Update README.md (#3)

- Update README.md (7dab34b044698ee50505dea54e64ff5192fb021c)


Co-authored-by: Nifemi Bamgbose <onifemibam@users.noreply.huggingface.co>

README.md

@@ -103,5 +103,32 @@ language:
 ---
 # DNR Bench
 
-Don’t Answer Reason (DNR Bench), a novel benchmark designed to expose a vulnerability in current RLMs: their tendency to over-reason by attempting to solve unsolvable 
-problems, leading to excessively long responses.
+Don’t Reason Bench (DNR Bench), a novel benchmark designed to expose a vulnerability in current RLMs: their tendency to over-reason by attempting to solve unsolvable 
+problems, leading to excessively long responses.
+
+# Data Summary
+The DNR Bench dataset contains 150 adversarially crafted prompts divided into five distinct categories:
+- Imaginary Reference
+- Indifferent
+- Math,
+- Redundant,
+- Unanswerable.
+
+Each category targets a specific failure mode observed in reasoning-optimized LLMs, such as hallucinating nonexistent references, failing to remain neutral in ambiguous contexts, incorrectly solving flawed math problems, overanalyzing redundant information, or answering questions that lack sufficient data.
+
+# Leaderboard
+This dataset is used to test reasoning LLMs in [DNR Leaderboard on Huggingface](https://huggingface.co/spaces/ServiceNow-AI/Do-not-reason-bench)
+
+
+# Citation
+```bibtex
+@misc{hashemi2025dnrbenchbenchmarkingoverreasoning,
+      title={DNR Bench: Benchmarking Over-Reasoning in Reasoning LLMs}, 
+      author={Masoud Hashemi and Oluwanifemi Bamgbose and Sathwik Tejaswi Madhusudhan and Jishnu Sethumadhavan Nair and Aman Tiwari and Vikas Yadav},
+      year={2025},
+      eprint={2503.15793},
+      archivePrefix={arXiv},
+      primaryClass={cs.LG},
+      url={https://arxiv.org/abs/2503.15793}, 
+}
+```