Papers by Taiwei Shi
STEER-BENCH: A Benchmark for Evaluating the Steerability of Large Language Models (2025.emnlp-main)
Copied to clipboard
| Challenge: | Large language models can adapt outputs to align with community-specific norms, perspectives and communication styles. |
| Approach: | They propose a benchmark to assess community-specific steering using contrasting reddit communities. |
| Outcome: | STEER-BENCH assesses how well large language models understand community-specific instructions, their resilience to adversarial steering attempts, and their ability to accurately represent cultural and ideological perspectives. |
How Susceptible are Large Language Models to Ideological Manipulation? (2024.emnlp-main)
Copied to clipboard
| Challenge: | Large Language Models (LLMs) have the potential to exert substantial influence on public perceptions and interactions with information. |
| Approach: | They examine how LLMs can learn and generalize ideological biases from their instruction-tuning data. |
| Outcome: | The LLMs show a startling ability to absorb ideology from one topic and generalize it to even unrelated ones. |
Safer-Instruct: Aligning Language Models with Automated Preference Data (2024.naacl-long)
Copied to clipboard
| Challenge: | annotating preference data by humans is resource-intensive and creativity-demanding . existing methods face limitations in data diversity and quality . |
| Approach: | They propose a pipeline for annotating large-scale preference data without human annotators. |
| Outcome: | The proposed pipeline outperforms models fine-tuned on human-annotated safety preference data while maintaining a competitive edge in downstream tasks. |
CoAnnotating: Uncertainty-Guided Work Allocation between Human and Large Language Models for Data Annotation (2023.emnlp-main)
Copied to clipboard
| Challenge: | Annotated data plays a critical role in training models and evaluating their performance. |
| Approach: | They propose a paradigm for Human-LLM co-annotation of unstructured texts at scale that utilizes uncertainty to estimate LLMs’ annotation capability. |
| Outcome: | The proposed model outperforms existing models on many text-annotation tasks with up to 21% performance improvement over random baseline. |
Can Language Model Moderators Improve the Health of Online Discourse? (2024.naacl-long)
Copied to clipboard
Hyundong Cho, Shuai Liu, Taiwei Shi, Darpan Jain, Basem Rizk, Yuyang Huang, Zixun Lu, Nuan Wen, Jonathan Gratch, Emilio Ferrara, Jonathan May
| Challenge: | Existing efforts to automate conversational moderation have focused on banning harmful comments or deleting them, but such efforts can inadvertently push users towards echo chambers that exacerbate polarization. |
| Approach: | They propose a framework to assess models’ moderation capabilities independently of human intervention and propose 'conversational moderation' they propose to use language models as conversational moderators to provide specific feedback on toxic behavior but struggle to influence users to increase their levels of respect and cooperation. |
| Outcome: | The proposed framework assesses models’ moderation capabilities independently of human intervention and shows that appropriately prompted models provide specific and fair feedback on toxic behavior but struggle to influence users to increase their levels of respect and cooperation. |
WildFeedback: Aligning LLMs With In-situ User Interactions And Feedback (2026.acl-long)
Copied to clipboard
Taiwei Shi, Zhuoer Wang, Longqi Yang, Ying-Chun Lin, Zexue He, Mengting Wan, Pei Zhou, Sujay Kumar Jauhar, Sihao Chen, Shan Xia, Hongfei Zhang, Jieyu Zhao, Xiaofeng Xu, Xia Song, Jennifer Neville
| Challenge: | Traditional alignment methods rely on human annotations and are subjective and misalignment with real-world user preferences. |
| Approach: | They propose a framework that leverages in-situ user feedback during conversations with LLMs to create preference datasets automatically. |
| Outcome: | The proposed framework identifies and classifies user feedback to LLM responses between conversation turns and creates examples of preferred and dispreferred responses according to user preferences. |
The Hallucination Tax of Reinforcement Finetuning (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Reinforcement finetuning (RFT) has become a standard approach for enhancing the reasoning capabilities of large language models (LLMs). |
| Approach: | They propose to incorporate 10% SUM into RFT to restore appropriate refusal behavior with minimal accuracy trade-offs on solvable tasks. |
| Outcome: | The proposed approach reduces model refusal rates by more than 80%, which significantly increases model’s tendency to hallucinate. |