LLM Variance in Open-Ended Information Elicitation
October 7, 2026
Evaluation of 11 LLMs reveals significant differences in how model families and parameter scales judge information value during sequential seeking. The study uses a controlled simulation to isolate model judgment from question generation and respondent behavior.
HOW THIS AFFECTS YOU
●
researcherYou should account for model-specific biases in how agents prioritize information during sequential decision-making tasks.