Interventions in LLM-simulated user experiments induce distribution shifts in latent attributes that create confounding bias, diagnosable with negative control outcomes and partially mitigated by adding setting-relevant persona details.
Proceedings of the 40th International Conference on Machine Learning , pages =
2 Pith papers cite this work. Polarity classification is still indexing.
2
Pith papers citing it
years
2026 2verdicts
UNVERDICTED 2representative citing papers
LLM simulations of user design preferences show significant systematic discrepancies from real aggregated user data across multiple experimental manipulations, with synthetic justifications lacking depth and relying on generic patterns instead.
citing papers explorer
-
The Illusion of Intervention: Your LLM-Simulated Experiment is an Observational Study
Interventions in LLM-simulated user experiments induce distribution shifts in latent attributes that create confounding bias, diagnosable with negative control outcomes and partially mitigated by adding setting-relevant persona details.
-
Distorted Perspectives of LLM-Simulated Preferences: Can AI Mislead Design?
LLM simulations of user design preferences show significant systematic discrepancies from real aggregated user data across multiple experimental manipulations, with synthetic justifications lacking depth and relying on generic patterns instead.