Pressure-Testing Deception Probes in LLMs: Scaling, Robustness, and the Geometry of Deceptive Representations

Documenti analoghi

• Entropy-Based Measurement of Value Drift and Alignment Work in Large Language Models
  di: Fadli, Samih
  Pubblicazione: (2025)
• Weakly Supervised Distillation of Hallucination Signals into Transformer Representations
  di: Salehmohamed, Shoaib Sadiq, et al.
  Pubblicazione: (2026)
• Contextual Integrity in LLMs via Reasoning and Reinforcement Learning
  di: Lan, Guangchen, et al.
  Pubblicazione: (2025)
• Counterfactual Likelihood Tests for Indirect Influence in Private Reasoning Channels
  di: Lorup, Alexander Boesgaard
  Pubblicazione: (2026)
• Adapting While Learning: Grounding LLMs for Scientific Problems with Intelligent Tool Usage Adaptation
  di: Lyu, Bohan, et al.
  Pubblicazione: (2024)