Sanju Tiwari

Layer-wise symbolic attention instability as a diagnostic signal for hallucination in large language models

A unified symbolic, behavioral, and mechanistic framework that connects symbolic triggers with internal failure dynamics in transformer architectures and provides an interpretable basis for diagnosing and stabilizing symbolic reasoning in LLMs is introduced.