Tag
This paper proposes using Natural Semantic Metalanguage (NSM) primes as a more basic explanation for emotions in LLMs, showing they are recoverable, causally controllable, and faithful, outperforming appraisal-based directions.