InvThink is a three-step premortem reasoning framework that improves LLM safety scores across general and domain-specific tasks while reducing the typical safety-performance trade-off.
(same with Figure 12)
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
-
InvThink: Premortem Reasoning for Safer Language Models
InvThink is a three-step premortem reasoning framework that improves LLM safety scores across general and domain-specific tasks while reducing the typical safety-performance trade-off.