6 papers
Discovering Universal Activation Directions for PII Leakage in Language Models
Leo Marchyok, Zachary Coalson, Sungho Keum +2
Modern language models exhibit rich internal structure, yet little is known about how privacy-sensitive behaviors, such as personally identifiable information (PII) leakage, are re…
Fail-Closed Alignment for Large Language Models
Zachary Coalson, Beth Sohler, Aiden Gabriel +1
We identify a structural weakness in current large language model (LLM) alignment: modern refusal mechanisms are fail-open. While existing approaches encode refusal behaviors acros…
Hard Work Does Not Always Pay Off: Poisoning Attacks on Neural Architecture Search
Zachary Coalson, Huazheng Wang, Qingyun Wu +1
We study the robustness of data-centric methods to find neural network architectures, known as neural architecture search (NAS), against data poisoning. To audit this robustness, w…
IF-GUIDE: Influence Function-Guided Detoxification of LLMs
Zachary Coalson, Juhan Bae, Nicholas Carlini +1
We study how training data contributes to the emergence of toxic behaviors in large language models. Most prior work on reducing model toxicity adopts reactive approaches, such as…
PrisonBreak: Jailbreaking Large Language Models with at Most Twenty-Five Targeted Bit-flips
Zachary Coalson, Jeonghyun Woo, Chris S. Lin +8
We study a new vulnerability in commercial-scale safety-aligned large language models (LLMs): their refusal to generate harmful responses can be broken by flipping only a few bits…
Harnessing Input-Adaptive Inference for Efficient VLN
Dongwoo Kang, Akhil Perincherry, Zachary Coalson +3
An emerging paradigm in vision-and-language navigation (VLN) is the use of history-aware multi-modal transformer models. Given a language instruction, these models process observat…