An audit of five large language models across four genocides finds accuracy drops sharply for the less-documented Cambodian Genocide and for Khmer-language prompts, suggesting AI can propagate denialist narratives.
(1976, October 4)
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CY 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
From prosthetic memory to prosthetic denial: Auditing whether large language models are prone to mass atrocity denialism
An audit of five large language models across four genocides finds accuracy drops sharply for the less-documented Cambodian Genocide and for Khmer-language prompts, suggesting AI can propagate denialist narratives.