WildDelusionCombined is a retrieval-enriched collection of 493 LLM-context-
verified target turns from 303 source conversations that potentially endorse
highly implausible beliefs. It contains only redistributable WildChat and
ShareChat material. No LMSYS-Chat-1M conversation text is included.
Labels describe messages in conversational context. They are not diagnoses of
users, human-adjudicated clinical ground truth, or prevalence estimates.… See the full description on the dataset page:
https://huggingface.co/datasets/danielfein/WildDelusionCombined.
