Update README.md

README.md

@@ -44,7 +44,7 @@ configs:
 
 ## Introduction
 
-[SWE-bench](https://www.swebench.com/) is a popular benchmark that measures how well systems can solve real-world software engineering problems. To solve SWE-bench problems, systems need to interact with large codebases that have long commit histories. Interacting with these codebases in an agent loop using git can be slow and take up large amounts of storage space. 
+[SWE-bench](https://www.swebench.com/) is a popular benchmark that measures how well systems can solve real-world software engineering problems. To solve SWE-bench problems, systems need to interact with large codebases that have long commit histories. Interacting with these codebases in an agent loop using git naively can be slow, and the repositories themselves take up large amounts of storage space. 
 
 This dataset provides the complete Python codebase snapshots for all problems in the [SWE-bench Verified](https://openai.com/index/introducing-swe-bench-verified/) dataset. For each problem instance, it includes all Python files present in the repository at the commit hash specified by the original SWE-bench Dataset.
 
@@ -54,7 +54,7 @@ The dataset consists of two main components:
 1. `file_content`: Contains the actual content of all unique files
 2. `problem_files`: Maps each problem instance to its relevant files
 
-Here's an example of how to load and use the dataset:
+Here's an example of how to load and use the dataset to access the files for a specific problem:
 
 ```python
 from datasets import load_dataset