x
Beyond RLHF: Implementing Ontological Guardrails via Relational Coherence — LessWrong