Large Language Model (LLM) alignment is intended to ensure that models remain helpful and safe, but its stability under input distributional shift is not yet fully understood. Prior work shows that aligned models can fail under jailbreak prompts, alternative encodings, and cross-lingual transfer, yet these failures are...
Mo-Han Li, Cheng-Yu Yu, Francesco Sovrano et al.· arXiv.org· 0 citations
The result is a deployed application and study workflow that integrates activity review, study data collection, privacy control, local prediction, and a participant-facing interface for XAI evaluation.
Cheng-Yu Yu, Leonor Costa, Zoja Anžur et al.· 0 citations
Synthetic semantic-preserving transformations that are rule-based and invertible are used as a probe of alignment generalization and suggest that semantic-preserving distribution shifts can expose a recurring gap in how utility and alignment generalize in current LLMs.
Mo-Han Li, Cheng-Yu Yu, Francesco Sovrano et al.· 0 citations
DreamGuard is proposed, a proactive guardrail for LLM agents built around a risk-aware world model that maintains a compact recurrent latent state over the trajectory and predicts future latent states from which DreamGuard derives immediate-hazard and prefix-risk evidence.
Wenhao Lin, Cheng-Yu Yu, Xingwei Lin et al.· 3 citations
We use cookies to run the site and, with your consent, for analytics and to show ads.
See our Cookie Policy.