Story
arxiv_cs_ai · Nov 3, 2026 · paper
Source brief
The Geometry of Ignorance: Detecting Suppressed Knowledge in LLMs through Activation Fragmentation
arxiv.orgNov 3, 2026
original source linked
In brief
Evaluating whether a large language model (LLM) genuinely knows a concept remains a fundamental problem, since behavioral metrics can be fooled by suppression: a model that refuses to answer may still harbor recoverable knowledge. The framework measures the geometric coherence of