Update README.md

README.md

@@ -19,6 +19,7 @@ BERT-Grader is a finetune of distil-bert, for performance.
 ## Biases and Issues
 For BERT-Grader to perform well, the input text should be an actual essay.
 Some biases may arise due to training data phrases, but there were mesures in place to limit bias.
+Giberrish is accounted for, and will be rather random. I suggest using a spell-checker with this model, and mesure tokens.
 
 ## Validation Metrics
 loss: 0.166