minor text updates

README.md

@@ -39,9 +39,9 @@ Note: The non-commercial variants of this model are explicitly not intended to b
 * Text in images produced by the model will likely be difficult to read.
 * The model struggles with more complex tasks that require compositional understanding
 * It may not accurately generate faces or representations of specific people.
-* The model primarily learned from English descriptions and may not perform as effectively in other languages.
+* CommonCatalog (the training dataset) contains synthetic captions that are primarily English-language text; our models may not perform as effectively when prompted in other languages.
 * The autoencoder aspect of the model introduces some information loss.
-* It may be possible to guide the model to generate objectionable content, i.e. nudity or other NSFW material.
+* It may be possible to guide the model to generate harmful content, i.e. nudity or other NSFW material.
 
 ## Intended Uses
 * Using the model for generative AI research
@@ -52,7 +52,7 @@ Note: The non-commercial variants of this model are explicitly not intended to b
 * Research on generative models.
 
 ## Usage
-We recommend using the MosaicML Diffusion Repo to finetune / train the model: https://github.com/mosaicml/diffusion . Example finetuning code coming soon. See the Gradio Demo [Here](https://github.com/mosaicml/diffusion/blob/main/scripts/gradio_demo.py)
+We recommend using the [MosaicML Diffusion repository](https://github.com/mosaicml/diffusion) to finetune / train the model. Example finetuning code coming soon. See the Gradio Demo [Here](https://github.com/mosaicml/diffusion/blob/main/scripts/gradio_demo.py)
 
 ## Evaluation/Validation
 We validated the model against Stability AI’s SD2 model and compared human user study