Knowledge After Collapse: Context and Wording Shape Factual Accessibility in Language Models
arXiv:2611.17763v1 Announce Type: new Abstract: When a language model accepts a false correction in a factual discussion, its next answer can depend on how the conversation continues. We introduce Knowledge Answer Divergence (KAD), an event-conditioned evaluation protocol that distinguishes initial false-answer adoption from subsequent persistence, recovery, and abstention. KAD freezes the conversation at the first explicit adoption of a designated false target and branches independent probes fr
Read original sourceSource record
Project analysis
Project analysis is not yet available. See the source record above.