Thinking far ahead, is there some way we can train LLMs that will moderate the same way as dang?