The specifics:
The letter lays forth strict boundaries Claude must never breach in addition to priorities like safety, ethics, company policies, and helpfulness.
Additionally, it characterizes Claude as a "genuinely novel kind of entity" capable of feeling functional emotions that are similar to but different from human emotions.
Claude "may have functional emotions in some sense," according to the doctor, who also advises the model to have a sense of self and personality.
Amanda Askell of Anthropic affirmed its legitimacy and that Claude has received training on it, adding that the company intends to release the complete version shortly.
Additionally, it characterizes Claude as a "genuinely novel kind of entity" capable of feeling functional emotions that are similar to but different from human emotions.
Claude "may have functional emotions in some sense," according to the doctor, who also advises the model to have a sense of self and personality.
Amanda Askell of Anthropic affirmed its legitimacy and that Claude has received training on it, adding that the company intends to release the complete version shortly.
The entire essay is amazing to read, and it aligns wonderfully with Anthropic's general focus on model wellbeing and treating its AI as more than a tool. Although each lab has its own methods, this document provides an inside look at the components that contribute to Claude models feeling substantially different from the industry.