diff --git a/README.md b/README.md
index e52213db80..01134ad458 100644
--- a/README.md
+++ b/README.md
@@ -39,8 +39,8 @@ We reimplemented all model architectures and training recipes from scratch for 4
 
 1. Remove all abstraction layers and have single file implementations.   
 2. Guarantee Apache 2.0 compliance to enable enterprise use without limits.    
-3. Optimized every detail of every model to get the fastest performance possible to lower cost and training speeds.    
-4. Config based for highly-optimized recipes.           
+3. Optimized each model architectural detail to maximize performance, reduce costs, and speed up training.    
+4. Highly-optimized recipe configs we have tested at enterprise scale.