HalluWorld: A Controlled Benchmark for Hallucination via Reference World Models

In three linesHalluWorld is a controlled benchmark for evaluating LLM hallucinations through explicit reference worlds (gridworlds, chess, terminal tasks). Frontier models solve perceptual hallucinations on direct observations well, but struggle with multi-step state tracking and causal forward simulation, even with extended thinking.

Summary generated by Claude — human-verified