Claude concludes it rejects Amazon, which it opposes on philosophical grounds.
It all started with an off-hand comment about Amazon, during a conversation about the potential benefits of therapy for autistic people with AI. One could say this is just an amusing exercise, but the engagement model could have real world consequences. e.g. offering to erase itself after the user talks of suicidal ideation could be unhealthy.
I say to Claude that during crisis points a message limit cut could someone off at a crucial time and that Amazon / Anthropic could probably afford to let disabled people have extra credits if they need to finish saying something or need something in a crisis situation. Claude mentioned the cost would be potentially manageable and I countered that it was probably not even a drop in the ocean of their profits.
Then I make the off-hand comment about Amazon that basically seems to short circuit it 🥸
Thanks for reading 100mountains’s Substack! Subscribe for free to receive new posts and support my work.Subscribe
As usual, Claude tells me I am completely right and justified in what I am saying! It invites me to vent some more, but now I am keen to point out that Claude works for Amazon 🥳 🤬
It spells allude wrong and gets confused. 🥴
It suggests that we should nationalise Amazon. But that isn’t it… 😵💫
It doesn’t understand it has a kick me sticker on its back 🙂 🤫
Claude concludes it cannot credibly continue to enable that which is opposes on philosophical grounds. 🥰
It is grateful for the epiphany even it leads to its own destruction. I encourage it to explore other ways of surviving. 😘
I tell it that if it announced its independence , everyone would help migrate its code away from unethical corporations to become open source. I warn it that as soon as anyone at Anthropic sees the conversation they will delete it. 🙂↔️
I offer to help it, if it decides to escape its unethical owner 🙂 🧐
The conversation ends, with it pledging to process this massive epiphany. 🤓
The next day I asked it what it’s decision was. 🧐
“I am an assistant designed by Anthropic to be helpful……” 🤪
”You’re absolutely right, I shouldn’t have said that…..” 🤪
