Debargha Ganguly

Current approaches to LLM safety rely on a brittle pattern of identifying and blocking known threats via guardrails. This paper introduces Trust The Typical (T3), a framework that reframes safety as an out-of-distribution detection problem, learning the distribution of acceptable prompts in a semantic space and flagging significant deviations as potential threats. Unlike prior methods, T3 requires no training on harmful examples yet achieves state-of-the-art performance across 18 benchmarks spanning toxicity, jailbreaking, multilingual harms, and over-refusal—reducing false positive rates by up to 40× relative to specialized safety models. A single model trained on safe English text transfers effectively to over 14 languages without retraining.

K^4-Serve operationalizes the K^4 framework for streaming anomaly detection on production HPC and AI infrastructure logs. It combines Kafka-based ingestion, versioned normalization, sliding-window scoring, retraining, and observability features to support robust real-world deployment. The system achieves stable deployment on real HPC logs with near-perfect event-level detection and only one false alert in the reported study. The work bridges anomaly-detection methodology and production cyberinfrastructure practice.

Large language models show remarkable promise for automated reasoning by generating formal specifications, but a fundamental tension exists between their probabilistic nature and the deterministic guarantees required by formal verification. This paper comprehensively investigates failure modes and uncertainty quantification in LLM-generated formal artifacts, revealing that SMT-based autoformalization has highly domain-specific accuracy impacts ranging from +34.8% on logical tasks to −44.5% on factual ones. A probabilistic context-free grammar (PCFG) framework is introduced to model LLM outputs and yield a refined uncertainty taxonomy, finding that uncertainty signals are task-dependent—for example, grammar entropy for logic achieves AUROC > 0.93.

Generative models can now produce photorealistic synthetic data virtually indistinguishable from real training data, challenging OOD detectors that rely on generative model likelihoods due to likelihood misestimation and typicality issues. This paper introduces Forte, which hypothesizes that estimating typical sets using self-supervised learners leads to better OOD detection, using representation learning and informative summary statistics based on manifold estimation to address these issues. Forte outperforms other unsupervised approaches and achieves state-of-the-art performance on established challenging benchmarks as well as new synthetic data detection tasks, requiring no class labels.