The primary issue is that language doesn't have as well of a defined "success" metric. So more data doesn't necessarily make better language. Without a human a

The primary issue is that language doesn't have as well of a defined "success" metric. So more data doesn't necessarily make better language. Without a human analyzing the transcripts it's very difficult for the chat bots to know which inputs it's receiving are "good" or "better." Even the idea of good language is subjective. We all know there is such a thing, but nearly everyone has different ideas of what this is.