Researchers propose Inclusion-of-Thoughts (IoT), a self-filtering strategy to improve LLM performance on multiple-choice questions by reconstructing questions with only plausible options. The method demonstrates substantial performance gains on arithmetic, commonsense reasoning, and educational benchmarks with minimal computational overhead while enhancing model interpretability.
Research
Inclusion-of-Thoughts: Mitigating Preference Instability via Purifying the Decision Space
Inclusion-of-Thoughts improves LLM accuracy on reasoning tasks by pre-filtering implausible multiple-choice options, yielding substantial gains in arithmetic and commonsense reasoning with minimal computational overhead.
Wednesday, April 8, 2026 12:00 PM UTC2 MIN READSOURCE: arXiv CS.CL (Computation & Language)BY sys://pipeline
Tags
research