michael Backes, Yang Zhang

Peering Behind the Shield: Guardrail Identification in Large Language Models (2026.findings-acl)

Challenge:	Misaligned large language models can magnify harm by exploiting them to undermine safety . et al., 2022b; Bai e.t., 2023): misalignment, realignment and model-specific resistance are important .
Approach:	They evaluate four methods to identify a mechanism asymmetry between attack and defense . they find that ORPO is most effective for misalignment, but DPO excels in realignment .
Outcome:	The proposed methods show a mechanism asymmetry between attack and defense . the proposed methods excel in realignment, but at the expense of model utility .

Copied to clipboard

Ziqing Yang, Yixin Wu, Rui Wen, Michael Backes, Yang Zhang

PRewrite: Prompt Rewriting with Reinforcement Learning (2024.acl-short)

Challenge:	Identifying guardrails in conversational AI agents is critical for identifying malicious content . identifying guardrail components in black-box AI agents poses security challenges .
Approach:	They propose a method that leverages guard-specific adversarial prompts to detect guardrails in black-box AI agents.
Outcome:	The proposed method achieves perfect classification accuracy in multiple scenarios.

Copied to clipboard

Weize Kong, Spurthi Hombaiah, Mingyang Zhang, Qiaozhu Mei, Michael Bendersky

Text Generation

Challenge:	Prompt engineering is done manually in a trial-and-error ad-hoc fashion, authors say .
Approach:	They propose a method to rewrite an under-optimized prompt to a more effective prompt.
Outcome:	The proposed method rewrites an under-optimized prompt to a more effective prompt.

PLaD: Preference-based Large Language Model Distillation with Pseudo-Preference Pairs (2024.findings-acl)

Copied to clipboard

Rongzhi Zhang, Jiaming Shen, Tianqi Liu, Haorui Wang, Zhen Qin, Feng Han, Jialu Liu, Simon Baumgartner, Michael Bendersky, Chao Zhang

Accelerating Code Search with Deep Hashing and Code Classification (2022.acl-long)

Challenge:	Knowledge distillation (KD) is a technique for transferring expertise from large teacher models to compact student models with reduced memory footprints and inference costs.
Approach:	They propose to transfer knowledge from large teacher models to compact student models by exploiting teacher-student capacity discrepancies to generate pseudo-preference pairs where teacher outputs are preferred over student outputs.
Outcome:	The proposed framework exploits teacher-student capacity discrepancy to generate pseudo-preference pairs where teacher outputs are preferred over student outputs.

Copied to clipboard

Wenchao Gu, Yanlin Wang, Lun Du, Hongyu Zhang, Shi Han, Dongmei Zhang, Michael Lyu

MCSE: Multimodal Contrastive Learning of Sentence Embeddings (2022.naacl-main)

Challenge:	Code search is to search reusable code snippets from source code corpus based on natural languages queries.
Approach:	They propose a method to accelerate code search with deep hashing and code classification by using deep hashes and code hash.
Outcome:	The proposed method can save 90% of retrieval time while preserving at least 99% of retrievals accuracy.

Copied to clipboard

Miaoran Zhang, Marius Mosbach, David Adelani, Michael Hedderich, Dietrich Klakow

Creator Context for Tweet Recommendation (2023.emnlp-industry)

Challenge:	Existing approaches to learning semantically meaningful sentence embeddings are limited by the complexity of pre-trained models.
Approach:	They propose a sentence embedding learning approach that exploits both visual and textual information via a multimodal contrastive objective.
Outcome:	The proposed approach improves the state-of-the-art average Spearman’s correlation by 1.7% on a variety of semantic textual similarity tasks.

Copied to clipboard

Spurthi Amba Hombaiah, Tao Chen, Mingyang Zhang, Michael Bendersky, Marc Najork, Matt Colen, Sergey Levi, Vladimir Ofitserov, Tanvir Amin

Challenge:	In tweets, people refer to the content it delivers, but also to the person behind it.
Approach:	They examine how creator context can be used to advance tweet understanding by recommending relevant tweets to news articles.
Outcome:	The proposed model can improve a news article's relevance by recommending relevant tweets to news articles.

Adapting LLM Agents with Universal Communication Feedback (2025.findings-naacl)

Copied to clipboard

Kuan Wang, Yadong Lu, Michael Santacroce, Yeyun Gong, Chao Zhang, Yelong Shen

Challenge:	Recent advances in large language models (LLMs) have demonstrated potential for LLM agents.
Approach:	They propose a universal buffer and iterative pipeline to store feedback and itersative pipelines to enable LLM agents to explore and update their policy in an environment.
Outcome:	The proposed approach outperforms supervised instruction fine-tuning baselines on four datasets.

Selectively Answering Ambiguous Questions (2023.emnlp-main)

Copied to clipboard

Jeremy Cole, Michael Zhang, Daniel Gillick, Julian Eisenschlos, Bhuwan Dhingra, Jacob Eisenstein

Question Answering

Challenge:	Prior work has focused on the case in which the question is clear and the answer is unambiguous but possibly unknown.
Approach:	They propose to use a sampled set of questions to calibrate answers to ambiguous questions with varying model scales.
Outcome:	The results show that sampling-based confidence scores help calibrate answers to relatively unambiguous questions, with more dramatic improvements on ambiguous ones.

Pruning Unsafe Tickets: A Resource-Efficient Framework for Safer and More Robust LLMs (2026.acl-long)

Copied to clipboard

Wai Man Si, Mingjie Li, Michael Backes, Yang Zhang

Open Schrödinger’s Closed Box: Identifying Retrieval Augmented Generation in API-Accessible Large Language Model Services (2026.acl-long)

Challenge:	Empirical evaluations on ML models show substantial reductions in unsafe generations and improved robustness against jailbreak attacks.
Approach:	They propose a resource-efficient pruning framework that directly identifies unsafe behaviors while preserving model utility.
Outcome:	The proposed framework reduces unsafe generations and improves robustness against jailbreak attacks with minimal utility loss.

Copied to clipboard

Yukun Jiang, Xinyue Shen, Michael Backes, Zheng Li, Yang Zhang

Rich Knowledge Sources Bring Complex Knowledge Conflicts: Recalibrating Models to Reflect Conflicting Evidence (2022.emnlp-main)

Challenge:	Large language models (LLMs) are powerful at question-answering but prone to hallucinations due to limited domain-specific or up-to-date knowledge.
Approach:	They propose a framework for IDentifying RAG properties in LLM services that integrates LLMs with retrieval systems and adds an external retriever and knowledge database to mitigate hallucinations.
Outcome:	The proposed framework detects RAG-enhanced LLMs with 99.97% accuracy with partial or no optional knowledge and nearly 100% when the LLM and database are known.

Copied to clipboard

Hung-Ting Chen, Michael Zhang, Eunsol Choi

AlphaEdit+: Model Editing in the Presence of Conflicting and Inconsistent Knowledge (2026.findings-acl)

Challenge:	Existing work on question answering models relies on retrieved documents for provenance, but recent studies show that models can retain vast amounts of factual knowledge . retrieval-based generation approaches combine parametric knowledge sources with a large number of retrieved evidence documents, achieving state-of-the-art performance on open retrieval datasets.
Approach:	They propose to use parametric and parametric knowledge to generate free-form questions from retrieved evidence documents.
Outcome:	The proposed model can use parametric and parametric knowledge to generate free-form answers from retrieved evidence documents.

Copied to clipboard

Qing Liu, Jianhao Zhang, Ou Wu, Michael Ng, Yi Du

Challenge:	Existing methods for knowledge editing struggle with knowledge conflicts and inconsistencies.
Approach:	They propose a new method for knowledge editing that relaxes null-space constraints and introduces a weighting scheme to mitigate conflicts between new and historical knowledge.
Outcome:	The proposed method outperforms existing methods on challenging datasets and outperformed existing methods.

On Second Thought, Let’s Not Think Step by Step! Bias and Toxicity in Zero-Shot Reasoning (2023.acl-long)

Copied to clipboard

Omar Shaikh, Hongxin Zhang, William Held, Michael Bernstein, Diyi Yang

Retrieval Augmented Generation or Long-Context LLMs? A Comprehensive Study and Hybrid Approach (2024.emnlp-industry)

Challenge:	Prior work has focused on logical reasoning tasks; it remains unclear whether improvements hold for more diverse types of reasoning, especially in socially situated contexts.
Approach:	They perform a controlled evaluation of zero-shot CoT reasoning in two socially sensitive domains: harmful questions and stereotype benchmarks.
Outcome:	The results show that zero-shot CoT reasoning increases model’s likelihood to produce harmful or undesirable output, but decreases with improved instruction following.

Copied to clipboard

Zhuowan Li, Cheng Li, Mingyang Zhang, Qiaozhu Mei, Michael Bendersky

Analyzing the Role of Semantic Representations in the Era of Large Language Models (2024.naacl-long)

Challenge:	Recent LLMs like Gemini-1.5 and GPT-4 show exceptional capabilities to understand long contexts directly.
Approach:	They propose a method that routes queries to RAG or LC based on model self-reflection.
Outcome:	The proposed method significantly reduces the computation cost while maintaining a comparable performance to RAG.

Copied to clipboard

Zhijing Jin, Yuen Chen, Fernando Gonzalez Adauto, Jiarui Liu, Jiayi Zhang, Julian Michael, Bernhard Schölkopf, Mona Diab

Jailbreaking Attacks vs. Content Safety Filters: How Far Are We in the LLM Safety Arms Race? (2026.findings-acl)

Challenge:	Existing studies show the benefits of semantic representations in NLP tasks . Existing work using AMR is concerned with trainable models .
Approach:	They propose an AMR-driven chain-of-thought prompting method that uses AMR . they propose to use it to predict which input examples AMR may help or hurt on .
Outcome:	The proposed method hurts performance more than it helps on five different tasks.

Copied to clipboard

Yuan Xin, Dingfan Chen, Linyi Yang, Michael Backes, Xiao Zhang

AdaPrompt: Adaptive Model Training for Prompt-based NLP (2022.findings-emnlp)

Challenge:	Existing studies have focused on the models, neglecting the full deployment pipeline . previous studies have underestimated the practical success of these attacks .
Approach:	They evaluate the effectiveness of jailbreak attacks targeting LLM safety alignment . they highlight critical gaps and call for further refinement of detection accuracy and usability .
Outcome:	The proposed attacks can detect at least one safety filter across the entire deployment pipeline.

Copied to clipboard

Yulong Chen, Yang Liu, Li Dong, Shuohang Wang, Chenguang Zhu, Michael Zeng, Yue Zhang

Rethinking Assessments of Prompt Injection Attacks (2026.findings-acl)

Challenge:	Prompt-based learning can tackle zero-shot and few-shot NLP tasks . authors propose a method that makes use of pre-trained language models .
Approach:	They propose to map NLP tasks into natural language prompts, which are then filled by pre-trained language models.
Outcome:	The proposed method outperforms standard prompt-based methods in few-shot settings.

Copied to clipboard

Chi Cui, Yixin Wu, Michael Backes, Yang Zhang

Discovering Language Model Behaviors with Model-Written Evaluations (2023.findings-acl)

Challenge:	Prompt injection attacks are recognized as one of the primary risks faced by LLM-integrated applications in recent years.
Approach:	They evaluate prompt injection attacks on LLM-integrated applications across 37 target tasks, 185 injected tasks, 21 attack instructions, and 143,745 queries.
Outcome:	The proposed framework provides a solid foundation for assessing vulnerabilities in LLM-integrated applications and evaluating the efficacy of defensive strategies.

Copied to clipboard

Ethan Perez, Sam Ringer, Kamile Lukosiute, Karina Nguyen, Edwin Chen, Scott Heiner, Craig Pettit, Catherine Olsson, Sandipan Kundu, Saurav Kadavath, Andy Jones, Anna Chen, Benjamin Mann, Brian Israel, Bryan Seethor, Cameron McKinnon, Christopher Olah, Da Yan, Daniela Amodei, Dario Amodei, Dawn Drain, Dustin Li, Eli Tran-Johnson, Guro Khundadze, Jackson Kernion, James Landis, Jamie Kerr, Jared Mueller, Jeeyoon Hyun, Joshua Landau, Kamal Ndousse, Landon Goldberg, Liane Lovitt, Martin Lucas, Michael Sellitto, Miranda Zhang, Neerav Kingsland, Nelson Elhage, Nicholas Joseph, Noemi Mercado, Nova DasSarma, Oliver Rausch, Robin Larson, Sam McCandlish, Scott Johnston, Shauna Kravec, Sheer El Showk, Tamera Lanham, Timothy Telleen-Lawton, Tom Brown, Tom Henighan, Tristan Hume, Yuntao Bai, Zac Hatfield-Dodds, Jack Clark, Samuel R. Bowman, Amanda Askell, Roger Grosse, Danny Hernandez, Deep Ganguli, Evan Hubinger, Nicholas Schiefer, Jared Kaplan

Communication Makes Perfect: Persuasion Dataset Construction via Multi-LLM Communication (2025.naacl-long)

Challenge:	Prior work creates evaluations with crowdwork or existing data sources, which are not always available.
Approach:	They generate evaluations automatically with language models (LMs) using crowdwork or existing data sources to find out how they behave .
Outcome:	The results show that large LMs repeat back a dialog user’s preferred answer and express greater desire to pursue concerning goals like resource acquisition and goal preservation.

Copied to clipboard

Weicheng Ma, Hefan Zhang, Ivory Yang, Shiyu Ji, Joice Chen, Farnoosh Hashemi, Shubham Mohole, Ethan Gearey, Michael Macy, Saeed Hassanpour, Soroush Vosoughi

Challenge:	Large Language Models (LLMs) have shown proficiency in generating persuasive dialogue, yet concerns about the fluency and sophistication of their outputs persist.
Approach:	They propose a multi-LLM communication framework that facilitates the efficient production of high-quality, diverse linguistic content with minimal human oversight.
Outcome:	The proposed framework excels in naturalness, linguistic diversity, and the strategic use of persuasion, even in complex scenarios involving social taboos.

Unsupervised Multi-Granularity Summarization (2022.findings-emnlp)

Copied to clipboard

Ming Zhong, Yang Liu, Suyu Ge, Yuning Mao, Yizhu Jiao, Xingxing Zhang, Yichong Xu, Chenguang Zhu, Michael Zeng, Jiawei Han

Information Extraction & Text Mining

Challenge:	Experimental results confirm the substantial superiority of GranuSum on multi-granularity summarization over strong baselines.
Approach:	They propose to rank events by their salience and annotate a benchmark for GranuSum that contains multiple summaries at different granularities for each document cluster.
Outcome:	The proposed framework is capable of producing multi-granular summaries in unsupervised manner over strong baselines.

Entity Cloze By Date: What LMs Know About Unseen Entities (2022.findings-naacl)

Copied to clipboard

Yasumasa Onoe, Michael Zhang, Eunsol Choi, Greg Durrett

Explanation-aware Soft Ensemble Empowers Large Language Model In-context Learning (2024.acl-long)

Challenge:	Existing literature provides benchmarks to measure LMs' knowledge about entities .
Approach:	They propose a framework to analyze what language models can infer about new entities that did not exist when they were pretrained.
Outcome:	The proposed framework shows that models more informed about the entities achieve lower perplexity on this benchmark.

Copied to clipboard

Yue Yu, Jiaming Shen, Tianqi Liu, Zhen Qin, Jing Nathan Yan, Jialu Liu, Chao Zhang, Michael Bendersky

Task Compass: Scaling Multi-task Pre-training with Task Prefix (2022.findings-emnlp)

Challenge:	Recent advances in natural language processing (NLP) have witnessed the remarkable capabilities of Large Language Models (LLMs).
Approach:	They propose an Explanation-Aware Soft Ensemble framework to empower in-context learning with Large language models.
Outcome:	The proposed framework can be used to enhance in-context learning on seven natural language understanding tasks and four varying-size LLMs.

Copied to clipboard

Zhuosheng Zhang, Shuohang Wang, Yichong Xu, Yuwei Fang, Wenhao Yu, Yang Liu, Hai Zhao, Chenguang Zhu, Michael Zeng

Can LMs Learn New Entities from Descriptions? Challenges in Propagating Injected Knowledge (2023.acl-long)

Challenge:	Existing studies show that multi-task learning with large-scale supervised tasks suffers from negative effects across tasks.
Approach:	They propose a task prefix guided multi-task pre-training framework to explore the relationships among tasks.
Outcome:	The proposed model can be used as a foundation backbone for a wide range of tasks and as augmentation tool for data augmentation with complementary tasks.

Copied to clipboard

Yasumasa Onoe, Michael Zhang, Shankar Padmanabhan, Greg Durrett, Eunsol Choi

Bridging the Preference Gap between Retrievers and LLMs (2024.acl-long)

Challenge:	Existing methods for updating knowledge show little propagation of injected knowledge.
Approach:	They propose to inject individual facts into LMs and evaluate whether they can propagate injected facts while not changing predictions on other contexts.
Outcome:	The proposed model can make inferences based on injected facts and propagate them . existing methods show little propagation of injected knowledge .

Copied to clipboard

Zixuan Ke, Weize Kong, Cheng Li, Mingyang Zhang, Qiaozhu Mei, Michael Bendersky

Reward Yourself: Efficient Self Rewards for Trustworthy Sampling (2026.findings-acl)

Challenge:	Existing studies on retrievers and LLMs treat them as separate components . a novel bridge model is proposed to optimize the relationship between the retriever and the LLM .
Approach:	They propose a framework that chains together supervised and reinforcement learning to train a bridge model that optimizes the connection between the retriever and the LLM.
Outcome:	Empirical results show that the proposed model optimizes the connection between the retriever and the LLM.

Copied to clipboard

Mingjie Li, Wai Man Si, Michael Backes, Yang Zhang

Causally Modeling the Linguistic and Social Factors that Predict Email Response (2025.naacl-long)

Challenge:	Retraining reward models to address privacy leaks and stereotypes is expensive . recent advances in large language models have led to improvements in understanding .
Approach:	They propose a lightweight intrinsic reward that can be used to prune existing LLMs to approximate an "untrust" and an ""untrust "" token distribution.
Outcome:	Experiments with two reward models and four LLMs show that selfRW improves trustworthiness with minimal impact on general utility benchmarks.

Copied to clipboard

Yinuo Xu, Hong Chen, Sushrita Rakshit, Aparna Ananthasubramaniam, Omkar Yadav, Mingqian Zheng, Michael Jiang, Lechen Zhang, Bowen Yi, Kenan Alkiek, Abraham Israeli, Bangzhao Shu, Hua Shen, Jiaxin Pei, Haotian Zhang, Miriam Schirmer, David Jurgens

Challenge:	a key intent behind many emails is to get a reply from the recipient.
Approach:	They propose to model the intents, expectations, and responsiveness in email exchanges by using a dataset containing 1800 emails annotated with nuanced types of intents and expectations.
Outcome:	The proposed model is based on 1800 emails annotated with nuanced types of intents and expectations . it shows that social status, argumentation, and strength of social connection influence email response rates .

INDUS: Effective and Efficient Language Models for Scientific Applications (2024.emnlp-industry)

Copied to clipboard

Bishwaranjan Bhattacharjee, Aashka Trivedi, Masayasu Muraoka, Muthukumaran Ramasubramanian, Takuma Udagawa, Iksha Gurung, Nishan Pantha, Rong Zhang, Bharath Dandala, Rahul Ramachandran, Manil Maskey, Kaylin Bugbee, Michael Little, Elizabeth Fancher, Irina Gerasimov, Armin Mehrabian, Lauren Sanders, Sylvain Costes, Sergi Blanco-Cuaresma, Kelly Lockhart, Thomas Allen, Felix Grezes, Megan Ansdell, Alberto Accomazzi, Yousef El-Kurdi, Davis Wertheimer, Birgit Pfitzmann, Cesar Berrospi Ramis, Michele Dolfi, Rafael Lima, Panagiotis Vagenas, S. Mukkavilli, Peter Staar, Sanaz Vahidinia, Ryan McGranaghan, Tsengdar Lee

When GPT Spills the Tea: Comprehensive Assessment of Knowledge File Leakage in GPTs (2025.acl-long)

Challenge:	Large language models trained on general domain corpora showed remarkable results on natural language processing tasks.
Approach:	They develop a suite of large language models trained on general domain corpora that address NLP tasks and smaller versions of them created using knowledge distillation.
Outcome:	The proposed models outperform general-purpose and domain-specific encoders on new and existing tasks and in industrial settings.

Copied to clipboard

Xinyue Shen, Yun Shen, Michael Backes, Yang Zhang

Reconstruct Your Previous Conversations! Comprehensively Investigating Privacy Leakage Risks in Conversations with GPT Models (2024.emnlp-main)

Challenge:	Existing studies show that adversarial prompts can induce GPTs to leak knowledge file content.
Approach:	They propose a workflow inspired by Data Security Posture Management to identify five leakage vectors for knowledge file leakage using 651,022 GPT metadata and 11,820 flows.
Outcome:	The proposed workflow analyzes 651,022 GPT metadata, 11,820 flows, and 1,466 responses to identify five leakage vectors: metadata, GPT initialization, retrieval, sandboxed execution environments, and prompts.

Copied to clipboard

Junjie Chu, Zeyang Sha, Michael Backes, Yang Zhang

PeerCheck: Enhancing LLM-Generated Academic Reviews Towards Human-Level Quality (2026.findings-acl)

Challenge:	Existing GPT models allow users to interact with them for multiple rounds to optimize the task execution.
Approach:	They propose a conversation reconstruction attack targeting the contents of previous conversations between GPT models and benign users, i.e., the benign users’ input contents during their interaction with GPT.
Outcome:	The proposed attacks demonstrate that GPT-4's defense mechanisms are ineffective against these attacks.

Copied to clipboard

Zeyuan Chen, Ziqing Yang, Yihan Ma, Michael Backes, Yang Zhang

Challenge:	Increasing use of large language models (LLMs) in academic review has raised concerns about quality and fairness.
Approach:	They propose a framework to improve the quality of LLM-generated reviews by using retrieval-augmented generation.
Outcome:	The proposed framework improves the human-level quality of LLM-generated reviews by adopting prompt engineering and retrieval-augmented generation.

Proofread: Fixes All Errors with One Tap (2024.acl-demos)

Copied to clipboard

Renjie Liu, Yanxiang Zhang, Yun Zhu, Haicheng Sun, Yuanbo Zhang, Michael Huang, Shanqing Cai, Lei Meng, Shumin Zhai