Reproduce “FATE-5: Faithful Alignment of Trait Expression for Continuous Big Five Personality Control” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Gunhee Cho, Hongyeob Kim, Yun-Gyung Cheong
FATE-5 controls chatbot personality using five adjustable Big Five scores. It encodes profiles as soft prompts, then applies preference alignment to improve naturalness while retaining trait expression. Human evaluation on Llama-3-8B found 78.7 trait accuracy and 67.8 naturalness; prompting scored 73.3 and 74.4, respectively. On Gemma-2-9B, questionnaire profile recovery RMSE was 8.9, but this does not establish human-like personality.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.