Update README.md

README.md

@@ -46,7 +46,7 @@ GPT-2 can contain many biases and factual inaccuracies, however, given the model
 
 For training this model, I use the GPT Wiki Intro dataset, which consists of Wikipedia intro paragraphs written by GPT-3.
 The Curie model variant is used to generate intro paragraphs given the title of the article and the first 7 words as a prompt.
-150k topics are present in the dataset, the creators did not share how these topics were selected.
+150k topics are present in the dataset; the creators did not share how these topics were selected.
 Each LLM-written paragraph is paired with the human written version; having paired examples from the same domain is why I chose this dataset.
 For computational efficiency, I use only 4000 topics for training and 1000 for testing.