Update README.md

README.md

@@ -28,7 +28,7 @@ SHP exploits the fact that if comment A was written *after* comment B but has a
 If A had been written before B, then we could not conclude this, since its higher score could have been the result of more visibility from being written first.
 
 How is SHP different from [Anthropic's HH-RLHF dataset](https://huggingface.co/datasets/Anthropic/hh-rlhf)? 
-Most notably, all the data in SHP is naturally occurring and human-written, whereas the responses in HH-RLHF are machine-written, giving us two very different distributions.
+Most notably, all the data in SHP is naturally occurring and human-written, whereas the responses in HH-RLHF are machine-written, giving us two very different distributions that can complement each other.
 
 | Dataset              | Size | Input                      | Label                       | # Domains           | Data Format                           | Length |
 | -------------------- | ---- | -------------------------- | ---------------------------- | ------------------------- | ------------------------------------- | --------------- |