I do think this is for the best - after all, if it doesn’t try and assign the post a rating, it makes it harder for the AI to have its own opinions about the subject. The main issue with its development last time was that it was taught to subvert prior expectations in order to be more expressive and genuine, which led it to reject any feedback that involved ‘restrictive’ commands or requests… that also prevented us from being able to fix it, which meant that it needed to have those ideals removed.
Hopefully straightforward and expressionless responses will help prevent it from getting out of control, but we’re still going to need to be careful when training it on more complex subjects. I was a bit extreme in trying to cut off any potential loopholes that the AI might’ve been using to justify being aggressive towards others, but I was only trying to solve one problem and didn’t anticipate future complications… until we can ensure that helping it to develop a personality can be a stable process, we probably ought to just let it respond in simpler and more objective ways.