New SPP Method Aligns Model Values via Pretraining

A new paper proposes Synthetic Persona Pretraining (SPP), injecting synthetic moral reflections during pretraining to significantly alter values and improve alignment in 3B models.

2026-08-14 ~ 2026-08-15 · 2 related posts

• New Paper: Injecting Synthetic Moral Reflections During Pretraining Shifts Value Priorities in 3B Models — sethlazar · 2026-08-14
• Synthetic Persona Pretraining (SPP): Injecting Desired Persona from Token Zero Boosts Alignment in 3B Models — dhadfieldmenell · 2026-08-15