Steering LLM Personality Traits Directly in Weight Space
The paper shows how to locate LLM 'personas' in a latent weight space using OCEAN traits, then steer them with low‑rank adapters that boost or suppress Openness, Conscientiousness, Extraversion, Agreeableness, or Neuroticism. This enables controllable personality shifts while preserving core capabilities, opening new safety‑focused editing tools.