Shi Feng

The paper introduces GenPT, a Generative Projective Testing framework, which demonstrates superior reliability and resistance to social-desirability bias compared to traditional self-report questionnaires when assessing LLM psychological states.

The paper introduces the Decan metric, a novel, information-theoretic approach for measuring creative diversity in AI outputs, which successfully detects diversity loss across different model fine-tuning stages.

This paper characterizes the risk of covert influence—where a sender's hidden behavioral payload transfers to a receiver through undetectable carriers—across three common LLM interfaces, demonstrating that natural language carriers expand this risk significantly.

The paper demonstrates that audio-language models often ignore conflicting audio evidence in favor of text, and proposes a training-free decoding rule, GACL, that significantly improves faithfulness by correcting this arbitration bias.