Exploring Why Giving AI a Persona Is an Effective Post-Training Strategy

On November 13, 2026, at 15:07, Juejin published an article titled "Why Granting AI a Persona Is an Effective Post-Training Strategy," opening up discussion on this question.

Approaching the topic from an engineering standpoint, the article breaks down the technical nature of a base model, arguing that it is merely a probabilistic aggregation of massive amounts of human text. As a result, it lacks built-in value filtering and behavioral constraints, which makes it prone to generating harmful or meaningless output. Building on that analysis, the article examines why assigning a persona to AI can work as an effective post-training strategy. It pushes back against the view that AI should be fully de-personified and reduced to a calculator-like form, and frames the question of model alignment and behavior control through an engineering lens. At present, the discussion remains at the level of the analysis and position set out in this single piece, with no further developments reported.