4 papers
FORGE: Verification-Gated Behavioral Repair for Generative Language Models
Hsin-Ling Hsu, Min-Yu Chen, Nai-Chia Chen +3
Generative large language models (LLMs) inherit undesirable behaviors from pre-training, including demographic bias and toxic generation, that often emerge only after deployment an…
OTTER: A Red-Teaming System for Toxicity-Evading Jailbreak Prompt Optimization
Jerry Wang, Hsin-Ling Hsu, Yi-Cheng Lai +2
Production LLMs increasingly rely on toxicity-based moderation filters as a primary defense, assuming that harmful intent correlates with toxic surface wording. We show this assump…
MedAction: Towards Active Multi-turn Clinical Diagnostic LLMs
Hsin-Ling Hsu, Zizheng Wang, Donghua Zhang +9
Most existing LLM diagnoses are evaluated on static, single-turn settings where complete patient information is provided upfront, an oversimplification of real clinical practice. W…
WARP: Guaranteed Inner-Layer Repair of NLP Transformers
Hsin-Ling Hsu, Min-Yu Chen, Nai-Chia Chen +3
Transformer-based NLP models remain vulnerable to adversarial perturbations, yet existing repair methods face a fundamental trade-off: gradient-based approaches offer flexibility b…