Zhouxiang Fang

Hanjie Chen, Zhouxiang Fang, Yash Singla et al.

LLMs have demonstrated impressive performance in answering medical questions, such as achieving passing scores on medical licensing examinations. However, medical board exams or general clinical questions do not capture the complexity of realistic clinical cases. Moreover, the lack of reference explanations means we cannot easily evaluate the reasoning of model decisions, a crucial component of supporting doctors in making complex medical decisions. To address these challenges, we construct two new datasets: JAMA Clinical Challenge and Medbullets.\footnote{Datasets and code are available at \url{https://github.com/HanjieChen/ChallengeClinicalQA}.} JAMA Clinical Challenge consists of questions based on challenging clinical cases, while Medbullets comprises simulated clinical questions. Both datasets are structured as multiple-choice question-answering tasks, accompanied by expert-written explanations. We evaluate seven LLMs on the two datasets using various prompts. Experiments demonstrate that our datasets are harder than previous benchmarks. In-depth automatic and human evaluations of model-generated explanations provide insights into the promise and deficiency of LLMs for explainable medical QA.

Zhouxiang Fang, Min Yu, Zhendong Fu et al.

Posts, as important containers of user-generated-content pieces on social media, are of tremendous social influence and commercial value. As an integral components of a post, the headline has a decisive contribution to the post's popularity. However, current mainstream method for headline generation is still manually writing, which is unstable and requires extensive human effort. This drives us to explore a novel research question: Can we automate the generation of popular headlines on social media? We collect more than 1 million posts of 42,447 celebrities from public data of Xiaohongshu, which is a well-known social media platform in China. We then conduct careful observations on the headlines of these posts. Observation results demonstrate that trends and personal styles are widespread in headlines on social medias and have significant contribution to posts's popularity. Motivated by these insights, we present MEBART, which combines Multiple preference-Extractors with Bidirectional and Auto-Regressive Transformers (BART), capturing trends and personal styles to generate popular headlines on social medias. We perform extensive experiments on real-world datasets and achieve state-of-the-art performance compared with several advanced baselines. In addition, ablation and case studies demonstrate that MEBART advances in capturing trends and personal styles.

Zhouxiang Fang, Jiawei Zhou, Hanjie Chen

Recent studies show that the safety alignment of large language models (LLMs) can be easily compromised even by seemingly non-adversarial fine-tuning. To preserve safety alignment during fine-tuning, a widely used strategy is to jointly optimize safety and task objectives by mixing in the original alignment data, which is typically inaccessible even for open-weight LLMs. Inspired by generative replay in continual learning, we propose Generative Replay for Safety Alignment Preservation (GR-SAP), a unified framework that synthesizes domain-specific alignment data from LLMs and integrate them during downstream adaption to preserve safety alignment. Theoretical and empirical analyses demonstrate this synthetic data serves as a reliable proxy for the original alignment data. Experiments across various models and downstream tasks show that GR-SAP substantially mitigates fine-tuning-induced safety degradation while maintaining comparable downstream performance. Our code is available at https://github.com/chili-lab/gr-sap.

Zhouxiang Fang, Aayush Mishra, Muhan Gao et al.

Recent works have suggested that In-Context Learning (ICL) operates in dual modes, i.e. task retrieval (remember learned patterns from pre-training) and task learning (inference-time ''learning'' from demonstrations). However, disentangling these the two modes remains a challenging goal. We introduce ICL CIPHERS, a class of task reformulations based on substitution ciphers borrowed from classic cryptography. In this approach, a subset of tokens in the in-context inputs are substituted with other (irrelevant) tokens, rendering English sentences less comprehensible to human eye. However, by design, there is a latent, fixed pattern to this substitution, making it reversible. This bijective (reversible) cipher ensures that the task remains a well-defined task in some abstract sense, despite the transformations. It is a curious question if LLMs can solve tasks reformulated by ICL CIPHERS with a BIJECTIVE mapping, which requires ''deciphering'' the latent cipher. We show that LLMs are better at solving tasks reformulated by ICL CIPHERS with BIJECTIVE mappings than the NON-BIJECTIVE (irreversible) baseline, providing a novel approach to quantify ''learning'' in ICL. While this gap is small, it is consistent across the board on four datasets and six models. Finally, we examine LLMs' internal representations and identify evidence in their ability to decode the ciphered inputs.