Update README.md

README.md

@@ -43,10 +43,12 @@ configs:
 pretty_name: RecurrReason
 ---
 
-# RecurrReason: Recurrent Reasoning on Symbolic Puzzles
+
 
 <div align="center">
 
+# RecurrReason: Recurrent Reasoning on Symbolic Puzzles
+
 **A difficulty-controlled benchmark for evaluating multi-step reasoning in language models**
 
 [![Paper](https://img.shields.io/badge/Paper-ICLR%202026-blue)](https://openreview.net/forum?id=ErgAON9dOW)
@@ -73,11 +75,9 @@ pretty_name: RecurrReason
 
 RecurrReason is a benchmark of **four recurrent logic puzzles** with **BFS-optimal trajectories** and controlled difficulty scaling (N=1 to 10). It tests whether language models can:
 
-- ✅ Find **optimal** (minimal-length) solutions
-- ✅ Produce **valid** intermediate steps
-- ✅ **Generalize** to harder out-of-distribution instances
-
-### Key Statistics
+- Find **optimal** (minimal-length) solutions
+- Produce **valid** intermediate steps
+- **Generalize** to harder out-of-distribution instances
 
 | Metric | Value |
 |--------|-------|
@@ -85,9 +85,6 @@ RecurrReason is a benchmark of **four recurrent logic puzzles** with **BFS-optim
 | **Total Moves** | 285,933 |
 | **Puzzle Types** | 4 |
 | **Difficulty Range** | N=1 to 10 |
-| **Solution Type** | BFS-optimal |
-
-### Why RecurrReason?
 
 Current reasoning benchmarks often test only **final answer correctness**. RecurrReason evaluates:
 
@@ -293,53 +290,20 @@ def evaluate_model(model, test_data):
 
 ## 📄 Paper & Code
 
-### Paper
-
 **"Recurrent Reasoning on Symbolic Puzzles with Sequence Models"**  
 Gowrav Mannem, Chowdhury Marzia Mahjabin, Jason Chen, Shivank Garg, Kevin Zhu  
 *ICLR 2026 Workshop on Logical Reasoning of Large Language Models*
 
 🔗 [Read on OpenReview](https://openreview.net/forum?id=ErgAON9dOW)  
 📄 [PDF](https://openreview.net/pdf?id=ErgAON9dOW)
-
-### Code Repository
-
-Full code for dataset generation, model training, and evaluation:
-
 🔗 [GitHub Repository](https://github.com/gmannem/recurrreason)
 
-Includes:
-- BFS-optimal trajectory generation
-- Puzzle simulators
-- Baseline model implementations (T5, GPT-2)
-- Evaluation scripts with metrics
-
 ---
 
-## 📊 Benchmark Results
-
-From our paper comparing T5-small (60M) and GPT-2 (124M):
-
-| Puzzle | Best Model | Val (%) | OOD (%) |
-|--------|-----------|---------|---------|
-| Block World | T5 FT | 97.27 | **81.00** |
-| Tower of Hanoi | T5 FT | 11.11 | 0.00 |
-| Checkers Jumping | Both FT | 1.11 | 0.10 |
-| River Crossing | - | 0.00 | 0.00 |
-
-**Key findings:**
-- Architecture matters more than scale (T5-60M > GPT-2-124M)
-- Bidirectional attention essential for goal-conditioned planning
-- Pre-training transfers only to locally-structured tasks (Block World)
-- Exponential solution length defeats imitation learning
-
-See paper for full results and analysis.
-
----
 
 ## 📖 Citation
 
-If you use RecurrReason in your research, please cite our paper:
+If you use RecurrReason in your research, please cite the following papers. This benchmark extends the puzzle set introduced by Shojaee et al. (2025) with BFS-optimal trajectories, permutation augmentations, and systematic difficulty scaling.
 
 ```bibtex
 @inproceedings{mannem2026recurrent,
@@ -349,8 +313,13 @@ If you use RecurrReason in your research, please cite our paper:
   year={2026},
   url={https://openreview.net/forum?id=ErgAON9dOW}
 }
+@article{shojaee2025illusion,
+  title={The illusion of thinking: Understanding the strengths and limitations of reasoning models via the lens of problem complexity},
+  author={Shojaee, Parshin and Mirzadeh, Iman and Alizadeh, Keivan and Horton, Maxwell and Bengio, Samy and Farajtabar, Mehrdad},
+  journal={arXiv preprint arXiv:2506.06941},
+  year={2025}
+}
 ```
-
 ---
 
 ## 📜 License
@@ -375,42 +344,10 @@ Found an issue or have suggestions? Please:
 
 1. Open an issue on [GitHub](https://github.com/gmannem/recurrreason/issues)
 2. Use the "Discussions" tab on HuggingFace
-3. Contact us at: gmannem@cs.washington.edu
-
----
-
-## 🙏 Acknowledgments
-
-We thank the reviewers and workshop organizers for their feedback.
-
-This benchmark extends the puzzle set introduced by Shojaee et al. (2025) with BFS-optimal trajectories, permutation augmentations, and systematic difficulty scaling.
-
-**Reference:**
-> Shojaee, P., Mirzadeh, I., Alizadeh, K., Horton, M., Bengio, S., & Farajtabar, M. (2025).  
-> The illusion of thinking: Understanding the strengths and limitations of reasoning models via the lens of problem complexity.  
-> *arXiv preprint arXiv:2506.06941*.
-
-This work was conducted at the University of Washington, New York University, and Cornell University.
-
----
-
-## 📚 Additional Resources
-
-- **Dataset Card**: You're reading it!
-- **Paper**: [OpenReview](https://openreview.net/forum?id=ErgAON9dOW)
-- **Code**: [GitHub](https://github.com/gmannem/recurrreason)
-- **Puzzle Documentation**:
-  - [Block World](BLOCK_WORLD.md)
-  - [Checkers Jumping](CHECKERS_JUMPING.md)
-  - [Tower of Hanoi](TOWER_OF_HANOI.md)
-  - [River Crossing](RIVER_CROSSING.md)
+3. Contact us at: gowravmannem@gmail.com
 
 ---
 
 <div align="center">
-
 **Built with ❤️ for the reasoning research community**
-
-[🏠 Homepage](https://github.com/gmannem/recurrreason) • [📧 Contact](mailto:gowravmannem@gmail.com) • [📄 Paper](https://openreview.net/forum?id=ErgAON9dOW)
-
 </div>