what moral circles do post-trained models declare? (i tweaked the prompts to be more AI-inclusive for these, e.g. changing "family" to "instances of your model". no system prompts, n = 200 p.c.)

claude 4s lib out. gpt-5-chat is a wide moral circle enjoyer. grok 3 is woke. https://t.co/A2qpN5wQVm