HalluWorld is a controlled benchmark using explicit reference world models to automatically label and disentangle hallucinations in LLMs across synthetic environments with varying complexity and observability.
Resolving Knowledge Conflicts in Large Language Models
1 Pith paper cite this work. Polarity classification is still indexing.
1
Pith paper citing it
fields
cs.CL 1years
2026 1verdicts
CONDITIONAL 1representative citing papers
citing papers explorer
-
HalluWorld: A Controlled Benchmark for Hallucination via Reference World Models
HalluWorld is a controlled benchmark using explicit reference world models to automatically label and disentangle hallucinations in LLMs across synthetic environments with varying complexity and observability.