Add license, task category, and paper/code links

README.md

@@ -1,4 +1,7 @@
 ---
+license: mit
+task_categories:
+- text-generation
 dataset_info:
 - config_name: Qwen--Qwen2.5-Math-1.5B-Instruct
   features:
@@ -470,8 +473,11 @@ configs:
   - split: test
     path: openai--gpt-oss-20b_medium/test-*
 ---
+
 # Generations Dataset: MATH
 
+This dataset accompanies the paper [LLMs Encode Their Failures: Predicting Success from Pre-Generation Activations](https://huggingface.co/papers/2602.09924). Code is available at [https://github.com/KabakaWilliam/llms_know_difficulty](https://github.com/KabakaWilliam/llms_know_difficulty).
+
 LLM-generated solutions across train/validation/test splits for multiple models.
 
 ## Columns
@@ -505,4 +511,4 @@ train = dataset["train"]
     author = {Lugoloobi, William and Foster, Thomas and Bankes, William and Russell, Chris},
     year = {2026},
 }
-```
+```