Research datasets from a 2-week investigation into personality steering in large language models. We mapped how personality traits (sarcasm, character voice, reasoning style) are represented and can be steered in Qwen3-VL-8B, Qwen3.5-27B, and GPT-OSS-20B.
GitHub: Atlas3DSS/Character-Creation