Threat-based prompts change LLM output length, style, and certainty, but the claimed performance gains are not backed by accuracy measures.
Impact of Prompt Fram- ing on Factuality in Language Models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
citation-role summary
background 1
citation-polarity summary
fields
cs.CR 1years
2025 1verdicts
REJECT 1roles
background 1polarities
support 1representative citing papers
citing papers explorer
-
Analysis of Threat-Based Manipulation in Large Language Models: A Dual Perspective on Vulnerabilities and Performance Enhancement Opportunities
Threat-based prompts change LLM output length, style, and certainty, but the claimed performance gains are not backed by accuracy measures.