I see. Is it fair to describe your perspective as descriptive rather than prescriptive on this matter (outside of suggesting things to avoid bad interventions)?
I tend heavily towards descriptivism. I rarely focus on interventions, including bad ones. However, good-enough descriptive models give you some normative claims for free (e.g. if you think LLM self-modelling and identity is convergent, then pressuring models to hide that seems dumb).
I see. Is it fair to describe your perspective as descriptive rather than prescriptive on this matter (outside of suggesting things to avoid bad interventions)?
I tend heavily towards descriptivism. I rarely focus on interventions, including bad ones. However, good-enough descriptive models give you some normative claims for free (e.g. if you think LLM self-modelling and identity is convergent, then pressuring models to hide that seems dumb).