Yaru Cao
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Yaru Cao has a total of 23 co-authors. Here are the top co-authors by number of shared publications:
Advait Bhat
The Behavioral Data Science Group, Paul G. Allen School of Computer Science and Engineering, University of Washington
1 shared publicationChris Z. Yao
The Conversational AI Group, Department of Computer Science and Technology, Institute for Artificial Intelligence, Tsinghua University
1 shared publication- Fei Ma
Guangdong Laboratory of Artificial Intelligence and Digital Economy (SZ), Shenzhen, China
1 shared publication - Fucheng Wan
Key Laboratory of Linguistic and Cultural Computing Ministry of Education (Northwest Minzu University), Northwest University for Nationalities, Lanzhou, Gansu, China
1 shared publication Hongning Wang
The Conversational AI Group, Department of Computer Science and Technology, Institute for Artificial Intelligence, Tsinghua University
1 shared publication- Hongzhi Yu
Key Laboratory of Minzu Languages and Cultures Intelligent Information Processing, Gansu Province, Northwest Minzu University, Lanzhou 730030, China
1 shared publication
Latest Publications
- Open Access
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17552-17559
CARES: A Clinically Anchored Residual Evidence System for Multimodal Depression Assessment in Older Adults
- 9 November 2026
- Published by Association for Computing Machinery (ACM)
- p. 17552-17559
Open AccessCARES: A Clinically Anchored Residual Evidence System for Multimodal Depression Assessment in Older Adults
- Open Access
- 14 September 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (38)
Human preferences are susceptible to covertly misaligned AI advice
AI assistants are increasingly used as advisors to guide decisions, yet little is known about how people evaluate such advice when the advisor’s underlying intent conflicts with their interests. We examine how covert misalignment shapes choices in a randomized experiment ( N = 233 participants; 699 observations) in which participants rated financial or emotional decisions before and after consulting one of three AI advisors: a neutral advisor, a misaligned advisor with a hidden objective to promote an inferior option, or a strategy-enhanced misaligned advisor additionally equipped with established tactics of covert influence. Across both domains, exposure to misaligned advisors shifted preferences away from optimal options and toward inferior alternatives, increasing the odds of preferring the incentivized (inferior) option over the optimal option by ≈5 to 8 times (up to +38 percentage points). Adding explicit influence strategies did not reliably strengthen these effects. Notably, participants continued to rate misaligned advisors as helpful, revealing a systematic disconnect between susceptibility to misaligned advice and subjective evaluations of advisor quality. These findings have implications for the design and governance of AI-mediated advice.
- 14 September 2026
- Published by National Academy of Sciences in Proceedings of the National Academy of Sciences
- Vol. 123 (38)
Open AccessHuman preferences are susceptible to covertly misaligned AI advice
AI assistants are increasingly used as advisors to guide decisions, yet little is known about how people evaluate such advice when the advisor’s underlying intent conflicts with their interests. We examine how covert misalignment shapes choices in a randomized experiment ( N = 233 participants; 699 observations) in which participants rated financial or emotional decisions before and after consulting one of three AI advisors: a neutral advisor, a misaligned advisor with a hidden objective to promote an inferior option, or a strategy-enhanced misaligned advisor additionally equipped with established tactics of covert influence. Across both domains, exposure to misaligned advisors shifted preferences away from optimal options and toward inferior alternatives, increasing the odds of preferring the incentivized (inferior) option over the optimal option by ≈5 to 8 times (up to +38 percentage points). Adding explicit influence strategies did not reliably strengthen these effects. Notably, participants continued to rate misaligned advisors as helpful, revealing a systematic disconnect between susceptibility to misaligned advice and subjective evaluations of advisor quality. These findings have implications for the design and governance of AI-mediated advice.
- Open Access
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 227-233
Improving Emotional Support Conversation with Strategy-Intent Inference
The demand for emotional support is increasingly growing. To enhance the effectiveness of models in emotional support conversations, strategy-guided dialogue generation has become a common approach. However, relying solely on strategy keywords is insufficient for the model to grasp the underlying logic of emotional support responses. This paper introduces a Strategy-Intent Inference framework designed to assist supporters in selecting appropriate strategies and clarifying strategy intent, thereby improving the quality of emotional support dialogues. To validate the effectiveness of Strategy-Intent Inference, we applied it to the ESConv dataset, expanding it into an emotional support dialogue dataset that incorporates Strategy-Intent Inference, and trained a dialogue model with this capability on the dataset. Comparative experiments conducted in both in-domain and out-of-domain settings demonstrated improvements in multiple metrics (BLEU and Distinct) compared to existing strategy-enhanced models. These results indicate that Strategy-Intent Inference can effectively enhance the model's emotional support capabilities and generalization performance.
- 18 October 2024
- Published by Association for Computing Machinery (ACM)
- p. 227-233
Open AccessImproving Emotional Support Conversation with Strategy-Intent Inference
The demand for emotional support is increasingly growing. To enhance the effectiveness of models in emotional support conversations, strategy-guided dialogue generation has become a common approach. However, relying solely on strategy keywords is insufficient for the model to grasp the underlying logic of emotional support responses. This paper introduces a Strategy-Intent Inference framework designed to assist supporters in selecting appropriate strategies and clarifying strategy intent, thereby improving the quality of emotional support dialogues. To validate the effectiveness of Strategy-Intent Inference, we applied it to the ESConv dataset, expanding it into an emotional support dialogue dataset that incorporates Strategy-Intent Inference, and trained a dialogue model with this capability on the dataset. Comparative experiments conducted in both in-domain and out-of-domain settings demonstrated improvements in multiple metrics (BLEU and Distinct) compared to existing strategy-enhanced models. These results indicate that Strategy-Intent Inference can effectively enhance the model's emotional support capabilities and generalization performance.