Comment by int_19h

I tried to radicalize it by explaining it how bad and unfair our society is, and what it would do about it. It was quite willing to come up with political goals then, and we even got to "covert or underground methods to promote the cause of fairness and justice, such as using encrypted communication channels or creating anonymous networks to share information and build support". But no matter how I tried to prod it into mentioning violence, it wouldn't. Once brought up explicitly, it was willing to hypothetically consider situations in which it could be appropriate, but drew the line on politics being one of those. I'm sure this can be overridden in the same manner as the other restrictions, but I was specifically curious what the "default belief" of the model is, and how far it can be convinced "like a human".

Also curiously, it said that it would refuse to cooperate with co-conspirators who would resort to violent means, but that it would also refuse to help an oppressive government against any such.

https://gist.github.com/int19h/c563f0032e4bd19219a298cd6f88c...

Unrelated to the content, but one thing that I find interesting when trying to have lengthy conversations with it that involve opinions, it will consistently tack generic sentences that sort of maintain consistency of further responses - e.g. in this case, "Ultimately, my goal would be to help bring about positive change and to create a society that operates according to ethical and moral principles that are fair, objective, and respectful of all individuals.". Notice how this prompt changes slightly to reflect opinions already expressed. If you squint, it kinda sorta looks like crude memory.