TPA attributes next-token probabilities in RAG to seven LLM components and uses POS-tag anomalies to detect hallucinations more comprehensively than prior binary FFN-context conflict views.
Summing over L layers, this costs O(L·T· |V| ·d)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
UNVERDICTED 1representative citing papers
citing papers explorer
-
TPA: Next Token Probability Attribution for Detecting Hallucinations in RAG
TPA attributes next-token probabilities in RAG to seven LLM components and uses POS-tag anomalies to detect hallucinations more comprehensively than prior binary FFN-context conflict views.