A new PDDL benchmark and modular test suite shows that LLMs reject clear safety violations but fail to plan around subtle physical hazards.
Title resolution pending
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.AI 1years
2025 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
Subtle Risks, Critical Failures: A Framework for Diagnosing Physical Safety of LLMs for Embodied Decision Making
A new PDDL benchmark and modular test suite shows that LLMs reject clear safety violations but fail to plan around subtle physical hazards.