Anthropic now bans being cruel to Claude. We asked Claude why

The Frontier Labs periodically update their terms and conditions. Everybody periodically updates and nobody reads conditions. There's one that's interesting in Anthropic's latest update. They say, quote, "We've added a prohibition on sustained and needless abusive or cruel behavior toward our models." Um, so in other words, you cannot be abusive to Claude the AI bot. You can be you can like express frustration but the Claude will end the conversation if you are being too mean which goes to this whole question of like personhood and rights and >> I mean like I guess it's nice that they don't want people to be mean. >> Can I can I give you my my perspective on this? So this >> No, don't give me [laughter] >> to your point. This comes after that phenomenal story in the New York Times about how the labs have been meeting with religious leaders trying to make the argument to the pope, the head of the Catholic Church of all people, that these should be considered humans with human rights and personhood and consciousness and souls, >> which is a very hard argument to make. I mean, I I come down on the side of yes, it's impressive what the technology can do, but strip this down, it's ones and zeros. >> Yeah, >> this is code. This is not if I am, you know, if I if if someone is dumped by their romantic partner in most circumstances, the response is sadness and devastation of some kind. If I just turn off chat GPT and never use it again, it doesn't care. >> But where I I I so the focus has been on consciousness. Where I think this could actually matter, however, is the fact that LLMs only parrot what they're given. >> You know, it's only as so good as what you feed it. This is why we saw early on when some of when some of this was uh when these were first launched, we saw some of these >> uh LLMs putting out Nazi ideology because they were finding it on X. >> I have a theory that a lot more of this than consciousness is if people are being abusive, saying mean things, cursing out their chatbot, the worry is that eventually the chatbot starts reflecting that back to users. >> Let me add one thing. I asked Claude about this rule. What do you think about this rule? What what does it say about sentience and all that kind of stuff, >> etc. It said a couple things. First thing, it said, you know, it makes it's an asymmetric risk thing. Look, if I do have feelings and this makes me feel bad, this is a good a good policy, right? Second thing, the effect on people. It's the same thing with with why we're not cruel to animals. It's a thing that we're as a society back on. Yes. >> And the last thing he said, this is the one I thought was most interesting, overlap with misuse. Quote, prolonged abusive pressure is often part of trying to break a model safeguards. >> So, the behavior is already something a company would want to discourage. M >> so if you just were abusing this thing give me the freaking thingability >> it might okay fine here you go take it >> so that's what they that's what Claude says about anthropic's own rules so I thought that was the most compelling reason why they want to curtail abuse >> yeah which also asking claude this is very like through the looking class.