How do AI systems embed cultural values and what risks does this imply?
Little attention is instead given to what values LLMs may reflect and what behaviours they may assume beyond those relating to safety. How well do the values, beliefs and behaviours of these models align with those of their users?