LLM-based regression has largely focused on predicting a single scalar value, while multi-output text regression—where a model predicts several related text outputs from one input—remains comparatively unexplored. A new arXiv preprint, DualView, aims to close that gap by introducing a parameter-efficient method for this task.

The method is designed to work with large language models without requiring extensive parameter updates, making it practical for settings where computational resources are limited. Based on the paper's title, DualView combines broad evidence with output-specific evidence, suggesting it draws on both general context and details relevant to each individual output.

Because the paper is a preprint, its claims have not yet undergone peer review. Still, the work highlights an important direction: extending LLM regression beyond simple scalar predictions to richer, multi-output scenarios.