Small Language Model Resolves LLM Prompt Ambiguity

Researchers propose a pre-inference prompt optimization mechanism using a small language model to resolve semantic ambiguity in LLM prompts. Large language models struggle with natural language prompts that lack proper syntactic rules, leading to multiple interpretations and incorrect reasoning paths. Prior work applied query editing during inference without addressing the root cause. The new method identifies semantic risks, checks multi-perspective consistency, and resolves conflicts before inference.