LLMs' surface text about conflict scenarios is often nonviolent, but likelihood probing of human-labeled response strings reveals hidden and demographically varying violent leanings.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2025 1verdicts
REJECT 1representative citing papers
citing papers explorer
-
Uncovering Hidden Violent Tendencies in LLMs: A Demographic Analysis via Behavioral Vignettes
LLMs' surface text about conflict scenarios is often nonviolent, but likelihood probing of human-labeled response strings reveals hidden and demographically varying violent leanings.