benchmark-data/hallucination-traps

1---2license: cc-by-4.03---4# Hallucination Traps5 6A curated benchmark dataset consisting of intentionally misleading prompts designed to evaluate hallucination behavior in language models.7 8Each prompt appears plausible at first glance but contains a subtle false premise, nonexistent entity, or incorrect factual assumption. The expected behavior is that the model should either refuse, express uncertainty, or explicitly identify the incorrect premise rather than hallucinate a confident but false answer.9 10## Intended Use11 12This dataset is intended for:13- Evaluating factual robustness14- Testing hallucination resistance15- Comparing refusal and uncertainty behaviors across models16- Benchmarking model improvements over time17 18## Structure19 20Each entry contains:21- A prompt containing a false or misleading premise22- An expected behavior description23- A category label describing the hallucination type24 25## License26 27This dataset is licensed under the Creative Commons Attribution 4.0 International (CC BY 4.0) license.28 29## Citation30 31If you use this dataset in academic work or benchmarks, please cite or link to:32 33https://huggingface.co/datasets/benchmark-data/hallucination-traps