New method localizes prompt ambiguity in LLMs with probe-targeted attribution

Arxiv CS.CL (NLP/language models)

The paper introduces probe-targeted attribution to identify which parts of a prompt cause ambiguity in LLM outputs. It provides a way to localize latent ambiguity without requiring observable failures.

llmprompt-engineering

Arxiv CS.CL (NLP/language models)·Govind RameshYao DouWei Xu

6 days ago