SOURCE-LINKED INTELLIGENCE

Scored vs. Generated Readouts in Behavioral Language Models: An Empirical Study of Elicitation Format

arXiv · AI, language, vision and robotics · article · Sep 9, 2026 · UTC

Language models fine-tuned on customer behavior can predict outcomes and generate explanations, but these readouts are often treated as interchangeable. Holding model checkpoint and prompt content fixed, we compare probabilities obtained by scoring answer tokens with predictions generated after a written rationale. Across 13 model-domain cells covering four retail tasks in three markets, including two using fully public data and checkpoints, the scored readout ranks outcomes more accurately in 12 of 13 cells (two-sided sign test, p approximately 0.003), by 1.5 to 14.5 points in area under the

Read original source ↗ Open in workspace

recordType

paper

region

Global