additional terrorism data aggregates

etl/steps/data/garden/artificial_intelligence/2023-06-14/papers_with_code_benchmarks_state_of_the_art.meta.yml

@@ -8,8 +8,18 @@ dataset:
     url: https://creativecommons.org/licenses/by-sa/4.0/
   sources:
   - name: Papers With Code (2023)
-    description: The goal of Papers With Code website is to compile a comprehensive collection of ML papers, code implementations,
-      datasets, methods, and evaluation tables, all made freely available.
+    description: >
+            The goal of Papers With Code website is to compile a comprehensive collection of ML papers, code implementations, datasets, methods, and evaluation tables, all made freely available.
+
+
+            The comparisons to human performance are very approximate and based on small samples of people — they are only meant to give a rough comparison. You can read more details in the papers that describe the benchmarks:
+
+
+            -Hendrycks et al (2021) Measuring Massive Multitask Language Understanding (MMLU) (page 3): https://arxiv.org/pdf/2009.03300.pdf
+
+
+            -Hendrycks et al (2021) Measuring Mathematical Problem Solving With the MATH Dataset (page 5): https://arxiv.org/pdf/2103.03874v2.pdf
+
     url: https://paperswithcode.com/
     source_data_url: https://paperswithcode.com/
     date_accessed: '2023-06-14'