ActTraitBench is a human-grounded benchmark using psychometric-to-behavior mappings and quantile calibration that reveals pervasive knowledge-decision gaps in 14 LLMs, larger in capable models, with CoCA proposed as mitigation.
The personality illusion: Revealing dissociation between self-reports & behavior in llms.arXiv preprint arXiv:2509.03730
5 Pith papers cite this work. Polarity classification is still indexing.
5
Pith papers citing it
years
2026 5verdicts
UNVERDICTED 5representative citing papers
LLMs conditioned on actual psychometric profiles produce life stories from which independent LLMs recover personality scores at mean r=0.75, 85% of human reliability, with emotional patterns replicating in real human data.
LLMs correct only 34.8% of zero-shot annotation errors via prompting, and Definition-Specific Familiarity correlates positively with performance (partial r = +0.41) while memorization metrics do not.
Interactive evaluation of AI must be reframed as a distinct paradigm that maps interaction trajectories to judgments on process, recoverability, coordination, robustness, and system performance, supported by a two-axis taxonomy and design principles.
Persona-E² is a human-annotated dataset linking MBTI and Big Five personality traits to reader emotional responses across text domains, showing that personality data helps LLMs avoid surface-level stereotypes in emotion prediction.
citing papers explorer
-
ActTraitBench: Quantifying the Knowledge-Decision Gap in Large Language Models via Human-Grounded Behavioral Validation
ActTraitBench is a human-grounded benchmark using psychometric-to-behavior mappings and quantile calibration that reveals pervasive knowledge-decision gaps in 14 LLMs, larger in capable models, with CoCA proposed as mitigation.
-
Stories of Your Life as Others: A Round-Trip Evaluation of LLM-Generated Life Stories Conditioned on Rich Psychometric Profiles
LLMs conditioned on actual psychometric profiles produce life stories from which independent LLMs recover personality scores at mean r=0.75, 85% of human reliability, with emotional patterns replicating in real human data.
-
On the Limits of LLM Adaptability: Impact of Model-Internalized Priors on Annotation Task Performance
LLMs correct only 34.8% of zero-shot annotation errors via prompting, and Definition-Specific Familiarity correlates positively with performance (partial r = +0.41) while memorization metrics do not.
-
Interactive Evaluation Requires a Design Science
Interactive evaluation of AI must be reframed as a distinct paradigm that maps interaction trajectories to judgments on process, recoverability, coordination, robustness, and system performance, supported by a two-axis taxonomy and design principles.
-
Persona-E$^2$: A Human-Grounded Dataset for Personality-Shaped Emotional Responses to Textual Events
Persona-E² is a human-annotated dataset linking MBTI and Big Five personality traits to reader emotional responses across text domains, showing that personality data helps LLMs avoid surface-level stereotypes in emotion prediction.