A new preprint on arXiv presents Big5-Scaler, a framework for giving large language models controllable Big Five personality traits. Rather than modifying model weights, the method works entirely through prompts, which could make personality conditioning easier to apply across different models and use cases.

The framework embeds numeric trait values into natural language prompts, allowing a user to specify a desired personality profile in the prompt itself. This lets the model adjust its responses according to the given trait scores, such as higher extraversion or lower neuroticism, without needing task-specific fine-tuning.

As a preprint, the paper has not yet undergone peer review, but it points to a growing interest in steering LLM behavior through lightweight, prompt-based interventions. The title emphasizes scaling, suggesting the authors aim to make personality control practical for large-scale deployment.